Ray: A Distributed Framework for Emerging AI Applications
arXiv:1712.05889
Abstract
The next generation of AI applications will continuously interact with the environment and learn from these interactions. These applications impose new and demanding systems requirements, both in terms of performance and flexibility. In this paper, we consider these requirements and present Ray---a distributed system to address them. Ray implements a unified interface that can express both task-parallel and actor-based computations, supported by a single dynamic execution engine. To meet the performance requirements, Ray employs a distributed scheduler and a distributed and fault-tolerant store to manage the system's control state. In our experiments, we demonstrate scaling beyond 1.8 million tasks per second and better performance than existing specialized systems for several challenging reinforcement learning applications.
17 pages, 14 figures, 13th USENIX Symposium on Operating Systems Design and Implementation, 2018
References in corpus (6)
- Caffe: Convolutional Architecture for Fast Feature Embedding
- Horovod: fast and easy distributed deep learning in TensorFlow
- Distributed Prioritized Experience Replay
- GraphLab: A New Framework For Parallel Machine Learning
- Massively Parallel Methods for Deep Reinforcement Learning
- Deep Learning with Dynamic Computation Graphs
Cited by in corpus (30)
- Tune: A Research Platform for Distributed Model Selection and Training
- Parsl: Pervasive Parallel Programming in Python
- Simple random search provides a competitive approach to reinforcement learning
- Qd-tree: Learning Data Layouts for Big Data Analytics
- numpywren: serverless linear algebra
- Effective Diversity in Population Based Reinforcement Learning
- Distributed Deep Reinforcement Learning: A Survey and A Multi-Player Multi-Agent Learning Toolbox
- AI in Human-computer Gaming: Techniques, Challenges and Opportunities
- HyperTendril: Visual Analytics for User-Driven Hyperparameter Optimization of Deep Neural Networks
- Efficient End-to-End AutoML via Scalable Search Space Decomposition
- ZOOpt: Toolbox for Derivative-Free Optimization
- Intelligence Beyond the Edge: Inference on Intermittent Embedded Systems
- Archipelago: A Scalable Low-Latency Serverless Platform
- Policy Gradient Search: Online Planning and Expert Iteration without Search Trees
- APACE: AlphaFold2 and advanced computing as a service for accelerated discovery in biophysics
- Learning data augmentation policies using augmented random search
- Teola: Towards End-to-End Optimization of LLM-based Applications
- Simple Improved Reference Subtraction for H4RG, H2RG, and H1RG Near-infrared Array Detectors
- Learning to rumble: Automated elephant call classification, detection and endpointing using deep architectures
- High Performance Dataframes from Parallel Processing Patterns
- A Framework for Democratizing AI
- Non-autoregressive Personalized Bundle Generation
- Asynchronous Stochastic Proximal Methods for Nonconvex Nonsmooth Optimization
- Towards Autonomous Pipeline Inspection with Hierarchical Reinforcement Learning
- Apache Submarine: A Unified Machine Learning Platform Made Simple
- Neural MMO v1.3: A Massively Multiagent Game Environment for Training and Evaluating Neural Networks
- POLO: a POLicy-based Optimization library
- Parallel and Multi-Objective Falsification with Scenic and VerifAI
- Mode and Ridge Estimation in Euclidean and Directional Product Spaces: A Mean Shift Approach
- A novel Cosmic Filament catalogue from SDSS data