2023 MLOps World Conference & Expo – Join Us

By Toronto Machine Learning Society (TMLS)

Meet Our Speakers

Advanced Technical / Research

Andreea Munteanu

Andreea Munteanu

AI/ML Product Manager, Canonical
Talk: MLOps on Highly Sensitive Data - Strict Confinement, Confidential Computing and Tokenization Protecting Privacy

Maciej Mazur

Maciej Mazur

Principal ML Engineer, Canonical
Talk: MLOps on Highly Sensitive Data - Strict Confinement, Confidential Computing and Tokenization Protecting Privacy

Jonas Mueller

Jonas Mueller

Chief Scientist, Cleanlab
Talk: Operationalizing Data-Centric Ai: Practical Algorithms + Software to Quickly Improve ML Datasets

Aishwarya Reganti

Aishwarya Reganti

Applied Scientist, Amazon
Talk: MLOps for Graph-Based Recommender Systems: Orchestrating Intelligent Connections

Goku Mohandas

Goku Mohandas

ML Lead, Anyscale
Talk: End-to-end Production Machine Learning Workflows

Matthew Honnibal

Matthew Honnibal

Founder & CTO, Explosion AI
Talk: How many Labelled Examples Do You Need for a BERT-sized Model to Beat GPT4 on Predictive Tasks?

William Falcon

William Falcon

Founder & CEO, Lightning AI
Talk: Gen AI and Lightning: Accelerating AI Innovation while Ensuring Scalability and Security

Rajiv Shah

Rajiv Shah

Machine Learning Engineer, Hugging Face
Talk: Evaluation Techniques for Large Language Models

Shreya Rajpal

Shreya Rajpal

Founder, Guardrails AI
Talk: Your AI Applications Need Guardrails: Here's How To Build Them

Sandeep Singh

Sandeep Singh

Head of Applied AI, Beans.AI
Talk: Stable Diffusion for Your Images: Custom Dream

Kumaran Ponnambalam

Principal Engineer - AI, Cisco Systems Inc., Emerging Tech & Incubation
Talk: Using Scouter Models to Monitor Model Drift - A Novel Approach

D. Sculley

CEO, Kaggle
Talk: Is It Too Much to Ask for A Stable Baseline?

Sophia Yang

Sophia Yang

Senior Data Scientist, Anaconda
Talk: Introduction to LangChain and Retrieval Augmented Generation (RAG)

Business Strategy

Dr. Ilyas Iyoob

Dr. Ilyas Iyoob

Faculty, University of Texas
Talk: Lessons Learned from Implementing GenAI at Large Enterprises

Reuven Cohen

Reuven Cohen

CTO Generative AI, EY Americas
Talk: Unlocking Potential: A Fireside Chat on Generative AI, Reasoning, and Logic in Enterprise Innovation

Hien Luu

Hien Luu

Head of ML Platform, DoorDash
Talk: LLMOps: An Emerging Stack to Productionalize LLM Applications

Norm Zhou

Norm Zhou

Engineering Manager, Meta
Talk: Beyond the Kaggle Paradigm: Future of End-to-End ML Platforms

Shaun Hillin

Shaun Hillin

Global Head of Solutions Architecture, Cohere
Talk: Behind the Hype....The Real Promise of LLM: Fundamentally Transforming Humanity’s Access to Innovation

David Bennett

David Bennett

Chief Customer Officer, Tenstorrent
Talk: Building Computers for AI

Muller Mu

Muller Mu

Solution Architect / Senior Scientist, Roche
Talk: Wanted: A Silver-Bullet Ml Ops Solution for Enterprise - Learnings from Implementing Sustainable Ml Ops in Pharma Research

Naiel Samaan

Naiel Samaan

Senior Product Owner, AI Platform, Ford Motor Company
Talk: Tales of Innovation within a 100 Year Old Company

Valmir Bucaj

Valmir Bucaj

AI/ML Platform Product Owner, Ford Motor Company
Talk: Tales of Innovation within a 100 Year Old Company

Liran Hason

Liran Hason

CEO & Co-Founder, Aporia
Talk: Removing the Roadblocks to Build Great GenAI Products

Stefan Krawczyk

Stefan Krawczyk

CEO & Co-Founder, DAGWorks Inc.
Talk: Getting Higher Roi on Ml Ops Initiatives: Five Lessons Learned While Building out The Ml Ops Platform for 100+ Data Scientists

Case Studies

Hagay Lupesco

Hagay Lupesco

VP Engineering, MosaicML
Talk: Training Large Language Models: Lessons from The Trenches

Prakash Putt

Prakash Putt

Staff Software Engineer, Instacart
Talk: Supercharging Search with LLMs: The Instacart Journey

Ryan McClelland

Ryan McClelland

Research Engineer, NASA Goddard Space Flight Center
Talk: Evolved Structures: Using AI and Robots to Build Spaceflight Structures at NASA

Vincent David

Vincent David

Senior Director - Machine Learning, Capital One
Talk: Low-latency Model Inference in Finance

Michael Meredith

Michael Meredith

Lead Software Engineer, Capital One
Talk: Low-latency Model Inference in Finance

Susrutha Gongalla

Susrutha Gongalla

Principal Machine Learning Engineer, Stitch Fix
Talk: Supercharging Recommender Systems: Unleashing the Power of Distributed Model Training

Aayush Mudgal

Aayush Mudgal

Senior Machine Learning Engineer, Pinterest
Talk: Evolution of ML Training and Serving Infrastructure at Pinterest

Ian Schweer

Ian Schweer

Staff Software Engineer, Riot Games
Talk: Amumu Brain; How League of Legends Uses Machine Learning an Applied Data Science

In-Person Workshop

Andreea Munteanu

Andreea Munteanu

AI/ML Product Manager, Canonical
Workshop: Build Your Own ChatGPT with Open Source Tooling

Greg Loughnane

Dr. Greg Loughnane

Founder & CEO, AI Makerspace
Workshop: Retrieval Augmented Generation (RAG) with LangChain: “ChatGPT for Your Data” with Open-Source Tools

Chris Alexiuk

Chris Alexiuk

Head of LLMs, AI Makerspace
Workshop: Retrieval Augmented Generation (RAG) with LangChain: “ChatGPT for Your Data” with Open-Source Tools

Virtual Workshops

Anouk Dutree

Anouk Dutree

Product Owner, UbiOps
Workshop: Deploying Generative AI Models: Best Practices and An Interactive Example

Eddie Mattia

Eddie Mattia

Data Scientist, Outerbounds
Workshop: LLMs in Practice: A Guide to Recent Techniques and Trends

Ville Tuulos

Ville Tuulos

CEO, Outerbounds
Workshop: LLMs in Practice: A Guide to Recent Techniques and Trends

Tibor Mach

Tibor Mach

Machine Learning Solutions Engineer, DVC
Workshop: Applying GitOps Principles at Every Step of An E2E MLOps Project - An Interactive Workshop

Niels Bantilan

Niels Bantilan

Chief ML Engineer, Union.ai
Workshop: Learn Your Codebase: Fine-tuning CodeLlama with Flyte… to Learn Flyte

Simba Khadder

Simba Khadder

Founder & CEO, Featureform
Workshop: Feature Stores in Practice: Train and Deploy an End-to-End Fraud Detection Model with Featureform, Redis, and AWS

David Talby

David Talby

CTO, John Snow Labs
Workshop: Applying Responsible AI with the Open-Source LangTest Library

Ramon Perez

Ramon Perez

Developer Advocate, Seldon
Workshop: Introduction to Building ML Microservices: A Hands-On Approach with Examples from The Music Industry

Aniket Maurya

Aniket Maurya

Developer Advocate, Lightning AI
Workshop: Finetuning a Large Language Model on A Custom Dataset

Fabiana Clemente

Fabiana Clemente

Chief Data Officer, YData
Workshop: Synthetic Data: Generative AI for Enhanced Data Quality in the Era of Foundational Models

Salina Wu

Salina Wu

Senior Machine Learning Infrastructure Engineer, Forethought
Workshop: Avoid ML OOps with MLOps: A Modular Approach to Scaling Forethought’s E2 E Ml Platform

Guanghua Shu

Guanghua Shu

Staff Machine Learning Engineer, Instacart
Workshop: Lessons Learned: The Journey to Real-Time Machine Learning at Instacart

Serg Masis

Serg Masis

Lead Data Scientist & Bestselling Author, Syngenta
Workshop: QA in ML

Greg Kuhlmann

Greg Kuhlmann

CEO, Sumatra
Workshop: Learning from Extremes: What Fraud-Fighting at Scale Can Teach Us About MLOps Across Domains

Dan Shiebler

Dan Shiebler

Head of Machine Learning, Abnormal Security
Workshop: How to Design and Build Resilient Machine Learning Systems

Alon Gubkin

Alon Gubkin

CTO & Co-Founder, Aporia
Workshop: Spend Less Time Troubleshooting ML Production Issues

Lightning Talks

Eddie Mattia

Eddie Mattia

Data Scientist, Outerbounds
Talk: LLMs from Hallucinations to Relevant Responses

Aurimas Griciūnas

Aurimas Griciūnas

Head of Product, Neptune AI
Talk: Building an end-to-end MLOps Pipeline

Andreea Munteanu

Andreea Munteanu

AI/ML Product Manager, Canonical
Talk: Generative AI: The Open Source Way

Richard Izzo

Richard Izzo

Tech Lead, Lightning AI
Talk: Supporting Community Competitions to Develop LLMs

Simba Khadder

Simba Khadder

Founder & CEO, Featureform
Talk: Feature Store are NOT about Storing Features

Ryan Turner

Ryan Turner

ML Solutions Engineer, DVC
Talk: How NOT to Get ML Models into Production

Gon Rappaport

Gon Rappaport

MLOps Solutions Architect, Aporia
Talk: The Evolution of ML Monitoring in Production: From ML 1.0 to LLMs

Jose Nicholas-Francisco

Jose Nicholas Francisco

ML Developer Advocate, Deepgram
Talk: LLMs, Big Data, and Audio: Breaching an Untapped Gold Mine

Speakers Corner

Daniel Lenton

Daniel Lenton

CEO, Ivy
Talk: Which Compilers Are Best for LLMs?

Dmitry Petrov

CEO, DVC
Talk: Data Versioning in Generative AI: A Pathway to Cost-Effective ML

More Speakers to be announced

Agenda

This agenda is still subject to changes.

Loading...

Tickets

Say Hello

Follow us

MLOps World est 2020

Follow us

Toronto Machine Learning Society est. 2017

Talk: MLOps on Highly Sensitive Data - Strict Confinement, Confidential Computing and Tokenization Protecting Privacy

Presenters:
Andreea Munteanu, AI/ML Product Manager, Canonical | Maciej Mazur, Principal ML Engineer, Canonical

About the Speakers:
I am a Product Manager at Canonical, leading the MLOps area. With a background in Data Science in various industries, such as retail or telecommunications, I used AI techniques to enable enterprises to benefit from their initiatives and make data-driven decisions. I am looking to help enterprises get started with their AI projects and then deploy them to production, using open-source, secure, stable solutions.

I am a driven professional, passionate about machine learning and open source. I always look for opportunities to improve, both myself and people within the teams that I am part of. I enjoy sharing my knowledge, mentoring young professionals and having an educational impact in the industry.

Talk Track: Advanced Technical/Research

Talk Technical Level: 4/7

Talk Abstract:
MLOps is used in various organizations that operate on very sensitive datasets. For instance, pharmaceutical and life science companies handle human DNA samples, healthcare institutions training models on patient data, or highly regulated environments like telecom and financial companies. End users, machine learning engineers or data scientists can be concerned about cloud-native workloads would expose them more to software vulnerabilities, data leaks, or any lack of data protection measures. In reality, it’s just the opposite. This presentation will cover how you can improve compliance and security with features like Kubernetes strict confinement, blockchain-based tokenization and privacy-enhancing technologies like confidential computing. The talk will feature a case study by a life sciences company that created customized treatments using these technology building blocks. After the talk, you will understand how you can apply them yourself on cloud envorinments MLOps using Kubeflow.

What You’ll Learn:
In this talk, we will discuss how you can set up a secure foundation for machine learning with open-source building blocks. We will cover how confidential computing on the public cloud helps you address run time insecurity. You will then learn how Kubernetes’ strict confinement helps you get complete isolation, up to a minimum access level to the host resources. Finally, we will cover how tokenization can enable you to avoid data leaks, and allows at the same time achieving high system productivity. We will demonstrate how this works in practice with a life sciences use case powered by Charmed Kubeflow, an open-source community-driven end-to-end MLOps platform.

Talk: Operationalizing Data-Centric Ai: Practical Algorithms + Software to Quickly Improve ML Datasets

Presenter:
Jonas Mueller, Chief Scientist, Cleanlab

About the Speaker:
Jonas Mueller is Chief Scientist and Co-Founder at Cleanlab, a software company providing data-centric AI tools to efficiently improve ML datasets. Previously, he was a senior scientist at Amazon Web Services developing AutoML and Deep Learning algorithms which now power ML applications at hundreds of the world’s largest companies. In 2018, he completed his PhD in Machine Learning at MIT, also doing research in NLP, Statistics, and Computational Biology. Jonas has published over 30 papers in top ML and Data Science venues (NeurIPS, ICML, ICLR, AAAI, JASA, Annals of Statistics, etc). This research has been featured in Wired, VentureBeat, Technology Review, World Economic Forum, and other media. He loves contributing to open-source, and helped create the fastest-growing open-source software for AutoML  and Data-Centric AI. An avid educator, he also taught the first-ever course on data-centric AI at MIT.

Talk Track: Advanced Technical/Research

Talk Technical Level: 4/7

Talk Abstract:
TBA

What You’ll learn:
How to best practice data-centric AI in real-world ML projects. This covers automated methods to check the dataset for various issues common in ML data as well as how to efficiently address the issues to improve the dataset and subsequent ML model. I will cover novel algorithms invented by our research team and case studies which showcase the benefits of data-centric AI in real-world ML applications.

The intended audience is folks with experience in supervised learning who want to develop the most effective ML for messy, real-world applications. Some of the content will be technical, but not require a deep understanding of how particular ML algorithms/model work (having completed one previous ML course/project should suffice). The topics should be of interest to anybody working in: computer vision, natural language processing, audio/speech or tabular data, and other standard supervised learning applications, as well as DataOps folks.

Talk: MLOps for Graph-Based Recommender Systems: Orchestrating Intelligent Connections

Presenter:
Aishwarya Reganti, Applied Scientist, Amazon

About the Speaker:
Aishwarya is an Applied Scientist in the Amazon Search Science and AI Org. She works on developing large scale graph-based ML techniques that improve Amazon Search Quality, Trust and Recommendations. She obtained her Master’s degree in Computer Science (MCDS) from Carnegie Mellon’s Language Technology Institute, Pittsburgh. She has over 6+ years of hands-on Machine Learning experience and 20+ publications in top-tier conferences like AAAI, ACL, CVPR, NeurIPS, EACL e.t.c. She has worked on a wide spectrum of problems that involve Large Scale Graph Neural Networks, Machine Translation, Multimodal Summarization, Social Media and Social Networks, Human Centric ML, Artificial Social Intelligence, Code-Mixing e.t.c. She has mentored several Masters and PhD students in the aforementioned areas. She has also served as a reviewer in various NLP and Graph ML conferences like ACL, EMNLP, AAAI, LoG e.t.c. She has worked with some of the best minds in both academia and industry through collaborations and internships in Microsoft Research, University of Michigan, NTU Singapore, IIIT-Delhi, NTNU-Norway, University of South Carolina e.t.c.

Talk Track: Advanced Technical/Research

Talk Technical Level: 4/7

Talk Abstract:
As graph-based recommender systems continue to gain prominence in large tech companies, the need for efficient management of machine learning operations (MLOps) becomes paramount. This talk highlights the specific challenges faced in operationalizing graph-based recommender systems and how MLOps addresses these complexities. We delve into techniques for efficient data ingestion, feature engineering, and model training within the graph context. We emphasize the significance of versioning and managing graph models, ensuring reproducibility and scalability in the MLOps pipeline.

What You’ll Learn:
The audience will learn about the unique advantages and challenges of implementing graph-based MLOps in recommender systems. They will gain insights into how graph structures enable more accurate recommendations. The talk will cover techniques for data preprocessing, feature engineering, and model training specific to graph data using the DGL library. Overall, attendees will gain a comprehensive understanding of the transformative potential of graph-based MLOps in building and managing recommender systems.

Talk: End-to-end Production Machine Learning Workflows

Presenter:
Goku Mohandas, ML Lead, Anyscale

About the Speaker:
TBA

Talk Track: Advanced Technical/Research

Talk Technical Level: 4/7

Talk Abstract:
We’ll start by breaking down the machine learning development lifecycle into experimentation (design + develop) and production (deploy + iterate). We’ll walk through the best practices for developing and executing ML workloads and iteratively build a production workflow around it (manual to CI/CD to continual learning). We’ll also take a look at the biggest obstacles in the way of taking machine learning in production (standardization, integrations, scaling workloads in Python, dev to prod transition, reliability, etc.) While our specific use case will be fine-tuning an LLM for a supervised NLP use case, all the content will easily extend to any algorithm (regression to LLMs), application (NLP, CV, tabular, etc.), tooling stacks and scales.

What You’ll Learn:
– Learn the best practices for developing ML workloads (data, train, tune, serve, etc.)
– Learn how to incorporate MLOps concepts to our data science work (experiment tracking, testing code, data + models, monitoring, etc.)
– Learn how to fine-tune an LLM for a supervised use case
– Learn how to iteratively put ML into production (manual deployment, CI/CD, continual learning)
– Learn how to execute ML workloads across multiple machines easily

Talk: How many Labelled Examples Do You Need for a BERT-sized Model to Beat GPT4 on Predictive Tasks?

Presenter:
Matthew Honnibal, Founder & CTO, Explosion AI

About the Speaker:
Major work includes:

– ExplosionAI GmbH: Founder and CTO
– spaCy: Open-source NLP library in use by thousands of companies, with over 100m downloads. Particularly known for efficiency and API design.
– Prodigy: Developer-focussed annotation tool, with active learning and scriptability features. Licenses purchases by almost 1000 companies.
– Thinc: Open-source ML library built for spaCy, designed around function composition.

Entered the field in 2005 as a linguist, transitioning towards computer science over PhD and post-doctoral research. Left academia in 2014. Originally from Sydney, now in Berlin.

Talk Track: Advanced Technical/Research

Talk Technical Level: 4/7

Talk Abstract:
Large Language Models (LLMs) offer a new machine learning interaction paradigm: in-context learning. This approach is clearly much better than approaches that rely on explicit labelled data for a wide variety of generative tasks (e.g. summarisation, question answering, paraphrasing). In-context learning can also be applied to predictive tasks such as text categorization and entity recognition, with few or no labelled exemplars.

But how does in-context learning actually compare to supervised approaches on those tasks? The key advantage is you need less data, but how many labelled examples do you need on different problems before a BERT-sized model can beat GPT4 in accuracy?

The answer might surprise you: models with fewer than 1b parameters are actually very good at classic predictive NLP, while in-context learning struggles on many problem shapes — especially tasks with many labels or that require structured prediction. Methods of improving in-context learning accuracy involve increasing trade-offs of speed for accuracy, suggesting that distillation and LLM-guided annotation will be the most practical approaches.

Implementation of this approach is discussed with reference to the spaCy open-source library and the Prodigy annotation tool.

What You’ll Learn:
TBA

Talk: Gen AI and Lightning: Accelerating AI Innovation while Ensuring Scalability and Security

Presenter:
William Falcon, Founder & CEO, Lightning AI

About the Speaker:
William Falcon is the creator of PyTorch Lightning, a deep learning framework. He is also the founder and CEO of Lightning AI, and was previously a co-founder and CTO of NextGenVest. He began working on these projects while completing a Ph.D. at NYU, which was funded by Google DeepMind and the NSF. Additionally, he worked as a researcher at Facebook AI and at Goldman Sachs.

Talk Track: Advanced Technical/Research

Talk Technical Level: 5/7

Talk Abstract:
As artificial intelligence (AI) continues to reshape industries and drive transformative advancements, the need for accelerated AI innovation has become paramount. Organizations across the globe are seeking novel ways to leverage AI technologies to gain a competitive edge, optimize processes, and deliver enhanced customer experiences. However, as AI applications grow in complexity and scale, so do the challenges related to scalability and security.

This talk aims to address the critical aspects of accelerating Enterprise AI innovation and will delve into the following key topics:

– Harnessing the Power of PyTorch and PyTorch Lightning: We will explore the benefits of modern AI frameworks, such as PyTorch, PyTorch Lightning, and Fabric, and the techniques that facilitate rapid development and deployment of AI solutions.
– Scalability: The exponential growth of data and increasing demands for AI applications necessitate scalable architectures. We will discuss strategies for designing AI systems that can effortlessly handle massive datasets and adapt to the evolving requirements of AI-driven applications.
– Security Considerations in AI: The integration of AI technologies introduces new security risks. This talk will highlight best practices for safeguarding AI systems against potential threats, ensuring data privacy, and maintaining compliance.
– Collaborative Ecosystems for Innovation: Accelerating AI innovation requires collaboration between diverse stakeholders, including researchers, developers, policymakers, and businesses. We will explore successful collaborative models that foster innovation.

What You’ll Learn:
TBA

Talk: Evaluation Techniques for Large Language Models

Presenter:
Rajiv Shah, Machine Learning Engineer, Hugging Face

About the Speaker:
Rajiv Shah is a machine learning engineer at Hugging Face who focuses on enabling enterprise teams to succeed with AI. Rajiv is a leading expert in the practical application of AI. Previously, he led data science enablement efforts across hundreds of data scientists at DataRobot. He was also a part of data science teams at Snorkel AI, Caterpillar, and State Farm.

Rajiv is a widely recognized speaker on AI, published over 20 research papers, been cited over 1000 times, and received over 20 patents. His recent work in AI covers topics such as sports analytics, deep learning, and interpretability.

Rajiv holds a PhD in Communications and a Juris Doctor from the University of Illinois at Urbana Champaign. While earning his degrees, he received a fellowship in Digital Government from the John F. Kennedy School of Government at Harvard University. He has recently started making short videos, @rajistics, with several million views.

Talk Track: Advanced Technical/Research

Talk Technical Level: 5/7

Talk Abstract:
Large language models (LLMs) represent an exciting trend in AI, with many new commercial and open-source models released recently. However, selecting the right LLM for your needs has become increasingly complex. This tutorial provides data scientists and machine learning engineers with practical tools and best practices for evaluating and choosing LLMs.

The tutorial will cover the existing research on the capabilities of LLMs versus small traditional ML models. If an LLM is the best solution, the tutorial covers several techniques, including evaluation suites like the EleutherAI Harness, head-to-head competition approaches, and using LLMs for evaluating other LLMs. The tutorial will also touch on subtle factors that affect evaluation, including role of prompts, tokenization, and requirements for factual accuracy. Finally, a discussion of model bias and ethics will be integrated into the working examples.

Attendees will gain an in-depth understanding of LLM evaluation tradeoffs and methods. Jupyter Notebooks will provide reusable code for each technique discussed.

What You’ll Learn:
Ways to quickly start evaluating models

Talk: Your AI Applications Need Guardrails: Here's How To Build Them

Presenter:
Shreya Rajpal, Founder, Guardrails AI

About the Speaker:
Shreya Rajpal is the creator and maintainer of Guardrails AI, an open source platform developed to ensure increased safety, reliability, and robustness of large language models in real-world applications. Her expertise spans a decade in the field of machine learning and AI. Most recently, she was the founding engineer at Predibase, where she led the ML infrastructure team. In earlier roles, she was part of the cross-functional ML team within Apple’s Special Projects Group and developed computer vision models for autonomous driving perception systems at Drive.ai.

Talk Track: Advanced Technical/Research

Talk Technical Level: 4/7

Talk Abstract:
Large Language Models (LLMs) such as ChatGPT have revolutionized AI applications, offering unprecedented potential for complex real-world scenarios. However, fully harnessing this potential comes with unique challenges such as model brittleness and the need for consistent, accurate outputs. These hurdles become more pronounced when developing production-grade applications that utilize LLMs as a software abstraction layer.

In this talk, we will tackle these challenges head-on. We introduce Guardrails AI, an open-source platform designed to mitigate risks and enhance the safety and efficiency of LLMs. We will delve into specific techniques and advanced control mechanisms that enable developers to optimize model performance effectively. Furthermore, we will explore how implementing these safeguards can significantly improve the development process of LLMs, ultimately leading to safer, more reliable, and robust real-world AI applications.

What You’ll Learn:
The audience will learn about the challenges and risks associated with Large Language Models (LLMs) and how the open-source platform, Guardrails AI, addresses these issues by providing specific techniques and advanced control mechanisms to optimize model performance, leading to safer, more reliable, and robust real-world AI applications.

Talk: Stable Diffusion for Your Images: Custom Dream

Presenter:
Sandeep Singh, Head of Applied AI, Beans.AI

About the Speaker:
Sandeep Singh is a leader in applied AI and computer vision in Silicon Valley’s mapping industry, and he is at the forefront of developing cutting-edge technology to capture, analyze and understand satellite imagery, visual and location data.

With a deep expertise in computer vision algorithms, machine learning and image processing and applied ethics, Sandeep is responsible for creating innovative solutions that enable mapping and navigation software to accurately and efficiently identify and interpret features to remove inefficiencies of logistics and mapping solutions.

His work includes developing sophisticated image recognition systems, building 3D mapping models, and optimizing visual data processing pipelines for use in logistics, telecommunications and autonomous vehicles and other mapping applications.

With a keen eye for detail and a passion for pushing the boundaries of what’s possible with AI and computer vision, Sandeep’s leadership is driving the future of applied AI forward.

Talk Track: Advanced Technical/Research

Talk Technical Level: 6/7

Talk Abstract:
Welcome to the “Stable Diffusion for Your Images: Custom Dream” workshop! In this one-hour session, we will dive into the fascinating world of stable diffusion techniques for creating custom images.

Dreambooth is an innovative platform that allows users to unleash their creativity and transform ordinary images into extraordinary works of art. Through stable diffusion, we will explore how to manipulate and enhance images in a visually stunning and captivating way.

During the workshop, participants will learn the fundamentals of stable diffusion and its applications in image editing. We will cover various techniques, including color manipulation, texture enhancement, and image blending, to create visually striking and unique compositions.

Additionally, we will delve into the intricacies of Dreambooth’s user-friendly interface, providing hands-on demonstrations and step-by-step guidance. Participants will have the opportunity to experiment with different filters, effects, and settings, unleashing their artistic potential and transforming their photos into mesmerizing masterpieces.

Whether you are a professional photographer looking to add an extra flair to your work or an amateur enthusiast eager to explore new creative avenues, this workshop is designed to inspire and empower you. Join us for an hour of exploration, experimentation, and artistic expression as we unlock the potential of stable diffusion with Dreambooth.

What You’ll Learn:
Stable Diffusion is very easy for anybody to use and customize for the purpose at hand.

Talk: Using Scouter Models to Monitor Model Drift - A Novel Approach

Presenter:
Kumaran Ponnambalam, Principal Engineer – AI, Cisco Systems Inc., Emerging Tech & Incubation

About the Speaker:
Kumaran Ponnambalam is a technology leader with 20+ years of experience in AI, Big Data, Data Processing & Analytics. His focus is on creating robust, scalable AI models and services to drive effective business solutions. He is currently leading AI initiatives in the Emerging Technologies & Incubation Group in Cisco. In this role he is focused on building MLOps and Observability services to enable ML. In his previous roles, he has built data pipelines, analytics, integrations, and conversational bots around customer engagement. He has also authored several courses on the LinkedIn Learning Platform in AI and Big Data.

Talk Track: Advanced Technical/Research

Talk Technical Level: 6/7

Talk Abstract:
Model Drift monitoring is a critical activity in the MLOps life cycle. Models in production may drift and decay due to several reasons. Identifying drift quickly and taking corrective action is critical for continued success of models. A severe limitation in drift monitoring in a number of cases, is the lack of true labels for production data. How do we monitor drift when true labels are not available? We built a technique called Scouter models, to identify concept drift, when true labels are not available. In this talk, we will talk about what a scouter model is, how to identify the right scouter feature and best practices in building and managing such models. This would be of great interest to data scientists and MLOps engineers.

What You’ll Learn:
Scouter model concepts, Building scouter models and best practices for managing them.

Talk: Is It Too Much to Ask for A Stable Baseline?

Presenter:
D. Sculley, CEO, Kaggle

About the Speaker:
D. is currently CEO of Kaggle and GM of 3P ML Ecosystems at Google. Prior to this role, he was a director of engineering in the Google Brain team, leading research teams working on robust, responsible, reliable and efficient ML and AI. During his 15 years at Google, he has worked on nearly every aspect of machine learning, and have led both product and research teams including those on some of the most challenging business problems. His work on machine learning and technical debt helped lay the foundation for the field of MLOps, and the book Reliable Machine Learning was named Best MLOps Book of 2022.

Talk Track: Advanced Technical/Research

Talk Technical Level: 3/7

Talk Abstract:
Evaluation and monitoring are the heart of any reliable machine learning system. But finding a stable reference point, a reliable comparison baseline, or even a decent performance metric can be surprisingly difficult in a world that is beset by changing conditions, feedback loops, and shifting distributions. In this talk, we will look at some of the ways that these conditions show up in more traditional settings like click through prediction, and then see how they might reappear in the emerging world of productionized LLMs and generative models.

What You’ll Learn:
Evaluation is hard, but not impossible, and with enough care we can probably say something useful about our models.

Talk: Introduction to LangChain and Retrieval Augmented Generation (RAG)

Presenter:
Sophia Yang, Senior Data Scientist, Anaconda

About the Speaker:
Sophia Yang is a Senior Data Scientist and a Developer Advocate at Anaconda. She is passionate about the data science community and the Python open-source community. She is the author of multiple Python open-source libraries such as condastats, cranlogs, PyPowerUp, intake-stripe, and intake-salesforce. She serves on the Steering Committee and the Code of Conduct Committee of the Python open-source visualization system HoloViz. She also volunteers at NumFOCUS, PyData, and SciPy conferences. She holds an M.S. in Computer Science, an M.S. in Statistics, and a Ph.D. in Educational Psychology from The University of Texas at Austin.

Talk Track: Advanced Technical/Research

Talk Technical Level: 4/7

Talk Abstract:
LangChain is an open-source framework for developing applications powered by language models. In this introductory talk, we’ll explore the core concepts behind LangChain including chains, memory, tools, agents, embeddings, and vector databases. Additionally, we’ll use LangChain and Panel to build LLM-based applications including answering questions using information from documents, a technique known as Retrieval Augmented Generation. Join us to learn LangChain!

What You’ll Learn:
Use LangChain to develop LLM applications

Talk: Lessons Learned from Implementing GenAI at Large Enterprises

Presenter:
Dr. Ilyas Iyoob, Faculty, University of Texas

About the Speaker:
Dr. Ilyas Iyoob is chief data scientist and global head of Research at Kyndryl. He pioneered the seamless interaction between machine learning and operations research in the fields of autonomous computing, fintech, and blockchain. As a successful entrepreneur at Gravitant, a start-up focused on optimizing the cloud journey, he helped build and sell the company to IBM in 2016. Dr. Iyoob currently advises over a dozen venture funded companies and serves on the faculty of the Cockrell School of Engineering at the University of Texas at Austin. He has earned a number of patents and industry recognition for cloud intelligence and was awarded the prestigious World Mechanics prize by the University of London.

Talk Track: Business Strategy

Talk Technical Level: 3/7

Talk Abstract:
In the rapidly evolving landscape of GenAI, large US enterprises face unique challenges when considering its implementation. Beyond the well-acknowledged concerns of data privacy, security, bias, and regulatory compliance, our journey in executing GenAI within mission-critical applications has revealed additional complexities. In this session we will walk through a number of real examples of failed implementations and the lessons learned from them.

What You’ll learn:
Learn from mistakes; new best practices for successful implementation of GenAI

Talk: Unlocking Potential: A Fireside Chat on Generative AI, Reasoning, and Logic in Enterprise Innovation

Presenter:
Reuven Cohen, CTO Generative AI, EY Americas

About the Speaker:
Reuven Cohen is an influential technology leader with over three decades of experience, particularly recognized as the CTO, Generative AI at EY Americas. In this role, he is spearheading the Generative AI strategy, transforming the workforce of nearly 400,000 and serving a vast portfolio of Fortune 500 clients.

His pioneering work in cloud computing is equally noteworthy. He founded Enomaly Inc., a trailblazing cloud computing company, and coined the term “”infrastructure as a service”” in 2005. As an inaugural member of the Amazon Web Services advisory board and co-founder of CloudCamp in 2008, Reuven’s impact on the cloud computing field is undeniable.

In the realm of artificial intelligence, Reuven served as an alpha/beta tester for OpenAI, contributing significantly to the growth and innovation of AI programming and engineering.

At EY, Reuven’s visionary leadership, combined with his extensive experience in both AI and cloud computing, continues to drive technological advancements and organizational success. His commitment to innovation and collaboration has solidified his reputation as a trusted authority in the industry, making him an indispensable ally for businesses, governments, and technology enthusiasts alike.

Talk Track: Business Strategy

Talk Technical Level: 5/7

Talk Abstract:
In a world where Generative AI is shaping enterprise innovation, this fireside chat explores the integration of reasoning, logic, and cutting-edge techniques like Language Models (LLM) and mixture of experts tactics. Led by Reuven Cohen, CTO of Generative AI at EY Americas, the discussion uncovers how LLM fuels new chain-of-thought processes, while the mixture of experts approach optimizes decision-making within businesses.

Join us for a thought-provoking conversation that delves into the real-world applications of these technologies, offering valuable insights into their transformative impact on modern enterprises. This talk is a must for anyone interested in the intersection of human intellect, artificial intelligence, and the future of business innovation.

What You’ll learn:
Attendees will leave with a deeper understanding of how Generative AI, combined with innovative techniques like Language Models (LLM) and the mixture of experts tactics, is revolutionizing reasoning, logic, and decision-making within modern enterprises.

Talk: LLMOps: An Emerging Stack to Productionalize LLM Applications

Presenter:
Hien Luu, Head of ML Platform, DoorDash

About the Speaker:
Passionate about the intersection of big data & machine learning. Extensive working experience in designing and building big data applications and scalable web-based applications. Have 3 plus years of technical leadership experience in managing multiple data infrastructure related projects. Passion for architecting scalable and highly available big data applications and systems.

Specialties: Big data, web application framework, cloud computing, RESTful web services and cryptography.

Part-time Passion:
* Instructor at UCSC Extension: Apache Spark, Apache Hadoop, Spring Framework, Design Patterns and Java Comprehensive.

Speaking Engagements:
* QCon NY 2019 – track host for ML for Developers track
* QCon (SF, Shanghai, London, AI)
– Building Recommender Systems w/ Apache Spark 2.x
– Recommender System: Powering LinkedIn Growth & Engagement
– Apache Spark 2.x Workshop
* Seattle Data Day – Extreme Streaming Processing at Uber
* 2015 – ArchSummit – Big Data Story : from an engineer’s perspective
* 2013 – Lucene/Solr Revolution – How Lucene Powers the LinkedIn Segmentation and Targeting Platform
* 2013 – Hadoop Summit – LinkedIn Segmentation & Targeting Platform: A Big Data Application
* 2010 – JavaOne – REST with Spring Framework and JAX-RS

Talk Track: Business Strategy

Talk Technical Level: 3/7

Talk Abstract:
TBA

What You’ll learn:
The unique challenges that LLM applications bring and an emerging LLMOps stack can help w/ those challenges to support and operationalize LLM applications.

Talk: Beyond the Kaggle Paradigm: Future of End-to-End ML Platforms

Presenter:
Norm Zhou, Engineering Manager, Meta

About the Speaker:
Innovative, adaptable, and highly technical leader. Always looking holistically to make the highest impact starting from first principle. My career has lead me over many interesting challenges and environments. From hardware Chip architecture in ASIC and FPGA startups, to Ads then AI Platforms in large internet companies. Currently I am leading multiple teams working on AutoML to Democratize AI at Meta.

I am interested in maximizing my impact on this world by working both on cutting edge research and translational work that improves people’s lives in the real world.

Talk Track: Business Strategy

Talk Technical Level: 4/7

Talk Abstract:
ML platforms help enable intelligent data-driven applications and maintain them with limited engineering effort. However the current approaches to ML system building is limited by the “Kaggle Paradigm” which focuses on the data to model transformation and the operationalizing the deployment of models into applications. This model centric view limits further increase engineering productivity for future ML Systems. We propose an alternative policy centric view as an alternative to model centric view. This policy centric view involves two major additions to model centric view. First is a fully managed unified data collection system extending upstream to establish a “full chain of data custody”. Second we propose downstream extension to A/B testing systems which will bridge online offline mismatch typically experienced in many ML practitioners. Together these approaches enable a fully end-to-end automation allowing for a future ML platform to directly improve business metrics and more fluently address changing business needs.

What You’ll Learn:
How to best practice data-centric AI in real-world ML Connecting ML to Business Impact. Short comings of a model first approach and an proposed alternative.

Talk: Behind the Hype....The Real Promise of LLM: Fundamentally Transforming Humanity’s Access to Innovation

Presenter:
Shaun Hillin, Global Head of Solutions Architecture, Cohere

About the Speaker:
TBA

Talk Track: Business Strategy

Talk Technical Level: 3/7

Talk Abstract:
This talk delves into the revolutionary potential of large language models, transforming humanity’s access to innovation. Computers have exponentially accelerated creativity and production across industries, but direct communication with computers has remained limited to a small sample of the population, restricting the flow of rich ideas. By removing the shadow between ideas and the innovation, through machines understanding natural language, billions of hours of human expertise can be unleashed. The talk envisions a future where innovators can directly communicate with machines, democratizing technology and harnessing the power of Artificial Intelligence and Natural Language Processing to propel innovation to new heights.

What You’ll learn:
The massive unlock of innovation that Large Language Models will enable.

Talk: Building Computers for AI

Presenter:
David Bennett, Chief Customer Officer, Tenstorrent

About the Speaker:
I am currently the Chief Customer Officer (CCO) for Tenstorrent Inc., a Toronto based unicorn developing software, silicon and systems to run AI and ML faster than anyone else. We are also developing a full line up of RISC-V CPUs.

I recently left Tokyo, Japan where I ran Lenovo’s Japan operations including both the Lenovo brand and market leading NEC Personal Computer; the leader in Consumer, Commercial and Enterprise solutions, Lenovo is the largest PC manufacturer in Japan.

Previously, I ran AMD’s Asia based Global MNC accounts and was the Asia Pacific and Japan Mega Region Vice President where the team has achieved 12 quarters of consecutive YoY revenue and share growth. I also lead a global team driving >$600M a year in Commercial Client, Thin Client and Server CPU and Graphics processor revenue. We focus on an end-to-sale beginning with our OEM partners, right through to our end customers.

I am known for having exceptionally high bandwidth, a sales methodology based on creativity, passion and reliability, and for delivering results. I am constantly ranked top of our management surveys, and I believe that to survive and thrive we must know more than the competition about our products, the industry and our customers.

Fascinated by the future of technology and computing, I leverage my knowledge of hardware and software design and programming to predict trends and to shape our future product development and regional deployment strategies.

Talk Track: Business Strategy

Talk Technical Level: 3/7

Talk Abstract:
Tenstorrent is an innovative hardware architecture designed to enhance the efficiency and scalability of artificial intelligence (AI) workloads. Addressing the compute-intensive nature of modern AI applications, Tenstorrent leverages a unique grid-based architecture to enable efficient execution of both sparse and dense computations. Its dynamic code generation and execution capabilities allow flexibility across various AI models and algorithms. Furthermore, Tenstorrent emphasizes scalability, with the potential to deploy individual units in massive, interconnected networks for larger AI tasks. The system is designed to optimize power efficiency and performance, making it an ideal solution for advanced machine-learning jobs. Architectures like Tenstorrent will be crucial in harnessing its full potential as AI drives technological progress.

What You’ll learn:
The audience will learn about hardware for AI

Talk: Wanted: A Silver-Bullet Ml Ops Solution for Enterprise - Learnings from Implementing Sustainable Ml Ops in Pharma Research

Presenter:
Muller Mu, Solution Architect / Senior Scientist, Roche & Scientific

About the Speaker:
Having amassed extensive experience in digitalization across different industries, Le (Muller) Mu is combining his IT experience with his knowledge in molecular biology to help accelerate the drug discovery process in Roche Pharma Research and Early Development (pRED). He is currently the tech lead on the Roche pRED MLOps Service team, driving the operationalization of many different machine learning models, which helps to bring better future medicines faster into the hands of the patients.

Talk Track: Business Strategy

Talk Technical Level: 4/7

Talk Abstract:
The pharmaceutical research and early development (pRED) business unit in Roche is facing an increasing demand for operationalization of Machine Learning models that are making critical impacts on its value chain. While this demand is putting pressure on IT departments to deliver end-to-end and scalable MLOps solutions, readily-available tools are still maturing. In particular, the fragmentation and volatility of the current landscape makes the selection of such tools and their maintenance a daunting challenge for many teams. In this talk, we would like to present key learnings and recommendations from our 4-year MLOps journey. We will present and discuss how we sustainably served a broad spectrum of users and use cases with diverse requirements, technological literacy and data modalities. Finally, we would also like to share some perspectives for the future of MLOps in our research organization.

What You’ll learn:
1. The audience will learn the diverse requirements and unique challenges in implementing MLOps in Pharma Research
2. The audience will learn how we have built business and technical solutions in addressing these challenges so far
3. The audience will learn how we have identified the gap between readily-available MLOps platforms/solutions (commercial or open-source) and the needs for successfully implementing sustainable MLOps in an enterprise context with diverse requirements (such as in Pharma Research)
4. The audience will learn how we think this gap can be bridged by adhering to some of the fundamental principles/concepts in software engineering.

Talk: Tales of Innovation Within a 100 Year Old Company

Presenters:
Naiel Samaan, Senior Product Owner, AI Platform, Ford Motor Company | Valmir Bucaj, AI/ML Platform Product Owner, Ford Motor Company

About the Speakers:
Naiel- As a seasoned product, people, and project manager, I thrive on tackling complex problems with a combination of data, analysis, leadership, and creativity. Currently, I lead a team of data scientists, tech leads, and software developers in building an enterprise machine learning operations platform for Ford Motor Company.

My expertise lies in leveraging data and machine learning, along with innovative ideas, to solve business challenges. I’ve successfully managed every aspect of the process, from evaluating customer needs to informing product development to leading cross-functional teams to success.

My passion for developing teams, combined with my entrepreneurial drive and expertise in data and technology, make me the ideal candidate for any company looking to solve complex challenges and achieve growth.

Valamir – Valmir Bucaj is a technical leader and product owner of Ford’s AI/ML Platform. He has 3+ years of experienced in leading cross-functional teams, who specialize in building multi-cloud enterprise MLOps frameworks to help ML Engineers and Data Scientists scale and productionize their machine learning projects faster and more efficiently. Valmir is passionate in building AI/ML products that customers both need and love. He previously used to work as a machine learning engineer, focusing on graph neural networks for social recommendations.

Valmir used to also work as an assistant professor of mathematics at West Point, where among other things, he designed and taught the Academy’s first Data Science course for their newly established major.

Valmir holds a PhD in Mathematics, from Rice University.

Talk Track: Business Strategy

Talk Technical Level: 3/7

Talk Abstract:
In this talk we will discuss Ford’s journey in building an AI/ML platform from inception to present day. Specifically, we will share the many lessons learned around starting a platform within an organization with deeply rooted traditions. This includes making decisions for product strategy around customer persona, technical stack, buy vs. build, on-prem vs. cloud, and centralized vs. decentralized. We will discuss the reasons behind our evolving product strategy, the challenges in finding the right product-market fit, and the many organizational challenges.

What You’ll learn:
The audience will learn how to navigate organizational challenges in starting a AI/ML ecosystem at a large company with deeply rooted traditions. This will include pitfalls to avoid, building for your customer, and navigating team topology that works best for your ecosystem.

Talk: Removing the Roadblocks to Build Great GenAI Products

Presenter:
Liran Hason, CEO & Co-Founder, Aporia

About the Speaker:
Liran Hason is the Co-Founder and CEO of Aporia, the AI Performance Platform. Aporia is trusted by Fortune 500 companies and data science teams in every industry to enable responsible AI and monitor, improve, and scale AI products. Prior to founding Aporia, Liran was an ML Architect at Adallom (acquired by Microsoft), and later an investor at Vertex Ventures. Liran founded Aporia after seeing first-hand the risks of AI without guardrails. In 2022, Forbes named Aporia as the “Next Billion-Dollar Company”.

Talk Track: Business Strategy

Talk Technical Level: 3/7

Talk Abstract:
Generative AI represents an exciting new frontier, but deploying successful generative AI products comes with challenges. In this talk, we’ll examine common roadblocks organizations face and provide a framework to overcome them. By having the right approach we can accelerate time-to-market and drive impact.

What You’ll learn:
The audience will learn how to navigate organizational challenges in starting a AI/ML ecosystem at a large company with deeply rooted traditions. This will include pitfalls to avoid, building for your customer, and navigating team topology that works best for your ecosystem.

Talk: Getting Higher Roi on Ml Ops Initiatives: Five Lessons Learned While Building out The Ml Ops Platform for 100+ Data Scientists

Presenter:
Stefan Krawczyk, CEO & Co-Founder, DAGWorks Inc.

About the Speaker:
A hands-on leader and Silicon Valley veteran, Stefan has spent over 15 years thinking about data and machine learning systems, building product applications and infrastructure at places like Stanford, Honda Research, LinkedIn, Nextdoor, Idibon, and Stitch Fix. A regular conference speaker, Stefan has guest lectured at Stanford’s Machine Learning Systems Design course and is an author of a popular open source framework called Hamilton. Stefan is currently CEO of DAGWorks, an open source startup that is enabling teams a standardized way to build and maintain data, ML and LLM pipelines without the coding nightmares.

Talk Track: Business Strategy

Talk Technical Level: 4/7

Talk Abstract:
MLOps is hard, because there’s so many “things” that you might want to integrate and connect with: A/B testing, feature stores, model registries, data catalogs, lineage systems, python dependencies, machine learning libraries, LLM APIs, orchestration systems, online vs offline systems, speculative business ideas, etc. In this talk I’ll cover five lessons that I learned while building out the self-service MLOps platform for over 100 data scientists at Stitch Fix. This talk is for anyone building their own, or buying it all off the shelf. Either way you’re still going to want everything to fit cohesively together, i.e. as a platform, and learning what to avoid/focus on will increase your ROI on MLOps initiatives.

What You’ll learn:
Five lessons that will help them with MLOps initiatives:
1. Build for immediate adoption.
2. Don’t build for every user equally. Let those with stronger SWE skills do more themselves.
3. Don’t give users direct access to vendor/cloud APIs.
4. Take time to ensure you can live in the shoes of your users.
5. Provide two layers of APIs to keep your development nimble: a foundational layer, and then an opinionated higher level layer.

Talk: Training Large Language Models: Lessons from The Trenches

Presenter:
Hagay Lupesko, VP Engineering, MosaicML

About the Speaker:
Hagay Lupesko is the VP of Engineering at MosaicML, where he focuses on making generative AI training and inference efficient, fast, and accessible. Prior to MosaicML, Hagay held engineering leadership roles at Meta AI, AWS AI, and GE Healthcare. He shipped products across various domains: from 3D medical imaging, through global scale web systems, and up to deep learning systems powering apps and services used by billions of people world wide.

Talk Track: Case Study

Talk Technical Level: 5/7

Talk Abstract:
Training large AI language models is a challenging task that requires a deep understanding of natural language processing, machine learning, and distributed computing. In this talk, we will go over lessons learned from training models with billions of parameters across hundreds of GPUs. We will discuss the challenges of handling massive amounts of data, designing effective model architectures, optimizing training procedures, and managing computational resources. This talk is suitable for ML researchers, practitioners, and anyone curious about the “sausage making” behind training large language models.

What You’ll learn:
– The challenges in training models with billions of parameters across hundreds of GPUs
– Tips and tricks to make the training work and the model to achieve high accuracy

Talk: Supercharging Search with LLMs: The Instacart Journey

Presenter:
Prakash Putt, Staff Software Engineer, Instacart

About the Speaker:
TBA

Talk Track: Case Study

Talk Technical Level: 4/7

Talk Abstract:
Discover how Instacart’s search journey has been revolutionized through the implementation of Language Models (LLMs). By leveraging the power of LLMs, we have achieved significant enhancements, transforming the search experience for Instacart users. Join us to discover real-world use cases, gain insights into our seamless integration strategies, and witness how LLMs have empowered us to overcome challenges, deliver personalized recommendations, and elevate the overall search experience at Instacart.
An example

What You’ll Learn:
The audience will gain valuable insights into the practical implementation of LLMs in real production use cases. While LLMs have garnered significant attention, their effective integration into live environments remains a challenge. By attending this talk, participants will learn firsthand about the successful utilization of LLMs to enhance Instacart’s search journey. Through real-world examples, they will discover how LLMs can be harnessed to supercharge search capabilities and derive actionable knowledge for their own production scenarios.

Talk: Evolved Structures: Using AI and Robots to Build Spaceflight Structures at NASA

Presenter:
Ryan McClelland, Research Engineer, NASA Goddard Space Flight Center

About the Speaker:
From a young age, Ryan McClelland has been captivated by futurism and technology, aspiring to contribute to a brighter future. As a Research Engineer in NASA GSFC’s Instrument Systems and Technology Division, he pursues the development and implementation of digital engineering technologies for space-flight mission. Ryan is particularly excited about the potential of Artificial Intelligence, Virtual Reality, Generative Design, and Digital Manufacturing to accelerate space systems development.

With a diverse background in technology development, Ryan’s previous research encompasses lightweight X-ray optics, aluminum foam core optical systems, and the investigation of non-linear effects in kinematic mechanisms. In addition to his research, Ryan has played a significant role in various flight missions, including designs currently on orbit aboard the Hubble Space Telescope and International Space Station. Recently, he served as the Roman Space Telescope Instrument Carrier Manager. Ryan holds a B.S. in Mechanical Engineering, summa cum laude, from the University of Maryland.

Talk Track: Case Study

Talk Technical Level: 4/7

Talk Abstract:
Come get a first hand account of how NASA is leveraging Generative Design to reduce the cost and increase the performance of spaceflight missions. Also, how the concept of AI Prompt Engineering can be practically applied to diverse fields, such as structures development.

What You’ll learn:
TBA

Talk: Low-latency Model Inference in Finance

Presenters:
Vincent David, Senior Director – Machine Learning, Capital One | Michael Meredith, Lead Software Engineer, Capital One

About the Speakers:
Experienced Machine Learning & Engineering leader with a history of working in Fintech and Entertainment. Strong data and quantitative background, with deep knowledge of complex systems and experiences working in multiple industries. Skilled in Machine Learning, Cloud Engineering. Passionate about using technology to solve high-value business problems.

Talk Track: Case Study

Talk Technical Level: 5/7

Talk Abstract:
Model Inference at Capital One as across the Fintech sector is a key aspect of the Model Development Lifecycle. In order to reap the benefits of machine learning models trained by Data Scientists, we are required to deploy said models in a production environment. This enables critical business applications, such as credit decisions or fraud detection. Increasingly, we are faced with demanding non-functional requirements for high resilience and low latency service response leading us to invest into service oriented architectures for model inference. Seldon has emerged over the last years a key solution to address these challenges.

In this presentation we will take a close look at the recently released Seldon V2 and our findings for its application for financial services at enterprise scale, comparing it to its V1 specification. We will summarize our findings as they relate to the benefits of novel improvements as well as challenges we see in establishing much needed controls to establish this new release in a highly regulated environment.

What You’ll learn:
Practical learnings, insights, and considerations for more effectively deploying models in a production environment – especially complex production environments.

Talk: Supercharging Recommender Systems: Unleashing the Power of Distributed Model Training

Presenter:
Susrutha Gongalla, Principal Machine Learning Engineer, Stitch Fix

About the Speaker:
Susrutha Gongalla is an experienced Machine Learning Engineer with over 8 years of experience in developing end-to-end machine learning models. She is currently a Principal Machine Learning Engineer in the recommender systems team at Stitch Fix, where she leads the optimization of the end-to-end model lifecycle stack. Her primary focus is on developing and improving the scoring model that generates purchase probabilities for clothing items. Prior to joining Stitch Fix, Susrutha worked as a tech lead at Intuit, where she led the development of recommender systems to improve customer engagement and reduce churn. Besides recommender systems, she also worked on projects using Natural Language Processing techniques to derive insights from unstructured data. Her passion lies in using machine learning to drive business impact and data-driven decision making.

Susrutha holds a Master’s degree from Carnegie Mellon University and a Bachelor’s degree from Indian Institute of Technology Indore. She has several patents in applications of machine learning and is a recognized leader in the field.

Talk Track: Case Study

Talk Technical Level: 6/7

Talk Abstract:
Stitch Fix utilizes a sophisticated multi-tiered recommender system stack, encompassing feature generation, scoring, ranking, and business policy decision-making. This presentation delves into the training architecture of the scoring model, a deep learning model that predicts the likelihood of a user purchasing an item. I will walk through our journey, detailing the transition from training on a single GPU to leveraging multiple GPUs through pytorch’s Distributed Data Parallel (DDP) strategy. Additionally, I will share empirical results highlighting the efficiency of GPU utilization as we scale up with DDP across multiple GPUs.

What You’ll learn:
I would like to show the implementation details of moving from using one GPU to multiple GPUs for model training. I hope this will give the attendees enough knowledge to implement it themselves, allowing them to train bigger (and better) machine learning models.

Talk: Evolution of ML Training and Serving Infrastructure at Pinterest

Presenter:
Aayush Mudgal, Senior Machine Learning Engineer, Pinterest

About the Speaker:
Aayush Mudgal is a Senior Machine Learning Engineer at Pinterest, currently leading the efforts around Privacy Aware Conversion Modeling. He has a successful track record of starting and executing 0 to 1 projects, including conversion optimization, video ads ranking, landing page optimization, and evolving the ads ranking from GBDT to DNN stack. His expertise is in large-scale recommendation systems, personalization, and ads marketplaces. Before entering the industry, Aayush conducted research on intelligent tutoring systems, developing data-driven feedback to aid students in learning computer programming. He holds a Master’s in Computer Science from Columbia University and a Bachelor of Technology in Computer Science from Indian Institute of Technology Kanpur.

Talk Track: Case Study

Talk Technical Level: 5/7

Talk Abstract:
Join us for an insightful talk as we delve into the fascinating evolution of training and serving infrastructure at Pinterest Ads over the past 5+ years. Witness the remarkable progression from logistic regression-based models to the cutting-edge implementation of large transformer-based models, efficiently served using GPU technology. Throughout this transformative journey, we encountered numerous challenges and invaluable lessons that have shaped the very core of this critical business undertaking. Prepare to be inspired by our experiences as we share the triumphs and tribulations that ultimately led to a revolution in Pinterest Ads’ capabilities.

What You’ll Learn:
1. How to best structure training and serving infrastruture.
2. How to balance infrastructure costs and performance
3. Learn from real industrial system serving users at scale and the design choices that were made.

Talk: Amumu Brain; How League of Legends Uses Machine Learning an Applied Data Science

Presenter:
Ian Schweer, Staff Software Engineer, Riot Games

About the Speaker:
Ian is a staff software engineer at Riot Games, working on the League Data Central team. Along with his team, Ian ships Machine Learning and Data products to millions of league of legends and tft players including in game recommendations, player behavior models, and internal decision science to help make the game a better place for all. Ian has worked on large data systems at Adobe and Doordash before coming to Riot Games. In his free time, he plays in metal bands and hangs out with his 2 year old daughter.

Talk Track: Case Study

Talk Technical Level: 4/7

Talk Abstract:
League of legends faces lots of interesting problems in the data space that are unique due to the video game aspect. How do you deploy and train models in a binary video game? What is the fundamental data model? How has the data and ML stack changed since the league’s inception in 2009? How do you do player-facing ML (Lane detection, feeding detection, etc.) and decision science at this scale?

What You’ll Learn:
My goal is to teach a very pragmatic approach to shipping machine learning in a more constrained environment. This will include problems around network saturation, how to measure at player scale, and the challenges of interpretability for game design.

Workshop: Build Your Own ChatGPT with Open Source Tooling

Presenter:
Andreea Munteanu, AI/ML Product Manager, Canonical

About the Speakers:
I am a Product Manager at Canonical, leading the MLOps area. With a background in Data Science in various industries, such as retail or telecommunications, I used AI techniques to enable enterprises to benefit from their initiatives and make data-driven decisions. I am looking to help enterprises get started with their AI projects and then deploy them to production, using open-source, secure, stable solutions.

I am a driven professional, passionate about machine learning and open source. I always look for opportunities to improve, both myself and people within the teams that I am part of. I enjoy sharing my knowledge, mentoring young professionals and having an educational impact in the industry.

Talk Track: Workshop

Talk Technical Level: 5/7

Talk Abstract:
LLMs are gaining huge popularity with projects such ChatGPT, LLaMA or PaLM being open to everyone. Yet, enterprises feel overwhelmed by the large number of applications that requires, at first sight, a complete reshuffle of the existing infrastructure, in order to accommodate the needs for powerful cloud computing.

Depending on the industry, there are various use cases, as well as legacy architectures in place. Generative AI, however, can be deployed on any environment, whether it ia a public or private cloud. From the very beginning, when doing estimations of the cluster size, to upper layers in the stack where inference infrastructure is considered, there are a bunch of key factors such as GPU types, MLOps tooling or artefact types that influence the infrastructure of the project.

This talk will cover how you can build your infrastructure for a generative AI project, with a focus on building your own conversational assistant. It will go through the entire stack, including hardware and software applications, that cover the entire machine learning lifecycle, focusing on open source tooling and models. The session will feature a case study by a finserv company that build their own chatbot, using their data. After this presentation, you will understand how you can build and automate your infrastructure for a genAI project, using open source tooling.

What You’ll Learn:
fine tuning LLMs using open source tooling

Workshop: Retrieval Augmented Generation (RAG) with LangChain: “ChatGPT for Your Data” with Open-Source Tool

Presenters:
Dr. Greg Loughnane, Founder & CEO, AI Makerspace | Chris Alexiuk, Head of LLMs, AI Makerspace

About the Speakers:
Dr. Greg Loughnane is the Founder & CEO of AI Makerspace, where he serves as lead instructor for their LLM Ops: LLMs in Production course. Since 2021 he has built and led industry-leading Machine Learning & AI bootcamp programs. Previously, he has worked as an AI product manager, a university professor teaching AI, an AI consultant and startup advisor, and ML researcher. He loves trail running and is based in Dayton, Ohio.

Chris Alexiuk, is the Head of LLMs at AI Makerspace, where he serves as a programming instructor, curriculum developer, and thought leader for their flagship LLM Ops: LLMs in Production course. During the day, he’s a Founding Machine Learning Engineer at Ox. He is also a solo YouTube creator, Dungeons & Dragons enthusiast, and is based in Toronto, Canada.

Talk Track: Workshop

Talk Technical Level: 4/7

Talk Abstract:
Retrieval Augmented Generation (RAG) – or “ChatGPT for your private data” is the most popular LLM application being built today. RAG systems are question-answering tools that return coherent, fact-checked answers. These answers are used to augment the initial question/prompt before it is fed into an LLM. During this workshop, we will walk through each component of a simple RAG system. You’ll learn about how vector stores, embedding models, and LLMs are held together with LLM Ops infrastructure, and you’ll get all of the code to do it yourself in no time! We will also evaluation of RAG systems and the emerging best-practices for optimizing the quality of RAG outputs. We will use LangChain tooling to embed own own documents using a [leading embedding model](https://huggingface.co/blog/mteb), which we’ll store in a Pinecone vector database, and following context retrieval we’ll pass our augmented prompts to Llama 2!

All demo code will be provided via GitHub and/or Colab!

What You’ll Learn:
To understand how to use LangChain to build complex LLM applications.
To build “ChatGPT for their own data”
To understand how LangSmith can be used for productionizing LLM apps.
What “LLM Ops” actually means!

Workshop: Deploying Generative AI Models: Best Practices and An Interactive Example

Presenter:
Anouk Dutree, Product Owner, UbiOps

About the Speaker:
Anouk is the Product Owner at UbiOps. She studied Nanobiology and Computer Science at the Delft University of Technology, and did a Master’s in Game Development at Falmouth University, which spiked her interest in Machine Learning. Next to her role at UbiOps, she frequently writes for Towards Data Science about various MLOps topics and she co-hosts the biggest Dutch data podcast, de Dataloog. Her efforts in tech have been awarded twice with the T500 award, in both 2020 and 2021

Talk Track: Workshop

Talk Technical Level: 4/7

Talk Abstract:
Generative AI models are all the hype nowadays, but how do you actually deploy them in a scalable way? In this talk we will discuss best practices when moving models to production, as well as show an interactive example of how to deploy one using UbiOps. UbiOps is a serverless and cloud agnostic platform for AI & ML models, built to help data science teams run and scale models in production. We will pay special attention to typical hurdles encountered in deploying (generative) AI models at scale. Python knowledge is all you need for following along!

What You’ll Learn:
Deployment at scale doesn’t have to be difficult. Participants will learn how to deploy a generative AI model to the cloud themselves, and how to select the right hardware for your use case (CPU,GPU,IPU etc.).

Prerequisite Knowledge:
Python knowledge and a basic understanding of AI/ML models

Workshop: LLMs in Practice: A Guide to Recent Techniques and Trends

Presenters:
Ville Tuulos, CEO, Outerbounds | Eddie Mattia, Data Scientist, Outerbounds

About the Speakers:
Ville has been developing infrastructure for machine learning for over two decades. He has worked as an ML researcher in academia and as a leader at a number of companies, including Netflix where he led the ML infrastructure team that created Metaflow, a popular open-source framework for data science infrastructure. He is a co-founder and CEO of Outerbounds, a company developing modern human-centric ML. He is also the author of the book, Effective Data Science Infrastructure, published by Manning.

Eddie Mattia is a data scientist at Outerbounds who began using Python to teach applied math in grad school. Since then, Eddie has worked at startups and at Intel building machine learning software

Talk Track: Workshop

Talk Technical Level: 5/7

Talk Abstract:
In this workshop, attendees will learn about methods for working with LLMs. Our stories will be guided by examples you can run on your laptop or in a (free) hosted cloud environment provided to attendees. Developers will expand their awareness of how researchers and product designers are working with LLMs, with emphasis on connecting high-level concepts such as fine-tuning and vector databases to the fundamental math and APIs data scientists should understand. Business-minded executives can either get hands – on or follow the higher-level stories to deepen their sense of what is possible with LLMs, the technicalities behind risks they introduce, and how they fit into the arc of ML. The primary value of this workshop will be as a guide to help teams set reasonable goals in the complex and fast-moving world of LLMs, and understand what you need to successfully support your team’s next LLM projects.

What You’ll Learn:
There are cheap (e.g., APIs) and expensive (e.g., fine-tuning, training) ways to build on top of LLMs. The methods you choose have consequences in apps you can build and how your dev team works. We will learn how to think about these choices as we develop basic apps you can use as templates for future genAI projects. Learners have the option to follow along in a provided dev environment where we will unpack these choices and make the tradeoffs and decision space concrete.

Prerequisite Knowledge:
– Basic knowledge of Python
– Ability to use the command line
– Ability to use common ML algorithms in a notebook environment

Workshop: Applying GitOps Principles at Every Step of An E2E MLOps Project - An Interactive Workshop

Presenter:
Tibor Mach, Machine Learning Solutions Engineer, DVC

About the Speaker:
Tibor Mach is a Machine Learning Solutions Engineer at Iterative.ai He has been working in ML and MLOps in the past 5 years. Tibor has a Ph.D in mathematics from the University of Göttingen and had published papers in the field of probability theory prior to refocusing to ML.

Talk Track: Workshop

Talk Technical Level: 4/7

Talk Abstract:
With the emergence of IaC (infrastructure as code) tools, we have seen GitOps become an increasingly popular DevOps pattern that facilitates automation, reproducibility, and security. While hugely beneficial, applying the same principles in MLOps is not straightforward due to the specific aspects of the field such as the need to work with large amounts of data and the experimental nature of ML development. In this talk, we will see how we can bridge these gaps by using tools such as DVC. Step by step, we will create an end-to-end MLOps pipeline which is centered around the git repository as its single source of truth.

What You’ll Learn:
In this largely interactive workshop you can learn how you can use your git repositories to keep track of your ML experiments, version data and models, maintain a model registry and handle model deployment

Prerequisite Knowledge:
Basics of working with git and conceptual understanding of GitHub Actions or GitLab CI.

Workshop: Learn Your Codebase: Fine-tuning CodeLlama with Flyte… to Learn Flyte

Presenter:
Niels Bantilan, Chief ML Engineer, Union.ai

About the Speaker:
Niels is the Chief Machine Learning Engineer at Union.ai, and core maintainer of Flyte, an open source workflow orchestration tool, author of UnionML, an MLOps framework for machine learning microservices, and creator of Pandera, a statistical typing and data testing tool for scientific data containers. His mission is to help data science and machine learning practitioners be more productive.

He has a Masters in Public Health with a specialization in sociomedical science and public health informatics, and prior to that a background in developmental biology and immunology. His research interests include reinforcement learning, AutoML, creative machine learning, and fairness, accountability, and transparency in automated systems.

Talk Track: Workshop

Talk Technical Level: 5/7

Talk Abstract:
Today, foundation LLMs can only be trained by a handful of organizations possessing the compute resources required to pre-train models with more than a hundred billion parameters over internet-scale data. These foundation models are then fine-tuned by the wider ML community for specific applications. Even though fine-tuning can be more compute and memory efficient than full parameter tuning, a significant challenge to fine-tuning is provisioning the appropriate infrastructure.

In this session, Niels will demonstrate how to use Flyte, a Linux Foundation open-source orchestration platform to fine-tune a LLM on the Flyte codebase itself 🤯. Flyte allows for the declarative specification of the infrastructure needed for a broad range of ML workloads, including fine-tuning LLMs with limited resources by leveraging multi-node, multi-gpu distributed training.

What You’ll Learn:
Attendees will gain hands-on experience using Flyte to leverage state-of-the-art deep learning tools such as `torchrun` distributed training, LoRA, 4/8-bit quantization, and FSDP, while benefiting from Flyte’s reproducibility, versioning, and cost management capabilities. At the end of this talk, you’ll be able to take the code and adapt it to learn your own code base to help to answer user-support questions, create boilerplate starter code, or whatever downstream task you’re interested in!

Prerequisite Knowledge:
Intermediate Python
Intermediate Machine Learning
Familiarity with Command-line Tools

Workshop: Feature Stores in Practice: Train and Deploy an End-to-End Fraud Detection Model with Featureform, Redis, and AWS

Presenter:
Simba Khadder, Founder & CEO, Featureform

About the Speaker:
Simba Khadder is the founder & CEO of Featureform. He started his ML career in recommender systems where he architected a multi-modal personalization engine that powered 100s of millions of user’s experiences. He later open-sourced and built a company around their feature store. Featureform is the virtual feature store. It enables data scientists to define, manage, and serve model features using a Python API. Simba is also a published astrophysicist, an avid surfer, and ran a marathon in basketball shoes.

Talk Track: Workshop

Talk Technical Level: 4/7

Talk Abstract:
In this workshop, we’ll go through the process of building a fraud detection model from scratch using Featureform’s open-source feature store alongside a handful of other tools like Redis and Sagemaker. We’ll both train and deploy the model through this workshop. We’ll deep dive into where feature stores fit into the MLOps stack, the value they provide, and how to use them in practice.

What You’ll Learn:
Participants will learn how to:
– Use Featureform to build, manage, and serve their model features from fraud detection data.
– Use Redis, Spark, and Sagemaker to train and deploy a random forest model.
– Use Terraform and other best practice DevOps, DataOps, and MLOps through the process.

Prerequisite Knowledge:
Basics of cloud, networking, databases, and machine learning.

Workshop: Applying Responsible AI with the Open-Source LangTest Library

Presenter:
David Talby, CTO, John Snow Labs

About the Speaker:
David Talby is the Chief Technology Officer at John Snow Labs, helping companies apply artificial intelligence to solve real-world problems in healthcare and life science. David is the creator of Spark NLP – the world’s most widely used natural language processing library in the enterprise. He has extensive experience building and running web-scale software platforms and teams – in startups, for Microsoft’s Bing in the US and Europe, and to scale Amazon’s financial systems in Seattle and the UK. David holds a Ph.D. in Computer Science and Master’s degrees in both Computer Science and Business Administration. He was named USA CTO of the Year by the Global 100 Awards and GameChangers Awards in 2022.

Talk Track: Workshop

Talk Technical Level: 5/7

Talk Abstract:
While there’s a lot of work done on defining the risks, goals, and policies for Responsible AI, less is known about what you can apply today to build safe, fair, and reliable models. This session introduces open-source tools and examples of using them in real-world projects – to address three common challenges.

The first is robustness – testing and improving a model’s ability to handle accidental or intentional minor changes in input that can uncover model fragility and failure points. The third is bias – testing that a model performs equally across gender, age, race, ethnicity, or other population groups. The third is data leakage, in combination with leakage caused by using personally identifiable information in training data. The open-source LangTest library is used to demonstrate how to generate tests, run tests, augment data, and integrate these evaluations into MLOps workflows.

This session is intended for data science practitioners and leaders who need to know what they can do today to build AI & LLM applications that work safely and reliably in the real world.

What You’ll Learn:
This session is intended for data science practitioners and leaders who need to know what they can & should do today to build AI systems that work safety & correctly in the real world.

Background Knowledge:
Basic familiarity with machine learning is assumed.

Workshop: Introduction to Building ML Microservices: A Hands-On Approach with Examples from The Music Industry

Presenter:
Ramon Perez, Developer Advocate, Seldon

About the Speaker:
Ramon is a data scientist and educator currently working in the Developer Relations team at Seldon in London. Prior to joining Seldon, he worked as a freelance data professional and as a Senior Product Developer at Decoded, where he created custom data science tools, workshops, and training programs for clients in various industries. Before Decoded, Ramon wore different research hats in the areas of entrepreneurship, strategy, consumer behavior, and development economics in industry and academia. Outside of work, he enjoys giving talks and technical workshops and has participated in several conferences and meetup events.

Talk Track: Workshop

Talk Technical Level: 3/7

Talk Abstract:
Serving a machine learning model is not particularly easy, especially if we add two or three models in parallel to the mix, in which case, a single model deployment recipe might start to crumble. To tackle the challenges around serving individual or multiple models in production, we have handy tools like MLServer and Seldon Core. The former is a python library that allows us create machine learning microservices with one or multiple models in the same service, and the latter allows us to build simple-to-complex inference graphs that can help us handle A/B testing, shadow and canary deployment, feature transformations, and model monitoring. If you want to learn how to use open-source tools to build microservices based on your different use cases and model recipes, come and join this hands-on workshop and get started with several of the key steps in the machine learning workflow as we walk through fun examples from the broader music industry.

What You’ll Learn:
The core of the workshop will teach participants how to create machine learning microservices and inference graphs, and how to monitor the predictions made by these services. The main use case we’ll follow throughout the workshop comes from the music industry, so this will be a fun and content-rich 3 hours to go through.

Throughout the workshop, we will be building a creative ML platform in several incremental steps. In the first 50 minutes of the workshop, we will set up the user interface and the back-end our application, and then we’ll spin up the first model we will interact with. In the second 50-minute section, we will start adding different functionalities to our platform by running new machine learning models inside our inference server. Lastly, we’ll create different replicas of each model, develop an inference graph to come up with unique tunes, and conduct AB testing on our service to assess and evaluate the output of different models when compared with real songs.

Within our 3 hours together, we’ll have two 10- to 15-minute breaks and there will be plenty of exercises for participants to complete.

Workshop: Finetuning a Large Language Model on A Custom Dataset

Presenter:
Aniket Maurya, Developer Advocate, Lightning AI

About the Speaker:
Aniket is a Developer advocate at Lightning AI. He is an open source enthusiast and contributor to some popular repos like Lit-GPT and Gradsflow.

Talk Track: Workshop

Talk Technical Level: 5/7

Talk Abstract:
This is a hands-on workshop for finetuning large language models using custom dataset. By the end of this workshop, you will learn about parameter efficient finetuning, optimised inference and tricks to finetune models at scale.

What You’ll Learn:
Parameter efficient finetuning and LLM optimisations for very large models.

Prerequisite Knowledge:
Python, PyTorch basics

Workshop: Avoid ML OOps with MLOps: A Modular Approach to Scaling Forethought’s E2 E Ml Platform

Presenter:
Salina Wu, Senior Machine Learning Infrastructure Engineer, Forethought

About the Speaker:
Salina Wu is a Sr. Machine Learning Infrastructure engineer at Forethought.ai. She works closely with the Machine Learning team to build and maintain their end-to-end training, serving, and data infrastructures. She is particularly motivated by introducing new ways to improve efficiency and reduce cost across the ML space. When not at work, Salina enjoys surfing, pottery, and being in nature.

Talk Track: Workshop

Talk Technical Level: 5/7

Talk Abstract:
As Machine Learning becomes more ubiquitous in business and product applications, the need for a cost-efficient, scalable, and automated infrastructure to support the end-to-end ML lifecyle becomes mission critical. However, a scalable and reusable ML Ops platform is often an afterthought in productionizing ML models, due to urgency of business needs and lack of resources or experience. A very common scenario is for ML Ops to be ad-hoc and de-centralized, with no good way to reproduce or automate ML processes. It can be challenging, especially for smaller teams, to identify and foresee specific ML Ops needs and understand how to address them.

Forethought is an enterprise company building AI-powered customer experience (CX) solutions. Our products require training customer-specific language models and deploying them on low-latency, high-uptime endpoints. With ML at the heart of our business, our infrastructure supporting it is pivotal to our growth and success. At Forethought, we took a close look at our initial ML infrastructure, aiming to identify key areas of improvement and anticipate future requirements. Through a step-by-step approach, we gradually replaced our existing infrastructure with improved, modular components to arrive at a much more mature system. This case study will dive into which areas we identified as critical to replace as well as the steps we took to enhance them. In particular, we will look at the following:

– Streamlining ML training and migrating to the Sagemaker training platform
– Achieving efficient model serving with Sagemaker Serverless and Multi-Model Endpoints
– Orchestrating our ML processes with automated pipelines on Dagster
– Centralizing ML feature engineering across our datalake using Spark
– Building intuitive model management tooling with Retool

What You’ll Learn:
– Understanding the different components of a solid ML infrastructure
– Identifying and proactively addressing bottlenecks and opportunities for growth in your ML lifecycle
– Learning how to improve and migrate your ML infrastructure in stages
– Understanding the goals and best practices of a stable end-to-end ML infrastructure

Workshop: Lessons Learned: The Journey to Real-Time Machine Learning at Instacart

Presenter:
Guanghua Shu, Staff Machine Learning Engineer, Instacart

About the Speaker:
Guanghua Shu is a Staff Machine Learning Engineer at Instacart, where he focuses on building end-to-end machine learning solutions to gain actionable insights from data in the e-commerce domain. Guanghua has spent over six years in applied machine learning, and worked on recommender systems for product recommendation, leveraging ML to improve cloud security, and Data/AI platforms.

Guanghua holds a PhD in ECE from University of Illinois at Urbana-Champaign. He has published over 30 research papers and received over 10 patents. Through academia and industry experience, Guanghua has explored different abstract levels of the technology stack, ranging from ASIC design, computer architecture, distributed software systems, and applied machine learning. He believes in the power of technology and its outsized impact on business, society and beyond.

Talk Track: Workshop

Talk Technical Level: 4/7

Talk Abstract:
Instacart incorporates machine learning extensively to improve the quality of experience for all actors in our “four-sided marketplace” — customers who place orders on Instacart apps to get deliveries in as fast as 30 minutes, shoppers who can go online at anytime to fulfill customer orders, retailers that sell their products and can make updates to their catalog in real time, and the brand partners that participate in auctions on the Instacart Advertising platform to promote their products.
A typical shopping journey at Instacart is powered by hundreds of machine learning models. Many decisions/actions happen in real time, which means leveraging machine learning in real-time can provide significant value to the business. One of the major changes we have gone through is transitioning many of our batch-oriented ML systems into real-time. In this talk, we describe ML platform at Instacart with a focus on the journey of real-time ML. We will discuss both fundamental infrastructures and important use cases, review main challenges and decisions, and draw important lessons that could help others learn from our experience.

What You’ll Learn:
Key problems to consider for real-time ML.
Important foundations to support real-time ML in a e-commerce platform.
Avoid pitfalls and take aways good lessons in building real-time ML from our experience.

Workshop: QA in ML

Presenter:
Serg Masis, Lead Data Scientist & Bestselling Author, Syngenta

About the Speaker:
Bestselling author of ML/AI books. Lead Data Scientist at multinational agribusiness company creating models for sustainable practices in agriculture.

Talk Track: Workshop

Talk Technical Level: 2/7

Talk Abstract:
Trust is mission-critical for any technology, so if AI/ML solutions are to supplant and complement software, AI must reach the reliability standards currently expected from software. The difference is Quality Assurance (QA) has existed in software for three decades, and the burgeoning field of ML has barely begun to perform quality controls:
1) We will take a journey through the history of QA, discuss why it is crucial, and what lessons from other disciplines and industries we can apply to machine learning.
2) Then, will discuss what important role Explainable AI methods, not to mention best practices in MLOps, data engineering, and data science, can play.
3) Lastly, we will discuss the challenge ahead. Given the many steps in an Machine learning (ML) and the many qualities to assess in an ML model, choreographing and standardizing tasks in a QA effort is a challenging undertaking. New roles for ML QA will likely appear within DevOps, SecOps, and MLOps teams to ensure increased reliability and robustness. Still, also, the roles of data scientist and Machine Learning engineer will evolve to enhance quality.
Thus session is ultimately about what business stakeholders and practitioners can do to make AI/ML more trustworthy to the end-users of this technology.

What You’ll Learn:
– What the history of quality assurance (QA) teaches us about how QA can be implemented in ML.
– What tools and roles already exist in ML that can enforce QA.
– What’s missing to make QA work much better.

Workshop: Learning from Extremes: What Fraud-Fighting at Scale Can Teach Us About MLOps Across Domains

Presenter:
Greg Kuhlmann, CEO, Sumatra

About the Speaker:
Greg Kuhlmann is Co-founder and CEO of Sumatra, a realtime customer data platform that helps growth teams optimize conversions through on-site personalization and recommendations. He formerly led data science teams for the App Store and Apple Pay. He holds a PhD in machine learning from UT Austin.

Talk Track: Workshop

Talk Technical Level: 5/7

Talk Abstract:
The engineers behind large-scale anti-fraud platforms, faced with extreme demands for low-latency inference, feature freshness, and agile redeployment, have been the quiet pioneers at the cutting edge of MLOps. One might assume the architectures and practices developed for these intense problems would be overkill in less operationally-demanding domains. However, we will challenge this assumption, and discuss how the real-time-first approach taken by these systems actually simplifies architectures by eliminating many complex pipelines. Further, we’ll show how the observability and replay technlogies developed to respond quickly to unpredictable attacks can be applied broadly to make ML teams more agile across the board.

What You’ll Learn:
– The canonical architecture for modern, large-scale, real-time fraud prevention systems
– A comparison of the “”real-time-first”” vs. “”make-batch-faster”” approaches
– How log-time denormalization, unified online/offline feature transformation engines, and backfill on demand, are the keys to rapidly deploying model improvements in non-stationary domains

Workshop: How to Design and Build Resilient Machine Learning Systems

Presenter:
Dan Shiebler, Head of Machine Learning, Abnormal Security

About the Speaker:
Hi, I’m Dan Shiebler. I like math, history podcasts, fantasy novels, riding my bicycle, and traveling. I live in NYC.

Today I work as the Head of Machine Learning at Abnormal Security. I lead our team of 40+ detection engineers to build AI systems that fight cybercrime. We use a combination of foundational data engineering and advanced ML to detect and remediate cyberattacks. Our technology protects many of the world’s largest companies.

Previously, I managed the Web Ads Machine Learning team at Twitter. Before that I worked as a Staff ML Engineer at Twitter Cortex and a Senior Data Scientist at TrueMotion.

I’ve also spent some time in Academia. My PhD at the University of Oxford focused on applications of Category Theory to Machine Learning (advised by Jeremy Gibbons and Cezar Ionescu). Before that I worked as a Computer Vision Researcher at the Serre Lab.

Talk Track: Workshop

Talk Technical Level: 4/7

Talk Abstract:
The real world is messy. Systems fail, pipelines break, services go down, engineers push bugs, and users behave erratically. Software is hard exactly because these problems always happen. Effective systems must gracefully handle these events and smoothly degrade without catastrophic failure.

Unfortunately, ML systems are more likely to break than bend. Just like a boxer who only punches a bag will fail in the ring, an ML model that only learns with clean data may fail in production. Most ML models are trained with clean data, and when failures occur feature distributions can shift in ways that the model has never seen during training. This can cause strange and unexpected behavior.

In this talk we will explore how to build resilience into ML systems. We will discuss several types of production-specific risks and how these risks tend to manifest. These risks are common across many domains, but we will primarily use examples from our experience at Abnormal Security to demonstrate how we can detect, mitigate, and overcome these risks.

What You’ll Learn:
How to design machine learning systems that are resilient to the kinds of problems that occur in production systems

Workshop: Spend Less Time Troubleshooting ML Production Issues

Presenter:
Alon Gubkin, CTO & Co-Founder, Aporia

About the Speaker:
Alon is the CTO and Co-Founder at Aporia, an ML Observability platform designed to empower organizations to trust their AI. Alon spent the last decade leading multiple software engineering teams, working closely with various organizations on their Data & Machine Learning platforms.

Talk Track: Workshop

Talk Technical Level: 4/7

Talk Abstract:
Business stakeholders are unhappy with the model decisions again? Manual triage takes up a lot of bandwidth from your team every single time? In this workshop, you’ll learn how ML leaders identify and troubleshoot ML issues in production faster than ever. By being more proactive about common types of ML-specific production issues such as model drift, you’ll be able to spend significantly less time troubleshooting and gain peace of mind to focus on cooler, mission-critical projects.

What You’ll Learn:
Spend less time and resources on troubleshooting ML production issues.
Model drift awareness and monitoring
Improving model decision processes

Prerequisite Knowledge:
Basic understanding of machine learning concepts including model development and deployment.
Experience with ML production environments and awareness of common issues such as model drift.
Familiarity with data science research processes and techniques.
Ability to apply analytical and problem-solving skills to identify and address production issues proactively.

Talk: LLMs from Hallucinations to Relevant Responses

Presenter:
Eddie Mattia, Data Scientist, Outerbounds

About the Speaker:
Eddie Mattia is a data scientist working on Metaflow and foundation models at Outerbounds. He began using Python to teach applied math in grad school. Since then, Eddie has worked at startups and at Intel building machine learning software.

Talk Abstract:
Present a taxonomy of methods for controlling LLMs. Listeners will learn the broad strokes of how apps based on retrieval-augmented generation (RAG) and instruction tuning work and where they fit into the big picture of generative AI. We focus on how these techniques can be used to make generated responses more relevant.

Talk: Building an End-To-End MlLOps Pipeline

Presenter:
Aurimas Griciūnas, Head of Product, Neptune AI

About the Speaker:
Aurimas has over a decade of work experience in various data-related fields: Data Analytics, Data Science, Machine Learning, Data Engineering, and Cloud Engineering. For a few years he also led teams working with Data and Infrastructure. Today, Aurimas is Head of Product at neptune.ai.

Talk Abstract:
The talk will be about MLOps and the lifecycle of ML projects. I will go through stages involved in the ML project lifecycle and some key highlights from each of them. I will also explain how CI/CD is different in Machine Learning project when compared to regular software and highlight how it evolves with the maturity of MLOps processes in the organisation. I will also ground the explanations with real life examples of building out MLOps capabilities, successes and failures.

Talk: Generative AI: The Open Source Way

Presenter:
Andreea Munteanu, AI/ML Product Manager, Canonical

About the Speaker:
Andreea is a Product Manager at Canonical, leading the MLOps area. With a background in Data Science in various industries, such as retail or telecommunications, Andreea used AI techniques to enable enterprises to benefit from their initiatives and make data-driven decisions. Andreea is looking to help enterprises get started with their AI projects and then deploy them to production, using open-source, secure, stable solutions.

Talk Abstract:
Generative AI is probably the topic of the year. Leaders across the world feel the pressure of missed opportunities related to the latest technology. Professionals are also left worried about the impact that latest technology will have on their roles. Data scientists and machine learning engineers are challenge and need to quickly upskill. With such a stretched picture in mind, will generative AI deliver up to the great promises that it has right now?

This lighting talk will depict the opportunities that open source gives to generative AI to accelerate innovation. Between anxiety and enthusiasms, the latest technologies bring a new angle to the market. Let’s learn together about genAI with open source: from models to tooling to applications

Talk: Supporting Community Competitions to Develop LLMs

Presenter:
Richard Izzo, Tech Lead, Lightning AI

About the Speaker:
Rick Izzo was a Ph.D. candidate at the University at Buffalo Endovascular Device Development laboratory before co-founding Tensor[werk] Inc, a startup focused on building core infrastructure to support ML model training & deployment. After developing RedisAI & Hangar (tensor storage & version control system), Tensor[werk] was acquired by Lighting AI, where he and the team have continued to build and improve the core infrastructure needed to develop and train ML/DL Models.

Talk Abstract:
Lightning AI Supported the NEURIPS LLM Challenge by providing the base LIT-GPT model upon which contestants were asked to improve training of & finetune. This talk describes what it’s like to support a supporting a community competition focused on llm efficiency; From setting up the leaderboards, to creating a model challenge that allows participants to demonstrate that you can meaningfully finetune a model on a single GPU (using quantization) in one day, and get to the best evaluation possible.

Talk: Feature Store are NOT about Storing Features

Presenter:
Simba Khadder, Founder & CEO, Featureform

About the Speaker:
Simba Khadder is the founder & CEO of Featureform. He started his ML career in recommender systems where he architected a multi-modal personalization engine that powered 100s of millions of user’s experiences. He later open-sourced and built a company around their feature store. Featureform is the virtual feature store. It enables data scientists to define, manage, and serve model features using a Python API. Simba is also a published astrophysicist, an avid surfer, and ran a marathon in basketball shoes.

Talk Abstract:
Feature Store is a misnomer. Their objective is not to simply store feature values but rather to facilitate the organization, collaboration, versioning, discovery, and serving of feature definitions. In this Ignite talk, we’ll break down the actual goal of feature stores, their value, and where they fit into the MLOps stack.

Talk: How NOT to Get ML Models into Production

Presenter:
Ryan Turner, ML Solutions Engineer, DVC

About the Speaker:
Ryan has worked as an ML engineer at companies like Uber and Twitter. He is now developing the platform at Iterative.AI. He grew up in Santa Cruz, CA. After spending a few years in the UK and then Canada, he now lives in Reno, NV.

Talk Abstract:
Ryan Turner will be presenting on the various pitfalls of productionizing ML models. Common issues include dependency management, correct testing processes, and over reliance on Python notebooks. There will be no shortage of satire and metaphor. The talk will draw upon the collective experience of several ML engineers at DVC.

Talk: The Evolution of ML Monitoring in Production: From ML 1.0 to LLMs

Presenter:
Gon Rappaport, MLOps Solutions Architect, Aporia

About the Speaker:
Gon Rappaport is an MLOps Solutions Architect at Aporia, the AI Performance Platform, providing Observability and Guardrails to drive responsible, high-performing AI products. A software engineer at heart, Gon works closely with ML teams in every industry, seamlessly blending his knack for meme-based executive reports with his commitment to overcoming the visible and hidden challenges of machine learning models in production.

Talk Abstract:
In this talk, we’ll explore the landscape of ML Monitoring in production, highlighting best practices for tracking real-world AI products like Recommender Systems and Fraud Detection models (“ML 1.0”) and LLM-based applications such as Chatbots. We’ll further explore the intricacies of LLMs, focusing on monitoring applications that use Retrieval Augmented Generation (enriching LLMs with external insights), LLMs serving as controllers capable of activating external APIs or other ML models, as well as LLM Guardrails.

Talk: LLMs, Big Data, and Audio: Breaching an Untapped Gold Mine

Presenter:
Jose Nicholas Francisco, ML Developer Advocate, Deepgram

About the Speaker:
Jose is a Developer Advocate at Deepgram, aiming to demystify the inner workings of AI. He has a background in software engineering, with projects focused on fraud detection and prevention. Jose earned a bachelor’s and master’s degree in computer science—with a specialization on AI and NLP—from Stanford University. He currently lives in San Francisco with his friends.

Talk Abstract:
Large language models like those in the GPT and Llama series are primarily trained on massive amounts of *text* data. However, the vast majority of language and communication doesn’t take place over text, but rather through voice. Cues in vocal tone carry information that the plaintext cannot convey—think about the last time you’ve witnessed or experienced a miscommunication over text/email/Slack. Thus, in this talk, I argue that training language models on audio data is the next step to improving them. Then, I’ll propose a way of integrating audio data with text data in a larger dataset that can then be used for training various LLMs.

Talk: Which Compilers Are Best for LLMs?

Presenter:
Daniel Lenton, CEO, Ivy

About the Speaker:
Daniel Lenton is the founder and CEO at Ivy, on a mission to accelerate AI development and deployment by unifying the fragmented AI stack. Prior to founding Ivy, Daniel was a PhD student in the Dyson Robotics Lab, where he conducted research at the intersection of Machine Learning, Computer Vision and Robotics. During this time, he also worked at Amazon on their drone delivery program. Prior to this, he completed his Masters in Mechanical Engineering at Imperial College, with Dean’s list recognition.

Talk Abstract:
We explore the current wave of new LLM compiler infra, including tools such as FlexGen, vllm, and ggml, as well as efficient kernels such as Flash Attention. We conclude with some general tips and tricks that we’ve uncovered while implementing LLMs efficiently.

Talk: Data Versioning in Generative AI: A Pathway to Cost-Effective ML

Presenter:
Dmitry Petrov, CEO, DVC

About the Speaker:
Dmitry Petrov is the CEO and co-founder of Iterative.ai, working on building data-centric MLOps tools. He’s an ex-data scientist at Microsoft with a Ph.D. in Computer Science and an active open-source contributor. He has written and open-sourced the first version of DVC.org (part of Iterative) – a data versioning and machine learning workflow management tool. He also implemented a wavelet-based image hashing algorithm (wHash) in the open-source library ImageHash for Python.

Talk Abstract:
For 5 years we have been building DVC and we know how data versioning helps teams. The evolving Generative AI workflows are different and require an evolution of versioning workflows to accomplish Generative AI goals. This new era thrives on vast amounts of unstructured data, which include everything from images, videos, and audio, to MRI scans, document scans, and plain text dialogues. This data, often scaling into billions of objects, together with the resource-consuming task of scoring models on expensive GPU hardware or using model APIs like ChatGPT, brings forth unique challenges in the field of data management and versioning.

In this talk, we will delve into data versioning in the context of generative AI. Our focus will be on strategies that assist businesses in minimizing their processing time and the volume of API calls to external models like ChatGPT, resulting in substantial cost savings. Furthermore, we will discuss effective methodologies for sharing datasets amongst ML researchers to promote seamless collaboration.

Lastly, we will examine the pivotal transformations generative AI has introduced to data versioning in the past year including annotations and embeddings versioning. Together, these insights will provide attendees with an in-depth understanding of the rapidly evolving data management landscape in the era of generative AI.