Pragma Stays at the Forefront of State-of-the-Art Systems Research

Our research work has been deployed in several production systems in Microsoft, including Bing search engine, Open Platform for AI (OpenPAI), and Azure, demonstrating the real-world impact of our work.

Discovered consensus bugs in Ethereum. One of the bugs led to the Ethereum hard fork on November 11, 2020, which was considered Ethereum's most significant challenge since the DAO fork in 2016 (News: Coindesk, Decrypt, Korean News).

Our Research Partner, Dr. John Youngseok Yang, is in the top 10 on Ethereum’s Execution Layer Bug Bounty leaderboard with 20k points.

Established a research group dedicated to publishing ambitious systems research projects with high impact potential. Key focus areas include Practical ML for Edge Devices (REP, Miro, CarM), System Solutions for Large-scale ML (Cascading KV Cache, HiP Attention, Metis, FusionFlow, EnvPipe, Sibylla, HUVM, Zico, Philly, Tiresias), and Fast and Scalable Big Data Analytics (Blaze, Sponge, SWAN, Jarvis, AOMG, StreamBox-HBM).

Aptos, Polygon PoS, RISE and other parallel EVMs use Software Transactional Memory (STM) for parallel transactions processings. Members of our team have built systems at the forefront of STM research, including systems for pauseless Java Garbage Collection, as well as parallel Actor Model message processing.

We have proposed methods for concurrent processing of order book messages, which are directly applicable to modern day blockchain matching engines.

Proposed methods for the processing of workloads on heterogeneous architectures applicable to distributed inference systems, including Exo (our portfolio project), which focuses on distributing inference workloads among everyday devices. Notably, we have also proposed Fast Automatic Distributed Training on Heterogeneous GPUs.