I'm seeking graduate and undergraduate students in computer science or related fields to participate in projects for academic credit. If the first project works out well, students may be funded for subsequent projects. Research activities include software design and implementation, application projects, and open-source software study. Research outcome may result in papers published at top International conferences and students will be funded to attend the conferences. Commitment of 9 to 12 hours per week is required.
Most projects require a computer science account. Students can apply for CS accounts on line. The fees for the CS accounts will be paid for by our lab.
The goal of a research project is often to answer an open research question, so it requires strong technical background and independent problem solving skills. An incomplete list of projects are shown below. If you are interested in helping with any of them, please drop me an email. Thanks!
- Robust and Secure Machine Learning. Recent advances in machine learning especially deep learning (DL) have achieved or surpassed human-level performance for a diverse set of tasks, leading to widespread adoption and deployment of ML in security- and safety-critical systems such as self-driving cars, malware detection, and aircraft collision avoidance systems. Unfortunately, DL systems, despite their impressive capabilities, often demonstrate unexpected or incorrect behaviors in corner cases. In safety- and security-critical settings, such incorrect behaviors can lead to disastrous consequences such as fatal collisions of self-driving cars. We are working on novel techniques to make DL more robust and secure. For instance, our recent ECCV work shows that multi-task learning increases adversarial robustness. Contact us to find out more.
- Systems for Machine Learning. An important ingredient driving the success of DL is the development of DL systems that efficiently support the task of learning and inference of potentially large models on diverse hardware and machine/cluster infrastructure. We are building some exciting systems along this direction. Contact us to find out more.
- Binary program analysis and hardening. Much software today, especially systems software, is distributed in binary form. For comprehensive analysis of all executable code, and fast turn-around time for transformations, it is essential to operate directly on binaries to enable profiling, security hardening, and architectural adaptation. Towards this goal, wxe are building a suite of new tools, some of which are powered by the state-of-art deep neural nets. Contact us to find out more.