Academic personal website from Tin Sum Cheng

I am a PhD student from Aurelien Lucchi’s group in University of Basel, Switzerland. My main interests are kernel methods and linear neural networks in machine learning theory. Feel free to check out my publications, blogs and CV. Our group welcomes all friendly networking and potential site visit. Please reach out to us and visit the beautiful Switzerland.

For my published research, please see Publication. Currently, I am researching on the following exciting topics:

• Optimization on Reasoning Task: Enhancing reasoning tasks through reinforcement learning, similar to those employed in DeepSeek R1, and comparing it to direct preference optimization
• Parameter-Efficient Fine-Tuning: Utilizing advanced methods such as LoRA ensembling and weight averaging to fine-tune LLMs/VLMs for downstream tasks.​ (Looking for Master Students, have a look on Proposal and please contact me via email if interested)
• Quantization Effects: Investigating the impact of quantization on model performance to optimize efficiency.​
• Implicit Biases: Investigating the implicit biases of optimizers and architecture under the lens of neural collapse (NC) and class imbalance. (Looking for Master Students, have a look on Proposal and please contact me via email if interested)
• Overlap between LoRA and NC: How LoRA variants affect NC metrics and hence OOD performance. (Looking for Master Students, have a look on Proposal and please contact me via email if interested)
• Kernel Method in Chemistry: Implementing practical kernel algorithms for molecular chemistry.

For more info

More info about our group can be found in our group website.