Suman Jana

Suman Jana

I am an associate professor of computer science at Columbia University.

My goal is a unified science of trustworthy computation: one that reasons about programs, neural networks, and AI agents with the same mathematics.

Programs are discrete; the mathematics behind modern learning is continuous. My work closes that gap from both sides. In one direction, we treat neural networks as programs to be tested and formally verified, from DeepXplore to the α,β-CROWN family of verifiers, winners of five consecutive neural network verification competitions. In the other, we make programs smooth enough for gradients and optimization to work on them, beginning with NEUZZ, which introduced neural program smoothing for fuzzing, and extending to invariant inference and binary analysis.

What ties these directions together is representation: finding compact descriptions of what a computational system can do, small enough to reason about and precise enough to verify.

Email suman (AT) cs.columbia.edu
Office Mudd 412, 500 W 120th St, New York, NY 10027

Blog My thoughts on AI security →

Earlier posts (1)

Classes

Earlier classes (15)

Preprints 4 on arXiv

All publications 73 papers, 2008–2026

20268 papers
20253 papers
20246 papers
20233 papers
20226 papers
20219 papers
20207 papers
20193 papers
20184 papers
20174 papers
20163 papers
20153 papers
20143 papers
20133 papers
20124 papers
2011 and older4 papers

A complete list is on Google Scholar. Feel free to contact me with questions or feedback.

Impact

Our research has led to more than 250 high-impact security vulnerabilities being reported and fixed in widely used software. α,β-CROWN, the widely used neural network verifier built with collaborators on our Beta-CROWN and GCP-CROWN work, has won the International Verification of Neural Networks Competition five years in a row (2021–2025). DeepXplore, the first whitebox testing framework for deep learning, won the SOSP 2017 Best Paper Award and was selected as a CACM Research Highlight. MoonShine was merged into Syzkaller, Google’s Linux kernel fuzzer, and NoFrak was integrated into Apache Cordova. Our work has received six best paper awards, including at SOSP, OSDI, and IEEE S&P.

Students

Awards & honors