Home | People | Research | Publication | About | Artifact

Our overall research focuses on developing high-performant, energy-efficient, and robust-and-reliable AI/ML systems, using a cross-layer full-stack approach spanning hardware, software, and algorithm.

Robust/Resilient AI/ML Systems

A “good” AI is not just about accuracy. It also needs to be robust/resilient against uncertainties, e.g., hardware faults/errors, adversarial input data, environmental variations, etc. How to make robust AI systems?
Algorithm-level Analysis
Hardware-level Analysis

Hardware Reliability –> Approximate Computing

As transistor becomes even smaller, hardware are more susceptible to errors/faults. How can we make reliable hardware?
Timing Error Prediction of Hardware
Approximate Computing Survey Part 1
Approximate Computing Survey Part 2

Fuzz Testing

A computing system needs comprehensive testing and validation. Fuzz testing is a popular testing method. Here are some examples: Fuzzing on Hardware
Fuzzing on AI Models
Fuzzing on Software
Fuzzing on Firmware

Brain-Inspired Hyperdimensional Computing

Hyperdimensional computing (HDC) is an emerging AI model family mimicking the “human brain”. It achieves success in various applications. Check this coverage from WIRED, and a few examples below.
In-memory Computing for Hyperdimensional Computing
Adversarial Attack on Hyperdimensional Computing
Energy-efficient Hyperdimensional Computing