This year's confirmed keynote speakers are: Smitha Shyam, Uber; Soumith Chintala, Meta AI; Anca Dragan, UC Berkeley; Dario Gil, IBM; Kim Hazelwood, Meta AI; Greg Brockman, OpenAI; Ion Stoica, Anyscale; and Robert Nishihara, Anyscale.<br><br>Please check back for more details on each keynote, coming soon.

Ray Summit Keynotes

Ray AI Runtime (AIR) is an open-source toolkit for building end-to-end machine learning applications. By leveraging Ray and its library ecosystem, Ray AIR brings scalability and programmability to ML platforms. In this session, we'll be exploring Ray AIR functionality and features, along with some benchmarks demonstrating its utility.

Engineering Manager

Software Engineer

Introducing Ray AI Runtime

Production RL and decision-making with RLlib

One of the challenges of using deep learning in production is managing the cost of loading models for inference. In this talk, we'll show how you can reduce this cost almost to zero by leveraging features of PyTorch and Ray. We'll introduce the concept of zero-copy model loading, storing the weights of a deep learning model in shared memory so that any process with access to the shared memory segment can load the model nearly instantaneously. We'll show, with code examples, how to implement zero-copy model loading with PyTorch and Ray. Then we'll introduce our open-source library, zerocopy, which lets you apply zero-copy model loading to your PyTorch model by changing two lines of Python code in your Ray application. We'll finish off with a simple benchmark study that shows how zero-copy model loading lets you run NLP models with stateless Ray tasks instead of heavyweight actors, giving you a self-tuning model deployment that delivers dramatically better performance than a naive deployment of the same models to Ray Serve.

Principal Research Staff Member

Zero-copy model loading with Ray and PyTorch

Predibase is a new kind of low-code machine learning platform from the creators of Ludwig and Horovod, combining large-scale deep learning with state-of-the-art model architectures across a range of modalities, including NLP, computer vision, tabular, and hybrid data. Unlike existing AutoML solutions, Predibase provides a modular "lego brick" experience that allows data scientists and engineers alike to iteratively build the best model for their data without needing to master a low-level framework like PyTorch or TensorFlow. In this talk, we'll explain how Predibase uses Ludwig and Ray under the hood to deliver an end-to-end, cloud-native solution encompassing data processing, model training, hyperparameter optimization, and model serving, all done at scale. We'll also explore how Predibase provides a unique serverless abstraction over Ray, allowing us to support multi-cloud and multi-tenant deployments capable of handling many concurrent users and requests.

Chief Technology Officer

Predibase: A low-code deep learning platform built for scale

Shuffle is a key primitive in large-scale data processing applications. The difficulty of large-scale shuffle has inspired a myriad of implementations. While these have greatly improved shuffle performance and reliability over time, it comes at a cost: flexibility. We show that contrary to the popular wisdom, shuffle can be implemented with high performance and reliability on a general-purpose system for distributed computing: Ray. In this talk we present Exoshuffle, an application-level shuffle system that outperforms Spark and achieves 82% of theoretical performance on a 100TB sort on 100 nodes. In Ray 2.0, we have integrated Exoshuffle with the Datasets library to provide high-performance large-scale shuffle for ML users.

Large-scale data shuffle in Ray with Exoshuffle

In this introductory session, we'll discuss the motivation behind Ray Serve, who's using Ray Serve and why, and recent features and updates in Ray 2.0.

State of Ray Serve in 2.0

Hamilton (https://github.com/stitchfix/hamilton) is an open source, declarative, general purpose, dataflow micro-framework, written in Python. It was originally created to manage complexities of scaling a team along with a time-series feature engineering code base past thousands of features at Stitch Fix. At a high level, in this talk, we'll cover: what Hamilton is and why it was created, how to use it for feature engineering, and how you can scale computation easily with the out-of-the-box Ray integration. At a low level, through code in the slides and a quick demo, you'll walk away with an understanding how a Data Science team at Stitch Fix scaled their team and code base with Hamilton, what Hamilton is and the declarative API paradigm it prescribes as opposed to traditional approaches, and lastly how the Ray integration with Hamilton works and how you can utilize it.

Mgr. Data Platform

Scalable feature engineering with Hamilton on Ray

At Riot Games, we leverage large-scale deep reinforcement learning to build bots that play our games at various skill levels to provide additional signals to our designers and ensure that we are releasing the best experiences for our players. As part of this experience we've experimented with multiple methods for controlling and tuning game servers to get the most out of our training time while the bots are learning to play the game. With our game Team Fight Tactics, we were able to take a unique approach that leveraged large neural networks to predict outcomes rather than have to rely on the server. In this talk, we will discuss the supervised learning process that we took and how we simplified and scaled with Ray Data, Ray Train, and Ray Tune.

Head of Tech Research

Large-scale deep learning to augment production RL workloads

Closing Comments

Centered on Ray, the open-source distributed computing framework, Ray Summit brings together ML practitioners, data scientists, and more to discuss the growing demand for distributed computing.

Ray Summit 2022 Logo Dark Blue

Ray Summit 2022 Logo

Ray Logo

Anyscale

Missed us in San Francisco for Ray Summit 2022? Don't fret, you can watch your favorite sessions on demand! 

Presented by Anyscale

Ray Summit collage of speakers from event

Ray Summit collage of speakers from event, stacked vertically

Ray Summit 2022 - Keynote Speakers

Ray Summit is a great opportunity to level up your Ray skills through hands-on training sessions with the builders and maintainers of Ray. We’re offering several pre-conference training sessions on August 22. Space is limited — [reserve your spot now](https://www.myeventi.events/raysummit22/attendee/)!

Training image desktop

Training image mobile

As the annual user conference for the [open source Ray community](https://www.ray.io/community), Ray Summit is a unique opportunity to enhance your Ray skills, build your professional network, and learn what's on the Ray roadmap.

Extend what you can do with Ray by attending hands-on, expert-led workshops and deep dives

Mingle with the world’s top minds in distributed computing, machine learning, and AI

Learn what’s on the Ray roadmap and connect with the builders and maintainers of Ray and Ray libraries

Get ready for our epic community bash at the Art of Tinkering exhibit at the San Francisco Exploratorium on August 23 from 5:30 to 9:30 p.m. Come explore, play, and celebrate with the Ray Community. This is an exclusive opportunity to experience this amazing interactive exhibition before it closes in September.

Man taking a photo on his phone

Ready to get involved in the Ray community before the conference? Ask a question in the forums. Open a pull request. Or share why you’re excited with the hashtag __#RaySummit__ on Twitter.

Ray Summit 2022

Come connect with the global community of thinkers and disruptors who are building and deploying the next generation of AI and ML applications.

Ray Summit 2022 horizontal logo

chucks

beanbags

register-bottom-mobile

Ray Summit 2022 - Virtual Agenda Americas

Are you passionate about Ray, scalable AI and distributed computing? Join us at Ray Summit!

From the creators of Ray, Anyscale is a framework for building machine learning applications at any scale originating from the UC Berkeley RISELab.

Virtual Agenda - Americas

Virtual - Americas

Tuesday, August 23

Ray Summit Keynotes

Introducing Ray AI Runtime

Production RL and decision-making with RLlib

Zero-copy model loading with Ray and PyTorch

Predibase: A low-code deep learning platform built for scale

Large-scale data shuffle in Ray with Exoshuffle

State of Ray Serve in 2.0

Scalable feature engineering with Hamilton on Ray

Large-scale deep learning to augment production RL workloads

Closing Comments

Wednesday, August 24

Ray Summit Keynotes

Introducing Ray AI Runtime

Large-scale deep learning to augment production RL workloads

Large-scale data shuffle in Ray with Exoshuffle

Scalable feature engineering with Hamilton on Ray

Production RL and decision-making with RLlib

State of Ray Serve in 2.0

Predibase: A low-code deep learning platform built for scale

Zero-copy model loading with Ray and PyTorch

Closing Comments

Ready to Register?

Join the Conversation