Publications

Semantics-preserving code generation improves transformation correctness by 6.9% and performance by 1.1× with the same training budget.

A realistic, developer-informed benchmark for evaluating code generation models on tasks that reflect real-world software development.

Sparse Fourier learning predicts feasible kernel race traces, finding 44 more races and five new bugs.

Static-analysis-guided prompting that drives LLMs to reason symbolically about execution paths, doubling coverage of generated regression tests.

Probabilistic lockset analysis finds kernel races 3× faster, uncovering 183 races, including 102 harmful ones.

A neural transformer paired with a test-oracle grammar that automatically generates bug-finding assertions, with a 170% improvement in bug detection over prior systems.

Data-driven inference of inductive invariants for verifying safety properties of distributed protocols.

Proximal gradients improve dataflow-tracking F1 by 20% on average, finding 22 bugs with under 5% overhead.

Gated Continuous Logic Networks learn nonlinear loop invariants, solving 26 of 27 benchmarks, three more than prior work.

Continuous Logic Networks learn explicit SMT invariants, solving all 124 Code2Inv problems 40× faster than prior methods.

Pixel Approximate Entropy predicts perceived line-chart complexity and the accuracy of at-a-glance judgments.

Realistic user bots inject malicious behavior to continuously test insider-threat detection in synthetic and live university networks.

Oversampling and filtering reduce vortex-shedding noise in current meters, improving signal-to-noise by 3×.