Parts II, III & Mphil Projects


My supervision style

This is a page to record my suggestions for Part II, III and Mphil Projects.

My strategy to design such a project is to have an achievable basis so that noboday fails, but also I make sure there is a research-orientated extension.

I have supervised several students in the past, and found this strategy extremely helpful to our students, with some of the students ended up writing a research paper (some published in Tier-1 venues like ICML) at the end of the project.

I am also happy with self-proposed projects, please write me an email to discuss about your ideas.

– For Cambridge Students

I am no longer taking any Part IIs, and only have very limited avalability for PartIII/Mphil projects.


Bit-level Manipulated GNNs (Graph/Hardware)

Data-model co-optimization using blackbox optimizers (Optimization)

Running GNNs on PYNQ (Graph/Hardware)

Dilation-based learning for better robustness (Security)

Backdooring advanced Reinforcement Learning (Security/RL)

Actively learned NAS using Graph Neural Networks (Efficiency)

Automated Machine Learning for Drug Design (Graph) [Taken]

Knowledge distillation for quantized transformers (Efficiency/NLP) [Taken]

Neural Networks for Complex Graphs (Graph) [Taken]

Data Augmentaiton in NLP (NLP) [Taken]

Attacking NLP prompting (NLP/Security) [Taken]

Quantized training for transformer models (Effciency/NLP) [Taken]

Dynamic Sequence Pruning in Transformers (Efficiency/MLP) [Taken]