Research

My current research interest lies in Human & Machine Visual Perception. My past research projects cover the topics of vision, deep learning, cognitive science, and Interactive Systems.

Papers

Harmonization

The rapid acceleration of progress in deep neural networks (DNNs) have researched or even surpassed human capacities in vision. However, we’ve found that these DNNs have become misaligned with humans in terms of representations, visually-evoked behaviors, robustness and neural responses. We’ve developed a training toutines to align models with biological intelligence to rectify this growing issue.

Advisors: Prof. Thomas Serre, Prof. Drew Linsely
Timeline: Mar. 2023 - Present
Key Words: Vision, Alignment, Deep Learning, Visual Cognition
Link: Paper - ArXiv, Paper (Coming soon)

Social AR

Social AR is developed to support real-time multiuser interaction for exploration of cultural heritage. In this research, we believe Social AR that projects cultural heritage objects projected onto the real environment can be a better experience in terms of depth of viewing. The work was carried out at the NVIDIA Joint-Lab on Mixed Reality, NVIDIA Technology Centre, supported by NVIDIA AI Technology Center (Singapore).

Advisors: Prof. Eugene Ch’ng
Timeline: Jun. 2021 - Aug. 2022
Key Words: Augmented Reality, Multi-user Interaction, Digital Heritage, Communication
Link: Paper - ACM Journal on Computing and Cultural Heritage

You Draw, I Guess

This undergraduate final-year project is about exploring the interpretability of AI-driven sketch recognition through human-AI interaction based on a lightweight interactive system – DRAW (Doodle Recognition in an Automated Way). A preliminary field study has been conducted to answer the research question: “Can humans interpret AI’s inference and draw machine-recognizable sketches through the gameplay?” According to the data analysis, the answer to this research question is positive, but not a 100% “Yes”.This project paves a way for future studies to explore the human understanding of AI’s decison making.

Advisors: Prof. Steve Benford
Timeline: Oct. 2020 - May. 2021
Key Words: Human-AI Interaction, Mobile Intelligence, Free-hand Sketch Recognition
Link: Undergraduate Thesis, Presentation

DeepRelics Dataset

Cultural heritage presents both challenges and opportunities for the adoption of deep learning in 3D digitalization. In this research, we attempted to reduce the need for manual labour by automated data generation via close-range photogrammetry models with a view for future cultural heritage activities, such as digital identification, sorting and asset management.

Advisors: Prof. Eugene Ch’ng
Timeline: Apr. 2020 - Jun. 2020
Key Words: Data Augmentation, Object Detection, Digital Heritage, Photogrammetry
Link: Paper - The 13th International Conference on Agents and Artificial Intelligence


Technology Patents

Automobile BLE Localization

This project was part of the work of my internship at NIO. It’s about the optimization of BLE positioning for “Passive Entry Passive Start (PEPS)” strategy. The algorithm was deployed in phone key and key fob of NIO car in order to shape a comfortable ingress-egress passenger experience. Due to NIO’s Confidentiality Agreement, the details will not be shown.

Timeline: Nov. 2021 - Apr. 2022
Key Words: Bluetooth Low Energy, Signal Localization, Optimization, Smart Key
Link: Patent (Aceepted)


Projects

Traffic Scene Understanding

This research project was conducted at Berkeley DeepDrive, where I worked as a student research assistant in the Summer of 2020. The work aims to define a framework that enables the scene understanding of the driving environment by a perception system. My contribution was to evaluate an inflated 3D convolutional neural network model (I3D) on TITAN dataset for identifying vehicle and pedestrian behavior, and compare the result with the scene understanding solution proposed by our team.

Advisors: Prof. Ching-Yao Chan, Dr. Pin Wang, Dr. Yuke Li
Timeline: Jun. 2020 - Sept. 2020
Key Words: Action Detection, Trajectory Prediction, Spatial-temporal Learning
Link: Internship Presentation


MIT License
Last updated on Dec 24, 2023 02:08 EST
Built with Hugo
Theme Stack designed by Jimmy