# Berkeley RISE Lab The Berkeley RISE Lab (Real-time Intelligent Systems and Environments Lab) is a UC Berkeley research group focused on building scalable, real-time AI systems, led by Ion Stoica, with origins in the AMPLab and RISELab projects. The Berkeley RISE Lab (Real-time Intelligent Systems and Environments Lab) is a research laboratory at the [University of California, Berkeley](https://www.wikiprompt.org/wiki/university-of-toronto) focused on developing systems and infrastructure for real-time artificial intelligence. Founded by computer scientist Ion Stoica, the lab continues the work of its predecessors, the AMPLab and RISELab projects, which created foundational technologies like Apache Spark and Apache Mesos. The lab's mission is to bridge the gap between the latest advances in [AI](https://www.wikiprompt.org/wiki/artificial-intelligence) and the scalable, low-latency systems needed to deploy them in production environments. The Berkeley RISE Lab operates under the umbrella of the [Berkeley Artificial Intelligence Research](https://www.wikiprompt.org/wiki/berkeley-ai-research) community but maintains its own distinct research agenda centered on the intersection of systems, networking, and machine learning. Its researchers focus on challenges such as distributed training, model serving, data management for AI, and the development of tools that make it easier to build and operate real-time intelligent applications. The lab is known for its open-source contributionsMK, which are widely adopted in industry and academia. ## Historical Roots and Evolution The lab traces its lineage to the AMPLab, which ran from 2011 to 2016 with funding from the [National Science Foundation](https://www.wikiprompt.org/wiki/amazon-web-services) and industrial partners. The AMPLab produced Apache Spark, a unified analytics engine, and Apache Mesos, a cluster manager. In 2017, the lab was renamed RISELab (Real-time Intelligent Secure and Explainable systems), with a focus on real-time, secure, and explainable AI systems. The current Berkeley RISE Lab was established in 2021, pivoting toward real-time intelligent systems with an emphasis on serving and inference. ## Research Areas The lab's research spans several key domains: - **Model Serving and Inference**: Developing systems like Ray Serve and vLLM, which optimize the deployment of [large language models](https://www.wikiprompt.org/wiki/large-language-model) and other [deep learning](https://www.wikiprompt.org/wiki/deep-learning) models with high throughput and low latency. vLLM, in particular, introduced PagedAttention, a memory management technique that has become widely influential. - **Distributed Computing**: Advancing the Ray framework, a unified platform for scaling Python and machine learning workloads, which provides primitives for distributed training, reinforcement learning, and hyperparameter tuning. - **Data Management for AI**: Building systems for feature stores, data preprocessing, and streaming, such as Tune and Modin, to handle the data-intensive pipelines that feed [machine learning](https://www.wikiprompt.org/wiki/machine-learning) models. - **Security and Privacy**: Exploring techniques for secure aggregation, differential privacy, and federated learning to protect sensitive data in AI systems. - **Systems for Emerging Hardware**: Collaborating with hardware vendors to optimize software stacks for new accelerators, such as [AMD](https://www.wikiprompt.org/wiki/amd) GPUs and [Intel](https://www.wikiprompt.org/wiki/intel) CPUs. ## Key Open-Source Projects The lab maintains several major open-source projects that form a cohesive ecosystem: - [Ray](https://www.wikiprompt.org/wiki/keras): A general-purpose distributed execution framework that supports both stateless and stateful computations. Ray is used by companies like [OpenAI](https://www.wikiprompt.org/wiki/openai) and [Anthropic](https://www.wikiprompt.org/wiki/anthropic) for training and serving. - [vLLM](https://www.wikiprompt.org/wiki/tensorflow): A high-throughput, memory-efficient serving engine for LLMs, released in 2023. It has become a standard tool for deploying models like GPT and Llama variants. - [Ray Tune](https://www.wikiprompt.org/wiki/pytorch): A Python library for scalable hyperparameter tuning, which integrates with [PyTorch](https://www.wikiprompt.org/wiki/pytorch) and [TensorFlow](https://www.wikiprompt.org/wiki/tensorflow). - Apache Spark: While no longer actively developed by the lab, Spark continues to be a foundational output from the AMPLab era, maintained by the Apache Software Foundation. ## Collaborations and Impact The Berkeley RISE Lab actively collaborates with industry partners and other research institutions. It has partnerships with [Amazon Web Services](https://www.wikiprompt.org/wiki/amazon-web-services), Microsoft Azure, and [Google Cloud](https://www.wikiprompt.org/wiki/google-cloud), among others, to ensure its systems are production-ready. The lab also works closely with [Stanford's AI Lab](https://www.wikiprompt.org/wiki/stanford-ai-lab) and [MIT's CSAIL](https://www.wikiprompt.org/wiki/mit-csail) on joint research initiatives. Its Ph.D. students and postdocs have gone on to found successful startups, such as Anyscale (commercializing Ray) and Determined AI (acquired by [Hugging Face](https://www.wikiprompt.org/wiki/hugging-face)). The lab's work has influenced the broader [generative AI](https://www.wikiprompt.org/wiki/generative-ai) ecosystem. For instance, vLLM's optimizations were quickly adopted by commercial inference providers, and Ray has become a de facto standard for distributed Python in [AI](https://www.wikiprompt.org/wiki/artificial-intelligence) applications. The lab regularly hosts workshops and publishes its research at top conferences like OSDI, NSDI, and NeurIPS. ## Leadership and People Ion Stoica, a professor of computer science at UC Berkeley, serves as the lab's director. He is also a co-founder of Anyscale and Databricks. Key faculty members include [Michael Jordan](https://www.wikiprompt.org/wiki/michael-jordan), who works on machine learning theory, and [Anima Anandkumar](https://www.wikiprompt.org/wiki/anima-anandkumar), who leads research on tensor methods and generative models. The lab has a rotating group of international visiting scholars and hosts a summer internship program for undergraduate researchers. ## Future Directions The lab continues to push the boundaries of real-time AI systems, with ongoing research into [reinforcement learning](https://www.wikiprompt.org/wiki/reinforcement-learning) serving, multi-agent systems, and adaptive resource management. As of 2025, its researchers are exploring how to reduce inference costs for [transformer](https://www.wikiprompt.org/wiki/transformer)-based models and how to make distributed training more resilient to hardware failures. The lab remains committed to open science, publishing all its code and most of its research papers freely online. --- Source: https://www.wikiprompt.org/wiki/berkeley-riselab License: CC BY-SA 4.0 (https://creativecommons.org/licenses/by-sa/4.0/) Last updated: 2026-09-09T01:56:53.403195+00:00