Publications

A One-Layer Decoder-Only Transformer is a Two-Layer RNN, With an Application to Certified Robustness

CodeFort, Robust Training for Code Generation Models

PECAN, A Deterministic Certified Defense Against Backdoor Attacks

Verified Training for Counterfactual Explanation Robustness under Data Shift

Reliability Assurance for Deep Neural Network Architectures Against Numerical Defects

Overwatch: Learning Patterns in Code Edit Sequences

BagFlip: A Certified Defense Against Data Poisoning

Certified Robustness to Programmable Transformations in LSTMs

Detecting Numerical Bugs in Neural Network Architectures

Robustness to Programmable String Transformations via Augmented Abstract Training

An Empirical Study on TensorFlow Program Bugs