Tianqi Chen
Carnegie Mellon University, PA·Computer Science·Early-career
AI models such as large language models are now used everywhere, but running them efficiently is hard. Each kind of hardware is different, whether a data-center GPU, a smartphone, a laptop or a web browser.
- Fields
Computer Science
Engineering
- Topics
machine learning compilers
ML compiler
AI deployment
- Grants
- 2 known grants, through Apr 2031
What they work on
- Build a unified program representation that captures both high-level computation and data formats and low-level hardware optimizations of AI models, as the foundation for a new machine learning compiler.
- Develop an AI-driven compiler agent that combines domain expertise with structured tools to iteratively optimize model programs for diverse hardware.
- Create end-to-end automated solutions for deploying AI models from data centers to smartphones and web browsers, reducing the effort needed to support new hardware.
- Grow an open-source ecosystem for generative AI deployment by unifying cloud and edge development flows with ML compilers and portable runtimes, and by setting up community governance.
- Build shared infrastructure to continuously monitor the quality and efficiency of deployed generative AI models, with documentation to help users trade off cost, privacy and accuracy.
More about the research
Each typically needs hand-tuned software to run a model quickly and cheaply. Machine learning compilers help with this. A compiler is software that translates a model into fast code for specific hardware. Both projects build compiler and runtime tools that let AI models, especially generative AI, run well across many platforms with less manual work.
Why it matters
Running AI models is expensive and often limited to large data centers. These tools could make generative AI cheaper to deploy and let it run on phones, laptops and browsers. On-device AI can keep personal data private. Researchers and smaller organizations could use customized models more easily. Hardware makers could also get AI software running on new chips with much less manual work.
Known grants
From NIH and NSF research grants. Labs often have other funding we can't see.
Closest labs
Minxuan Zhou
Illinois Institute of Technology, IL·Computer Science
Design a domain-specific programming language that makes it easier to write general fully homomorphic encryption (FHE) applications across different algorithmic schemes.
David Andrew Hyde
Vanderbilt University, TN·Computer Science·Early-career
Develop optimized GPU implementations of grids, particles, octrees and linear solvers as the foundation of COSTA, an open cloud platform for physics simulation.
Willow Ahrens
Georgia Institute of Technology, GA·Computer Science
Form four coordinated working groups to develop core ecosystem components for sparse array programming — shared designs, data formats, performance metrics and optimization strategies — aligned through a joint committee.
The school
Carnegie Mellon University: 28 labs on PI Pages, $505M a year on research, 346 research doctorates a year.
Reach out
A start for your email. Fill in the brackets.