Main content

Research Highlights

Summary & Vision

Monographs & collaborative vision/positioning papers that provide a higher-level summary of our research results and vision.

external pageDistributed Learning Systems with First Order Methods

In this monograph (external pageFoundations and Trends® in Databases series), we summarize the theoretical foundation of the zip.ml project: system relaxation techniques for distributed learning, such as lossy communication compression, decentralization, and asynchronization. In zip.ml, we combine these theoretical results together with carefully designed ecosytem (DB, Spark, Serverless, etc.) specific optimizations and the power of modern hardware (e.g., FPGAs).

           

external pageOpportunities for Data Management Research in the Era of Horizontal AI/ML

Our research is deeply rooted in the database and data management community. This abstract summarizes the panel at VLDB 2019. Together with fellow panelists, we discuss, in our opinion, the amazing future of data management researches in the era of AI/ML. At DS3Lab, the ultimate goal of our research is to build a principled, high level, easy-to-use abstraction for ML, just like what SQL provides for data processing.

           

external pageMLSys: The New Frontier of Machine Learning Systems

The emerging research around machine learning system requires us to bring together multiple research communities such as machine learning, systems, data management, architecture etc. Over the years, we have been partcipating in the community effort of MLSys and this paper summarizes the vision. 

Projects

The goal of our research is to improve the usability of modern machine learning platforms. We tackle different aspects of this fundemental and emerging problem: from speed & scalability, automation, to process management and human interaction. We tackle each aspect with one umbralla system, which is often a collaborative efforts involving multiple DS3Lab members and (international) collaborators. 

           

Ease.ml: End-to-end Process Management for Machine Learning

Project Page

Zip.ml: Eco-system Friendly Scalable Learning w/ Modern Hardware

Project Page

Passive.ml: Observational & Passive Supervision

Project Page

sh¡ft!: Transfer Learning-Centric ML Platform

Stay Tuned!

Applications: ML for Better World & Science

Project Page

JavaScript has been disabled in your browser