Projects

Research software for neural network verification, model analysis, and ONNX workflows.

OOPSLA'25 VerificationActivation FunctionsConvex HullFormal Methods

WraAct: Convex Hull Approximation for General Activation Functions

Efficiently constructing tight over-approximations for activation function hulls, with reported average 400x faster construction on benchmarked Sigmoid, Tanh, MaxPool, and related cases.

GitHub Paper

POPL'24 VerificationReLUConvex HullFormal Methods

WraLU: Fast and Precise ReLU Hull Approximation

A fast and precise approach to over-approximating the convex hull of the ReLU function, reporting 10x-10^6x runtime improvements and up to 50% fewer constraints in the evaluated benchmarks.

GitHub Paper