Skip to content

Research

Research Focus

My research focuses on developing robust and generalizable machine learning systems for ubiquitous sensing and wearable technology.


Research Directions

🤖 LLM-Driven Multimodal Sensing Agents

Exploring the synergy between large language models (LLMs) and multimodal sensor data to build context-aware agents that can: - Reason about human activities - Adapt to novel scenarios - Provide personalized assistance

🔄 Adaptive Continual Learning for HAR

Designing algorithms that enable models to incrementally learn new user behaviors without catastrophic forgetting: - Meta-learning approaches - Dynamic memory architectures - Long-term personalization

🎯 Domain Generalization for Sensor-Based HAR

Enhancing cross-user HAR accuracy through: - Meta-feature alignment - Feature disentanglement - Seasonal style projection

Projects

Cross-Domain Generalization for Sensor-Based Human Activity Recognition

Timeline: Jun. 2025

  • Developed a PyTorch-based domain generalization framework integrating meta-learning and feature disentanglement
  • Validated on multiple public benchmarks, achieving state-of-the-art robustness against distribution shifts
PyTorch Meta-Learning HAR

Lightweight Large Language Model Implementation & Training

Timeline: Feb. 2026

  • Built a 26M-parameter LLaMA-style LLM from scratch
  • Implemented core Transformer components (Multi-Head Attention, RMSNorm, RoPE)
  • Covered pre-training, SFT, LoRA fine-tuning, and vLLM deployment
LLM Transformer DPO

Privacy-Preserving ML Inference via Garbled Circuits

Timeline: Jun. 2021

  • Led the design of a privacy-preserving neural network inference framework
  • Combined garbled circuits with PyTorch for secure, high-fidelity inference
Privacy-Preserving ML Garbled Circuits

Collaboration

I'm open to collaborations in:

  • Human Activity Recognition
  • Domain Generalization
  • LLM applications