Yuhao Zhu
University of Rochester, NY·Computer Science·Early-career
Cameras, headsets and robots increasingly feed images to computers rather than to people. The chips that process those images are running into limits on speed and energy.
- Fields
Computer Science
Engineering
- Topics
computer architecture
hardware accelerators
accelerator-level parallelism
- Grants
- 3 known grants, through Sep 2027
What they work on
- Develop the foundations for in-sensor machine vision that combines computational metasurfaces (flat nanostructures that process light) with analog encoders inside CMOS image sensors, so images are partly processed before they are digitized.
- Build an end-to-end design framework that models trade-offs across optics, image sensors and vision processors, linking simulation tools across physical domains and using machine-learning feature selection to decide where each computation should happen.
- Run psychophysical experiments to identify which aspects of visual scenes shape both the viewing experience and the computing cost in AR/VR, and turn the results into lightweight computational models of human perception.
- Use these perceptual models to cut work across the AR/VR pipeline, from algorithms and systems to the display, and tune the whole system with a data-driven learning approach that meets perception and efficiency targets.
- Reduce the overhead of data communication and task coordination between hardware accelerators on a chip by designing new hardware structures and run-time mechanisms that detect and adapt to data and task flows.
- Create programming interfaces and compiler support that expose data flow, task flow and metadata-sharing opportunities across accelerators, so algorithms can share information instead of each accelerator working alone.
More about the research
A common thread across these projects is designing hardware and software together, so that each piece of computation happens in the most efficient place. That place might be the optics and image sensor, a group of specialized chips working together, or a decision guided by what the human eye can actually notice. Target uses include augmented and virtual reality (AR/VR), smart sensing, self-driving cars and mobile robotics.
Why it matters
Self-driving cars, AR/VR headsets, mobile robots and always-on environmental sensors all need to process large amounts of visual data quickly on small energy budgets. Moving computation into optics and sensors, coordinating specialized chips, and skipping work the eye cannot notice could make these devices faster and far more energy efficient. Possible beneficiaries include healthcare, education, remote communication and training.
Known grants
From NIH and NSF research grants. Labs often have other funding we can't see.
Collaborative Research: FuSe: Metaoptics-Enhanced Vertical Integration for Versatile In-Sensor Machine Vision
NSF · Oct 2023 – Sep 2027
Collaborative Research: CNS Core: HCC: Small: Enabling Efficient Computer Systems for Augmented and Virtual Reality: A Perception-Guided Approach
NSF · Nov 2022 – Apr 2027
CAREER: Systems and Architectural Support for Accelerator-Level Parallelism
NSF · May 2021 – Apr 2027
Closest labs
Yingyan Lin
Georgia Institute of Technology, GA·Computer Science·Early-career
Create algorithms that summarize streaming data on each edge device to cut memory use and latency while preserving privacy, as the basis for decentralized incremental learning.
Shaahin Angizi
New Jersey Institute of Technology, NJ·Engineering·Early-career
Design reconfigurable circuits and architectures for a multi-modal in-sensor accelerator that moves AI computation onto sensor nodes as efficiently as possible.
Tian Guo
Worcester Polytechnic Institute, MA·Computer Science·Early-career
Design QCloud, a quantum cloud framework with job placement, job scheduling and network scheduling algorithms built on models of quantum circuits, gates and remote gates executed across networked machines.
The school
University of Rochester: 38 labs on PI Pages, $492M a year on research, 308 research doctorates a year.
Reach out
A start for your email. Fill in the brackets.