·
AI Infrastructure & Inference
Richard Liaw
Product Leader / Co-creator of Ray Tune |Anyscale
Co-creator of Ray Tune, co-author of the Ray OSDI paper and O'Reilly's 'Learning Ray'. UC Berkeley RISELab researcher turned Anyscale product leader.
5papers
GitHubrichardliawResearch Papers
5 papersarXiv
Tune: A Research Platform for Distributed Model Selection and Training
1807.05118·2018-07-13
DOI
Ray: A Distributed Framework for Emerging AI Applications
·2018-10-08
DOI
RLlib: Abstractions for Distributed Reinforcement Learning
·2017-12-26
DOI
HyperSched: Dynamic Resource Reallocation for Model Development on a Deadline
·2019-11-20
DOI
RubberBand: Cloud-based Hyperparameter Tuning
·2021-04-21
Biography
Deep Research
Richard Liaw is a co-creator of Ray Tune and a Product leader at Anyscale, the company commercializing the Ray distributed computing framework. A UC Berkeley RISELab graduate researcher under Joseph Gonzalez, Ion Stoica, and Ken Goldberg, Liaw is a co-author of the foundational Ray paper (OSDI 2018, 1,560 citations), the RLlib reinforcement learning library (ICML, 1,006 citations), and the Tune hyperparameter tuning platform (1,079 citations). He co-authored 'Learning Ray' (O'Reilly, 2023) and has an h-index of 13 with over 4,300 total citations across 20 publications. His research spans distributed systems, hyperparameter optimization (HyperSched, RubberBand), and scalable reinforcement learning.
Distributed Hyperparameter TuningRay TuneRay Distributed FrameworkScalable Reinforcement Learning (RLlib)AutoML & Neural Architecture SearchCloud-Based Model TrainingKubernetes + Ray (KubeRay)Population-Based Training (PBT)Distributed Systems for MLOpen-Source AI Infrastructure
Timeline
11 Research11 total
2023
2022
2021
2020
2019
2018
Key Contributions
Ray Tune
Co-created Tune, a unified framework for distributed hyperparameter tuning and model selection that supports ASHA, BOHB, PBT, and integrates with PyTorch, TensorFlow, XGBoost, and Keras. Over 1,079 citations.
Ray (OSDI 2018)
Co-authored the foundational Ray paper introducing a general-purpose distributed framework for emerging AI applications with fault-tolerant primitives. 1,560 citations.
RLlib
Co-authored the scalable reinforcement learning library providing composable abstractions for multi-agent environments and policy serving. 1,006 citations.
Learning Ray (O'Reilly, 2023)
Co-authored the definitive O'Reilly book on Ray with Max Pumperla and Edward Oakes, covering distributed Python for machine learning including training and cluster management.
HyperSched
Developed a dynamic application-level resource scheduler that reallocates resources to best-performing hyperparameter trials under deadline constraints. Published at ACM SoCC 2019.
RubberBand
Cloud-based hyperparameter tuning system for cost-efficient model selection, published at EuroSys 2021.
Notable Quotes
“
Ray is a compute engine for scaling AI workloads, offering libraries for training, serving, and data processing.
“
Developing locally and scaling into large-scale contexts without needing a distributed systems PhD was a key design principle.
“
Ray Data makes batch inference and image processing pipelines 100 times easier than forcing them into weird setups.
12 sources(click to expand)
richardliaw on GitHub (114 repos, 269 followers)Richard Liaw on Semantic Scholar (20 publications, h-index 13, 4,347 citations)Richard Liaw on Google ScholarTune: A Research Platform for Distributed Model Selection and Training (arXiv)Ray: A Distributed Framework for Emerging AI Applications (OSDI 2018)Learning Ray: Flexible Distributed Python for Machine Learning (O'Reilly)Kubernetes Podcast Episode 235 - Ray & KubeRay with Richard LiawSoftware Engineering Daily - Ray Applications with Richard LiawRichard Liaw on Anyscale BlogRichard Liaw - Anyscale | LinkedInRichard Liaw on DBLPRichard Liaw - Engineering Manager @ Anyscale (Crunchbase)
Research generated March 19, 2026
AI Infrastructure & Inference/Richard Liaw
All Profiles