I am interested in machine learning, especially in the intersection with security/robust concerns such as anomaly detection. To understand more, I adopt 2 approaches: the applied side of problem formulation (XAI in anomaly detection, adversarial robustness) and the learning theory perspective (generalization bounds, convergence guarantees).

A list of my projects is below:

1. Anomaly Detection Theory (Convergence guarantees, expressivity with lower generalization error)

2. Adversarial Attack/Defence

3. Explainable Anomaly Detection in Power Systems (Graphical Models, Statistics, Causal Analysis, Anomaly Localization)

4. Model Predictive Control (Deep Learning, Self-Attention, Implicit Optimization)

I have also given academic talks on:

• Self-Attention

• Variational Autoencoders with Reinforcement Learning

• Anomaly Transformers

• In-context Learning of Transformers as Statisticians: Learning 2-Layer Neural Networks, Algorithm Selection

• Implicit Bias of Gradient Descent for Logistic Regression and 2-Layer Neural Networks