Yingyan Lin
Georgia Institute of Technology, GA·Computer Science·Early-career
Powerful AI models need far more computing power, memory and energy than phones, drones, headsets and other everyday devices can supply. Sending data to a central server instead adds delay, uses energy and raises privacy concerns.
- Fields
Computer Science
Engineering
- Topics
edge computing
on-device AI
federated learning
- Grants
- 3 known grants, through Jul 2027
What they work on
- 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.
- Develop decentralized optimization and communication methods for training models across networks of heterogeneous edge devices, and validate them with co-designed energy-efficient hardware on real devices.
- Build a Differentiable Hardware Accelerator Search (D-HAS) engine, using a design-space description and performance predictor, to quickly find efficient deep learning accelerator chip designs.
- Search jointly for the best neural network and accelerator pairs to maximize hardware efficiency, and benchmark the results on Rice University's ASTRO platform.
- Co-design algorithms and hardware for instant training and real-time inference of neural 3D reconstruction, covering both single-scene fitting and cross-scene learning from few views.
- Develop distributed learning of 3D reconstruction models across edge devices for private data, and resource-adaptive inference scheduling, then evaluate them in realistic systems.
More about the research
The common thread of this research is closing that gap by designing AI algorithms and the hardware that runs them together, a strategy called co-design. Three projects apply this idea to learning from streaming data across many devices, to automatically designing chips for deep learning, and to 3D scene reconstruction on devices.
Why it matters
Self-driving cars, drones, smart homes and augmented and virtual reality headsets all need AI that runs quickly on the device itself. This work could make on-device AI faster, more energy-efficient and more private by keeping raw data local. It could also cut the time needed to design specialized AI chips. Lower energy use would support greener AI.
Known grants
From NIH and NSF research grants. Labs often have other funding we can't see.
SHF: Medium:DILSE: Codesigning Decentralized Incremental Learning System via Streaming Data Summarization on Edge
NSF · Jan 2024 – Jul 2027
SHF: Medium: Cross-Stack Algorithm-Hardware-Systems Optimization Towards Ubiquitous On-Device 3D Intelligence
NSF · Jul 2023 – Jun 2027
CAREER: Differentiable Network-Accelerator Co-Search Towards Ubiquitous On-Device Intelligence and Green AI
NSF · Dec 2023 – Mar 2027
Closest labs
Yuhao Zhu
University of Rochester, NY·Computer Science·Early-career
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.
Miao Pan
University of Houston, TX·Computer Science
Design software-hardware co-designed model pruning and adaptive quantization schemes that exploit the sparsity of model updates to reduce computation and memory access during federated learning on a single 5G mobile device.
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.
The school
Georgia Institute of Technology: 50 labs on PI Pages, $1.5B a year on research, 572 research doctorates a year.
Reach out
A start for your email. Fill in the brackets.