About me
I am a Ph.D. student @ Kahlert School of Computing, University of Utah. I am jointly advised by Dr. Daniel Brown (ARIA Lab), & Dr. Guanhong Tao (SaLT Lab).
My research focuses on machine learning security and robustness at the intersection of generative AI and robotics. I am currently focused on securing generative models (LLMs, VLMs, Diffusion) against malicious manipulation, whether through adversarial prompts or compromised models, and on developing robust and generalizable foundations for trustworthy generation. My work also involves applying these security frameworks and identifying vulnerabilities in autonomous and robotic systems that rely on transformer-based architectures.
Ultimately, my goal is to uncover both vulnerabilities and defenses in AI systems, and to promote more robust, secure, and thoughtful model development as the field advances.
I received my M.S. in Computer Science from the University of Mississippi, under the supervision of Dr. Charles Walter @ the SPARC Lab, where I focused on exploiting the vulnerabilities and of the state-of-the-art object detection models to develop generalizable adversarial attacks.
Publications
Evading Intellectual Property Model Protections via Limited Fine-tuning - Soumil Datta, Shih-Chieh Dai, Leo Yu, Guanhong Tao. (Under Review)
Dataset Poisoning in Behavioral Cloning Policies - Akansha Kalra, Soumil Datta, Ethan Gilmore, Duc La, Guanhong Tao, Daniel S. Brown. EAI SmartSP 2025 (Accepted as Short Paper)
Generalized Loss-Function-Based Attacks for Object Detection Models - Soumil Datta, and Charles Walter. HICSS 2025 (Nominated for Best Paper)
EllipScape: A Genetic Algorithm Based Approach to Non-Photorealistic Colored Image Reconstruction for Evolutionary Art - Soumil Datta, and Charles Walter. HICSS 2024 (Best Paper Award, Top 1%)
Awards & Honors
- HICSS 2024 Conference Best Paper Award, Jan 2024
- SAP Computer Science Award, Sep 2023
- Marcus Elvis Taylor Memorial Medal, April 2022
- Outstanding Student in Computer Science Award, Mar 2021
Services
Subreviewer
- S&P (IEEE Symposium on Security and Privacy): 2025, 2026
- SaTML (IEEE Conference on Secure and Trustworthy Machine Learning): 2026
- CCS (ACM Conference on Computer and Communications Security): 2025
- NeurIPS (Conference on Neural Information Processing Systems): 2024, 2025

University of Utah