Constructed negation dataset generating pipeline and negation testing benchmarks on CLIP..
Developed AIVariant, a deep learning model which can detect single nucleotide variant (potential tumor) even in low sequence depth.
Developed ProPILE, which can be used both for LLM users and providers to check Personally Identifiable Information(PII) leakage.
Devised ILVR, which can be used to guide a diffusion model to generate images based on the reference.
Developed a new unlearning technique (OPT-OUT), a technique that removes a certain entity’s data without seriously harming model’s performance.
