Ray Summit 2020
2020-09-30..2020-10-01
-- online --
[Ray Summit 2020, Ray Summit 2021, Ray Summit 2024, Ray Summit 2026]
| title | author | length |
|---|---|---|
| Keynote: Anyscale Product Demo | Edward Oakes | 31m |
Demo builds a movie recommender end-to-end: color-palette KNN search with FAISS and a BERT-based plot model fine-tuned via Ray Tune, then deployed with Ray… | ||
| Keynote: Building a Fusion Engine with Ray | Charles He | 10m |
Ant Group builds a fusion engine on Ray to unify AI, graph, streaming and OLAP workloads and claims production success at scale. The stack runs on Ray with a… | ||
| Keynote: Programming the Cloud as Easily as Your Laptop | Ion Stoica | 17m |
Claims that training compute grows ~35x every 18 months and model size grew ~20x (200x in the last two years), while single‑core speed and even GPUs/TPUs… | ||
| Keynote: The Future of Ray | Robert Nishihara | 14m |
Community grows from a handful to over 350 contributors from 75 companies, and the release of Ray 1.0 locks core API stability while adding automatic… | ||
| Ray: A General Purpose Serverless Substrate? | Eric Liang | 26m |
Ray 1.0 positions Ray as a general-purpose serverless substrate that supports stateful actors, tasks, and a shared-memory object store so ML training and data… | ||
| Hyperparameter Tuning and Visualization in Deep Learning | Lukas Biewald | 25m |
Practical recipe for automated tuning: use Weights & Biases Sweeps with Ray Tune integration to run random, Bayesian, and population-based searches, add… | ||
| Easy Access to SOTA NLP Models with Ray and Hugging Face | Thomas Wolf | 25m |
Hugging Face demonstrates running pretrained Transformers with the Tokenizers and Datasets libraries plus Ray Tune hyperparameter search to fine‑tune on GLUE… | ||
| EfficientBERT: Trading off Model Size and Performance | Meghana Ravikumar | 21m |
Distills a fine-tuned BERT teacher into smaller student models on SQuAD 2.0, seeding students with DistilBERT weights and running a simultaneous architecture… | ||
| Building a Python Web Service with Ray | Philipp Moritz | 19m |
Shows how Ray integrates with Python web serving by using Ray tasks and actors to keep state, cache connections, and recover from failures; demonstrates… | ||
| Keynote: Machine Learning & A.I. At Uber | Zoubin Ghahramani | 17m |
Uber runs a production ML stack called Michelangelo platform that serves online and offline workloads, trains ~36,000 models per month, deploys ~12,000,000… | ||
| Keynote: Easier Machine Learning Thoughts From Scikit-Learn | Gael Varoquaux | 21m |
Scikit-learn focuses on concrete wins: algorithmic changes that speed real workloads, not vague scaling promises. Examples include randomized PCA (≈10x speedup… | ||
| Keynote: Scalable Python We Can Afford | Wes McKinney | 14m |
Pandas often requires 10× RAM because data must be RAM resident and many algorithms remain single threaded and inefficient for strings; scaling by naively… | ||
| Keynote: Artificial Innovation Accelerates Artificial Intelligence | David Patterson | 26m |
Machine learning training needs far more compute but Moore’s Law slows, so architects build domain-specific hardware. Google ships TPUv1–v3: 256x256 MACs with… | ||
| Keynote : Modern App Development | Adrian Cockcroft | 18m |
Serverless-first development claims dramatic speedups: teams build functional prototypes in a day and one team rewrote a stalled nine‑month project over a… | ||
| Keynote: ML for Systems and Chip Design | Azalia Mirhoseini | 20m |
Google Brain ML for Systems applies reinforcement learning and graph neural nets to real engineering tasks: device placement for neural nets, graph… | ||
| Keynote: AI for Financial Services | Manuela Veloso | 18m |
Company researchers present concrete systems: a realistic OTC market multi-agent reinforcement learning simulator trains market-maker policies using Ray RLlib… | ||
| Dask-on-Ray: Using Dask and Ray to Analyze Petabytes of Remote Sensing Data | Clark Zinzow | 29m |
A Dask-on-Ray scheduler converts each Dask task into a ray.remote call so full Dask graphs run on Ray’s decentralized scheduler and Plasma object store… | ||
| Ray Internals: A Peek at `ray.get` | Stephanie Wang | 22m |
Deep dive into what ray.get actually does: precise lifecycle of an object, how small objects stay at the owner while large objects live in a distributed store… | ||
| What's New with Ray Libraries: Tune | Richard Liaw | 22m |
Ray Tune 1.0 adds concrete, usable features: a toggleable population based training backend (async for speed, sync for stability) plus reproducible replay of… | ||
| Introducing Ray Serve: Scalable and Programmable ML Serving Framework | Simon Mo | 23m |
Ray Serve is a distributed Python web framework on Ray that replaces Flask-or-TFServing tradeoffs by letting HTTP APIs own preprocessing/postprocessing while… | ||
| AI In Production Key Challenges and Use Cases | Jonathan Berte | 25m |
Robovision builds a no-code platform that treats each model as one ray actor and composes them as ML framework independent compute cells inside… | ||
| Debugging and Observability for Distributed Ray Applications | SangBin Cho | 22m |
Describes concrete debugging and observability features added to Ray: a memory view that shows object references and call sites, custom metrics recorded via a… | ||
| Keynote: Every Five Decades: A New Field of Engineering | Michael I. Jordan | 14m |
Keynote: Every Five Decades: A New Field of Engineering - Michael I. Jordan, Distinguished Professor, University of California at Berkeley
Machine learning… | ||
| Project Zouwu: Scalable AutoML for Telco Time Series Analysis using Ray and Analytics Zoo | Ding Ding | 27m |
Project Zouwu delivers scalable AutoML for telco time series by wiring Ray Tune on Ray into Analytics Zoo so feature generation, model selection and… | ||
| Turbocharging State-of-the-art Natural Language Processing on Ray | David Talby | 30m |
A Python library named nlu wraps Spark NLP so users run production-grade models with a one-line API that returns whichever dataframe type is passed (pandas… | ||
| Leveraging Ray to Enhance Machine Learning Models in Finance | James Dunworth-Crompton | 21m |
Goldman Sachs builds RayLab (kaniko + Kubernetes + Jupyter Lab) to run Ray on private cloud and uses Tune config-driven experiments so research and production… | ||
| RLlib: Scalable RL for TensorFlow, PyTorch, and Beyond | Eric Liang | 14m |
RLlib presents a unified API and cluster runtime that runs TensorFlow and PyTorch at scale, with RLlib 1.0 claiming 100% PyTorch parity, NVF-MPO and Dreamer… | ||
| Distributed Black-Box Model Explanation with Ray | Alexandru Coca | 24m |
Parallelizes Kernel SHAP via Alibi across Ray actors to explain many independent predictions, packaging the explainer as a distributed actor pool and running… | ||
| AlphaDow: Reinforcement Learning for Industrial Production Scheduling | Adam Kelloway | 23m |
Reinforcement learning trains agents that generate production schedules for a 5-product plant, enforcing action masking to block illegal transitions and… | ||
| Ray + Arize - Why you Need Model Observability | Aparna Dhinakaran | 14m |
Demo shows a Lending Club loan-initiation fraud classifier served with Ray Serve and logged to Arize, capturing inputs, predictions and delayed actuals with… | ||
| Building Complex Data Analytics Pipelines with Ray | Qingqing Mao | 24m |
Describes using Ray to convert Python preprocessing, document-level entity extraction, corpus-level clustering and entity linking into remote tasks so data… | ||
| Connecting Reinforcement Learning to Simulation Software | Max Pumperla | 14m |
Pathmind connects AnyLogic simulation models to Ray RLlib by adding an AnyLogic plugin Pathmind Helper that exports a standalone model to a web app, trains… | ||
| Distributed Reinforcement Learning for Robotic Assembly | Rodger Luo | 21m |
Three projects show concrete results: Project Touch trains using Apex DDPG plus preserved human demos and deploys policies that succeed on 75° and 60° tight… | ||
| The Java API and Cross Language Programming on Ray | Hao Chen | 22m |
Ant Group reports a Java API released with Ray 1.0 that turns Java static methods and classes into Ray tasks and actors with ObjectRef futures, Maven… | ||
| Cloudstate—Towards Stateful Serverless | Jonas Bonér, James Roper | 25m |
Cloudstate claims to add managed, addressable state to serverless by delegating state to a runtime and supports event sourcing, CRDTs, and key/value backends… | ||
| Advanced Ray RLlib: New Features and New Possibilities | Sven Mika | 21m |
RLlib now ships PyTorch for all algorithms plus native TensorFlow 2 eager support and Windows wheels, and adds model-based methods Dreamer and MB-MPO. Dreamer… | ||
| Utilizing Reinforcement Learning Techniques for Supply Chain with Amazon SageMaker | Anna Luo | 31m |
Claim: Amazon SageMaker RL runs Ray RLlib and TensorFlow/PyTorch in managed containers, uses managed spot instances with checkpoint resume, and supports… | ||
| Using Ray On Large-scale Applications at Ant Group | Jiaying Zhou | 26m |
Ant Group replaces a glue of Flink/Spark/TF with Ray as a single engine, adding plugin runtimes for streaming, parallel batch and AI (TensorFlow-on-Ray)… | ||
| Project Bonsai: How Microsoft is Using Ray and RLlib Inside a Machine Teaching Service | Edilmo Palencia | 20m |
Project Bonsai uses RLlib and Ray for distributed training and scales many simulators to build deployable control models called "brains." A domain-specific… | ||
| Adopting Ray @ LinkedIn | Jonathan Hung, Nitin Pasumarthy | 29m |
Integrates Ray on YARN via TonY and Azkaban by having the TonY application master request containers, reserve ports, broadcast a head+worker topology, and… | ||
| Keynote: A Secure Collaborative Learning Platform | Raluca Ada Popa | 10m |
MC^2 is an open-source platform that runs secure collaborative learning by combining Python DSL policy-aware workflows with two execution backends MPC and… | ||
| Data Scientists are More Valuable than Hardware: The Design of Modin | Devin Petersohn | 24m |
Claim: data scientists are more valuable than hardware, so Modin makes pandas code run at cluster scale without rewriting. Modin is a drop-in pandas… | ||
| State Management in Distributed Online Learning with Ray | Edmon Begoli | 29m |
Experimental framework uses Ray actors and Ray Serve with gRPC to manage continuously updating distributed models for suicide prevention, overdose and… | ||
| ML Engineering Kubeflow | Pavel Dournov | 29m |
Kubeflow provides a Kubernetes-native pipeline system that composes containerized tasks into reproducible ML workflows using Kubeflow Pipelines and TFX… | ||
| Keynote: AlphaStar: How Google Infrastructure Made it Possible | Oriol Vinyals | 19m |
AlphaStar is a full StarCraft II agent built on Google-scale infrastructure that reached Grandmaster in human ladder play with no game restrictions. The… | ||
| Information Gain Regulation in Reinforcement Learning with the Digital Twins' Level of Realism In RlLib | Géza Szabó, József Pető | 20m |
Information Gain Regulation in Reinforcement Learning with the Digital Twins' Level of Realism In RlLib - Géza Szabó, Ericsson Research & József Pető, Budapest… | ||
| Human-First AI: Because AI Needs The Human Eye | Christopher Nguyen | 11m |
Claims an open-source Human-First AI framework that jumpstarts industrial ML where data is scarce by combining rule/physics-based data generators, an execution… | ||
| Distributed Deep Learning with Horovod on Ray | Travis Addair | 27m |
Horovod runs data-parallel training for TensorFlow PyTorch and MXNet with near 90% scaling efficiency (demonstrated on IBM Summit with ~27,000 GPUs and a… | ||
| nothing matches | ||