SEPAL @ Georgia Tech

Our group SEPAL¹ is focused on the design and analysis of algorithms for machine learning, with a focus on security and privacy. We span theory and practice, with particular interests in adversarially robust algorithms, formal verification, cryptography, and privacy-preserving algorithm design.

Our work is motivated by impactful, real-world problems in the security and privacy of machine learning systems.

Research Questions. Can we trace the provenance of data used to train a model, and can that provenance survive adversarial tampering with provable guarantees? How robust are learning systems to poisoning and covert control attacks, and what guarantees can we give against them? Can we verify and certify properties of neural networks, such as robustness and fairness, while preserving the confidentiality of the model and its training data?

We are lucky to have a great group of PhD students, undergraduate and Master’s students, and collaborators working on these topics and related AI safety and security topics.

PhD Students

• Jehad Hussein
• Caroline Huang
• Zedian Shao
• Kevin Dai
• Sankha Das, co-advised with Vlad Kolesnikov

Undergraduate / Master’s Students

• Yash Chauhan

Alumni

• Jason Zhang (→ Adobe, San Jose)
• Adrian Cheung (→ Amazon, Seattle)
• Varkey John, PhD student in ECE GaTech

Collaborators

• Evangelos Froudakis, PhD student in ECE GaTech
• Sidong Guo, PhD student in ECE GaTech

Funding

We gratefully acknowledge the support of NSF CICI and Cisco, as well as the Alan and Anne Taetle Professorship of Teodora Baluta, for making our work possible.