JASPR Lab launches at Dartmouth
We launched JASPR (Joint AI Security Privacy Research) at Dartmouth Computer Science. The lab is recruiting students at all levels. See the Join page for current openings.
Security, privacy, and trustworthy AI.
Model provenance, generative AI, adversarial ML, and protection against AI misuse.
The JASPR Lab (Joint AI Security Privacy Research) at Dartmouth College is directed by Professor Shawn Shan. Our research spans the security, privacy, and trustworthiness of modern AI systems, with occasional ventures into human-computer interaction and measurement. The lab includes PhD and undergraduate researchers and is actively recruiting students at all levels.
News
Research Areas
Attribution, Provenance, and Auditing
We trace where models and their outputs come from: which model produced a given image, whether an image is human-made or AI-generated, and which training data drove a model's behavior.
Generative Models and Training Data
We study how training data shapes what generative models learn and produce. Glaze and Nightshade, two tools from this work, measure how edits to training images affect a model's output.
Security and Privacy in Machine Learning
We work on the security and privacy of deployed machine learning: detecting backdoors, defending against black-box query attacks, and tools like Fawkes that limit what models can learn about people.
- Neural Cleanse: Identifying and Mitigating Backdoor Attacks (IEEE S&P 2019)
- Blacklight: Defense against Query-Based Black-Box Attacks (USENIX Security 2022)
- Fawkes: Protecting Personal Privacy against Unauthorized Deep Learning (USENIX Security 2020)
- SoK: Anti-Facial Recognition Technology (IEEE S&P 2023)