Michael Luo

I am a PhD student in the UC Berkeley EECS department advised by Prof. Ion Stoica. My research interests are in Artificial Intelligence and Systems. I am associated with SkyLab and Berkeley Artificial Intelligence Research (BAIR).

Before that, I earned my M.S. in EECS under Ion Stoica and Ken Goldberg from AUTOLab during 2021. I also earned a B.S. from UC Berkeley with a double major in EECS and Business Administration in 2020.

Email  /  CV  /  Google Scholar  /  LinkedIn  /  Github

Currently, my research involves building scalable systems for ML pracitioners that will fulfill the Sky Computing vision. We are open source. This research involves virtualizing GPUs to scale DL training to trillions of parameters and designing learnable scheduling policies for migrating jobs across different and regions clouds (incl. on-premise).

Previously,my masters's and undergraduate research primarily focused on practical problems and applications for reinforcement learning (RL), including NLP, query optimization for databases, and video streaming.

An end2end query optimizer trained via deep RL that exceeds the query-performance of expert solvers by up to 2.8x.

Safe RL algorithm that meta-learns from offline datasets to safely adapt to unseen environments.

An algorithm for distributed reinforcement learning that tunes the tradeoff between distributed data collection and learning sample efficiency to optimize for training speed by combining the sample efficiency of PPO and the data throughput from IMPALA.

We introduce a framework for using contextual information about a task to guide the initialization of task-specific models before adaptation to online feedback, which leads to faster adaptation to online feedback than that of zero-shot multitask approaches.

An algorithm for safe reinforcement learning which utilizes a set of offline data to learn about constraints before policy learning and a pair of policies which separate the often conflicting objectives of task directed exploration and constraint satisfaction to learn contact rich and visuomotor control tasks.

We propose RLFlow, a hybrid actor-dataflow programming model for distributed RL, that leads to highly composable and performant implementations of RL algorithms, which results to faster training and significant code reductions.

An intelligent application of RL that tunes the parameters of existing Quadratic Program (QP) solvers and improving solving times by up to 3x.

We propose the first domain-independent unsupervised / self-supervised learner that discovers high-quality autoregressive orders through fully-parallelizable end-to-end training without domain-specific tuning.

An algorithm for interactive imitation learning that learns to minimize human context switching through sustained interventions and maintains the same supervisor burden for prior algorithms.

We investigate different seed placement and pruning algoritms in a polyculture garden simulator to jointly maximize diveristy and coverage of various plants types.

CS 189: Introduction to Machine Learning
Teaching Assistant: Fall 2019

CS 162: Operating Systems and System Programming
Reader: Fall 2018

Website template from Jon Barron