Sanjay Das

About Me

I am a Postdoctoral Research Associate at Oak Ridge National Laboratory working at the intersection of AI for Science & Operations, AI system reliability, and hardware-aware machine learning. My research focuses on building robust, trustworthy, and safety-aware AI systems that can be reliably deployed in mission-critical environments.

A central theme of my research is the use of AI for operational safety, particularly in complex and high-risk domains such as chemical and industrial systems. I develop AI-driven frameworks for risk-aware decision support, anomaly detection, and causal analysis, enabling models to reason about system behavior in a structured and physically grounded manner rather than relying solely on statistical correlations.

More broadly, I am interested in enabling the “research lab of the future”—where AI systems assist in autonomous scientific discovery, hypothesis generation, and experimental design. My work focuses on ensuring that such systems are grounded, interpretable, and aligned with domain constraints, allowing them to integrate reliably into complex scientific workflows.

At the same time, I study the limitations and risks of AI integration, including:

• Brittleness under distribution shifts and adversarial perturbations
• Lack of causal understanding in data-driven models
• Error propagation in multi-stage automated workflows
• Safety and verification challenges in high-stakes environments

My goal is to develop principled frameworks that enable safe, reliable, and interpretable AI integration—ensuring that advanced AI systems are not only powerful, but dependable and aligned with real-world operational and scientific requirements.

Education

Current Research

1. AI-Driven Scientific Discovery & Hypothesis Generation

Currently, as a Postdoctoral Researcher, I am developing novel workflows that leverage state-of-the-art AI tools and foundation models to assist in scientific hypothesis generation. My work focuses on building structured, reproducible AI-driven pipelines that augment domain experts in generating testable scientific insights across interdisciplinary research domains.

2. AI for Workplace Safety in Mission-Critical Environments

Designing intelligent systems that enhance operational safety in mission-critical work sites. This includes risk-aware AI frameworks, structured safety knowledge integration, and automated decision-support systems aimed at reducing hazards and improving situational awareness.

3. Hardware-Induced Vulnerability in Large Language Models

Investigating how targeted bit-flips in quantized LLM weights can induce adversarial behavior. Developing constrained optimization methods to identify minimal critical weight subsets for feasible hardware-level attacks.

4. Robustness Evaluation & Defense Frameworks

Building production-ready evaluation pipelines to test AI systems under structured perturbations, adversarial noise, and hardware faults. Integrating statistical validation and adversarial training mechanisms.

Selected Publications

• AttentionBreaker: Adaptive Evolutionary Optimization for Unmasking Vulnerabilities in LLMs through Bit-Flip Attacks
  Sanjay Das, Swastik Bhattacharya, Souvik Kundu, Shamik Kundu, Anand Menon, Arnab Raha, and Kanad Basu.
  Transactions on Machine Learning Research (TMLR), 2025. [Paper]
• CONCEAL: Covert NoC Exploitation in In-Memory Computing-Based DNN Accelerators
  Sanjay Das, Shamik Kundu, Sumit Kumar Mandal, and Kanad Basu.
  In IEEE International Conference on Omni-layer Intelligent Systems (COINS), 2025. [Paper]
• Machine Learning-Driven STL Generation for Enhancing Functional Safety of E/E Systems
  Sanjay Das, Swastik Bhattacharya, Anand Menon, Shamik Kundu, Pooja Madhusoodhanan, Prasanth Viswanathan Pillai, Rubin Parekhji, Arnab Raha, Suvadeep Banerjee, Suriya Natarajan, and Kanad Basu.
  In Proceedings of the 62nd ACM/IEEE Design Automation Conference (DAC), 2025. [Paper]
• Enhancing AMS Circuit Reliability: An Anomaly Dataset for Functional Safety Research in Automotive SoCs
  Sanjay Das, Anand Menon, Omar Abiola Abioye, Afreen Fatimah Khazi-Syed, Jonathan Edward Lee, Ayush Arunachalam, Shamik Kundu, et al.
  In Great Lakes Symposium on VLSI (GLSVLSI), 2025. [Paper]
• Bit-by-Bit: Investigating the Vulnerabilities of Binary Neural Networks to Adversarial Bit Flipping
  Shamik Kundu, Sanjay Das, Sayar Karmakar, Arnab Raha, Y. Makris, and K. Basu.
  Transactions on Machine Learning Research (TMLR), 2024. [Paper]
• Graph Learning-Based Fault Criticality Analysis for Enhancing Functional Safety of E/E Systems
  Sanjay Das, Shamik Kundu, Pooja Madhusoodhanan, Prasanth Viswanathan Pillai, Rubin Parekhji, Arnab Raha, Suvadeep Banerjee, Suriya Natarajan, and Kanad Basu.
  In Proceedings of the 61st ACM/IEEE Design Automation Conference (DAC), 2024. [Paper]
• Explainability to the Rescue: A Pattern-Based Approach for Detecting Adversarial Attacks
  Sanjay Das, Shamik Kundu, and Kanad Basu.
  In IEEE International Symposium on Hardware Oriented Security and Trust (HOST), 2024. [Paper]
• Analyzing and Mitigating Circuit Aging Effects in Deep Learning Accelerators
  Sanjay Das, Shamik Kundu, Anand Menon, Yihui Ren, Shubha Kharel, and Kanad Basu.
  In IEEE VLSI Test Symposium (VTS), 2024. [Paper]

News & Achievements

• April, 2026
  Reviewed a paper in Design & Test conference
  
• September, 2025
  Paper Accepted at TMLR
  "AttentionBreaker: Adaptive Evolutionary Optimization for Unmasking Vulnerabilities in LLMs through Bit-Flip Attacks" accepted to Transactions on Machine Learning Research (TMLR).
• March, 2025
  DAC 2025 Publication
  Presented work on ML-driven STL generation at the 62nd ACM/IEEE Design Automation Conference.
• April, 2025
  COINS 2025 Presentation
  Presented "CONCEAL" on covert NoC exploitation in AI accelerators.
• 2024
  HOST 2024 Paper Presentation
  Presented adversarial detection research at IEEE HOST 2024.
• 2024
  Research Award / Recognition
  Recognized for contributions to hardware security and AI reliability research.

Teaching & Mentorship

Community & Professional Activities

Contact Information

Email: dass3@ornl.gov; sanjay.das@utdallas.edu

Google Scholar: Google Scholar Profile

LinkedIn: LinkedIn Profile

Orcid Id: Orcid Profile