Kanak Das

Manuscript under review

Beyond the Binary Verdict: Quantitative Neural Network Verification

Kanak Das, Bor-Yuh Evan Chang, Sasa Misailovic, Gagandeep Singh, Manu Sridharan

We study a quantitative formulation of neural network verification that goes beyond a binary robustness verdict by quantifying violating inputs and examining what richer verification questions can enable.

33rd Static Analysis Symposium (SAS 2026) · Preprint

Uncovering the Limits of Proof Sharing for Neural Networks

Kanak Das, Shubham Ugare, Bor-Yuh Evan Chang, Sasa Misailovic, Gagandeep Singh, Manu Sridharan

We systematically study the limits of proof sharing for neural network verification, use jointly stable neurons to explain when template reuse is likely to help, and introduce an automatic procedure for deciding whether, where, and how to reuse templates.

• Paper
• Artifact

39th European Conference on Object-Oriented Programming (ECOOP 2025)

Practical Type-Based Taint Checking and Inference

Nima Karimipour, Kanak Das, Manu Sridharan, Behnaz Hassanshahi

TaintTyper combines a practical type-based taint checker with annotation inference, enabling modular and efficient checking on large Java programs.

• Paper
• Extended
• Code
• Artifact

7th International Conference on Networking, Systems and Security (NSysS 2020)

An Integrated Inspection and Visualization Tool for Accurate Android Collusive Malware Detection

Ajoy Das, Kanak Das, Md. Shohrab Hossain

AndroCap combines Android inter-component communication analysis with visualization to inspect potential collusive behavior across applications.

• Paper