Trustworthy AI
Understanding and controlling how modern AI systems learn and behave.
- Mechanistic interpretability and backdoor analysis
- Machine unlearning and privacy risk mitigation
- Content safety for generative models, including NSFW content and copyright
- Hallucination detection and mitigation
Generative AI & NLP
Developing and evaluating language and generative models for reliable use.
- Large language models (LLMs), vision-language models (VLMs), and vision-language-action models (VLAs)
- Diffusion models
Applied AI
Translating AI methods into effective systems for real operating environments.
- Clinical and medical imaging AI
- Wafer image enhancement and defect detection
- Reliable multi-agent and routed AI systems
Computer Vision
Advancing visual models from image understanding to image reconstruction.
- Vision transformers and interpretable classification
- Medical image generation and segmentation
- MRI reconstruction and image enhancement
