Liu Yang Joins HKU: From Ambient Light Privacy Leaks to Physical AI Imaging

Background and Appointment

Caltech postdoctoral researcher Liu Yang has been appointed as a tenure-track assistant professor in the Department of Electrical and Computer Engineering at the University of Hong Kong (HKU-ECE) . Liu earned his bachelor's and master's degrees from the Department of Automation at Tsinghua University , where his master's thesis received the Tsinghua University Outstanding Thesis Award . He then pursued a Ph.D. in Electrical Engineering and Computer Science at MIT , supported by the MIT-Takeda Fellowship and the MIT Stata Family Presidential Scholarship . His doctoral committee included his advisor Frédo Durand , Bill Freeman , and Kaiming He .

After completing his Ph.D., Liu joined Katie Bouman 's Computational Imaging Group at Caltech as a postdoctoral scholar. Bouman is a core contributor to the algorithm that produced the first image of a black hole.

Key Research Contributions

Privacy Dual Imaging: Ambient Light Sensors as Covert Cameras

In his Privacy Dual Imaging work, Liu discovered that the ambient light sensor (ALS) embedded in smartphone and laptop screens — a component that only adjusts screen brightness — can be repurposed to infer hand gestures and scenes in front of the screen without accessing the front-facing camera or triggering any permission prompts . This finding reveals a previously overlooked privacy risk.

The paper was accepted by Science Advances in 2024 and covered by over 30 media outlets including Forbes , WIRED , Fox News , and Ars Technica . Following this work, Liu participated in drafting the W3C Ambient Light Sensor Standard to mitigate the vulnerability.

Vaccine Record Encoding via Microneedle QR Codes

In collaboration with MIT's vaccine delivery team, Liu developed a method to encode vaccination records into error-corrected 2D barcodes that are delivered into the skin via microneedle patches . A deep learning pipeline robustly decodes the patterns from skin images. This work was published in Nature Materials in 2025 .

PnP-SCI Toolchain: A Unified Framework for Snapshot Compressive Imaging

Beyond these two high-profile applications, the backbone of Liu's research is the PnP-SCI (Plug-and-Play Snapshot Compressive Imaging) toolchain, a line of work that evolves progressively:

2019 (TPAMI): Nuclear-norm minimization for SCI, enabling reconstruction of high-dimensional data cubes from 2D measurements.

2020 (CVPR Oral): Large-scale SCI using Plug-and-Play (PnP) framework with deep denoisers as priors.

2021 (Photonics Research): Extensions to spectral imaging and megapixel-scale SCI.

2022 (TPAMI): Extension to video SCI .

Diffusion Models as Universal Priors

The final chapter of Liu's Ph.D. thesis explores "Large Model Universal Priors" : when traditional PnP priors fail, pre-trained diffusion models can serve as off-the-shelf plug-in priors for SCI and non-line-of-sight (NLOS) imaging. During an internship at Meta Reality Labs , he applied the same principle to elevate the realism of CG avatars in VR/AR to human-level fidelity.

Responsible Imaging Workshop and Future Directions at HKU

Liu co-founded the first Responsible Imaging Workshop at ICCV 2025 (Honolulu), bringing together global teams to discuss ethical imaging and sensing under a unified framework.

At HKU, he organizes his upcoming research into three pillars:

Physical AI Imaging Algorithms

Physical AI Perception Applications

Multi-Agent Physical AI

His long-term vision is to achieve intelligence beyond visual senses , using physical AI as an anchor to develop multi-agent AI systems that reach or exceed human expert performance.