Lookahead Branching for Neural Network Verification

ICLR 2026 Conference Submission13744 Authors

Keywords: Neural Network Verification, AI Safety Robustness, Formal Methods

TL;DR: This paper explores incorporating lookahead, a well-known branching strategy from SAT/SMT solving, into neural network verification.

Abstract: In this paper, we investigate the effect of lookahead branching strategy in neural network verification. We present a general recipe for integrating lookahead into any branch-and-bound search framework, and also describe how in addition to guiding branching, lookahead can generate additional lemmas that accelerate verification. We instantiate the method in two representative branch-and-bound-based verifiers (Marabou and $\alpha$-$\beta$-CROWN), and demonstrate consistent reductions in overall verification time across both systems.

Primary Area: alignment, fairness, safety, privacy, and societal considerations

Submission Number: 13744