Yang Liu

PhD student at Caltech

About Me

I have a diversity of interests and experience in vision neuroscience, computer vision, AR/VR technologies, and robotics.

I worked on soft robotics, vehicle crash safety, electric vehicle warning systems, and auditory neuroscience as an undergraduate student at Tsinghua University, China.

As a PhD student in Computation and Neural Systems at Caltech, I lead projects including: an AR cognitive assistant for the blind, real-time animal tracking, calcium imaging in mice, and synthetic data generation for improving rare class generalization with deep convolutional neural networks.


California Institute of Technology

Graduate Research Assistant

  • Unity 3D/C# based pipeline for generating annotated photo-realistic images for improving rare animal classification with deep neural networks.
  • An augmented reality powered cognitive assistant for the blind with Microsoft HoloLens.
  • A computer vision system for realtime rodent motion tracking with Python/OpenCV.

XRDC 2018

Guest Speaker

Invited Talk: Powering a cognitive assistant for the blind using AR.

California Institute of Technology

Graduate Teaching Assistant

  • CNS 187, Neural Computation
  • Instructor: Pietro Perona

Tsinghua University

Undergraduate Research Assistant

  • Developed an ultra-compliant liquid metal electrodes for artificial muscles and soft electronics


California Institute of Technology, Pasadena, CA, USA

Sept 2014 - June 2020

Doctor of Philosophy in Computation and Neural Systems

Advisor: Dr. Markus Meister
Selected Coursework:

Tsinghua University, Beijing, China

Sept 2009 - Sept 2013

Bachelor of Engineering in Vehicle Engineering

Department of Automotive Engineering
Thesis Advisor: Dr. Qing Zhou


Synthetic examples improve generalization for rare classes

Deep convolutional neural networks, like most other modern computer vision systems, struggle to categorize objects they have seen only rarely during training, and collecting a sufficient number of training examples of rare events is often challenging and expensive, and sometimes outright impossible. We explore in depth an approach to this problem: complementing the few available training images with ad hoc simulated data.

Preprint on arXiv

Augmented reality powered cognitive assistant for the blind

We combine augmented reality technology with computer vision algorithms and spatial sound to create goggles that can help blind people navigate unfamiliar spaces. We had great success with initial testing with blind subjects and hope the technology can one day be offered by places like banks, grocery stores, museums, and more.

Peer-reviewed paper on eLife
Media: Caltech News , Digital Trends , MIT Tech Review , Tech Crunch

Real-time rodent tracking

Video tracking is widely used for studying rodent behaviors. However the majority of tracking methods does not work in real time, and therefore can not be used for closed-loop experiments. With python and OpenCV, I developed a lightweight, fast, and simple blob tracking system that performs real-time mouse position tracking, video display and recording, as well as streaming position via UDP.

Ultra-compliant liquid metal electrodes for artificial muscles and soft electronics

We developed a type of liquid metal electrodes and the method of directly printing it on dielectric elastomers for dielectric elastomer actuators (DEA). Such soft electrodes enable dielectric elastomer films to approach its maximum strain and stress at relatively low voltages. And its unique capability of achieving two-dimensional in-plane self-healing by simple actuation allows more tolerance to fault and resilience to abusive environments. This actuator has a wide spectrum of applications ranging from artificial muscle, flexible electronics to smart clothings.

Peer-reviewed paper published on Applied Physics Letters


Get in Touch