SnareNet: Flexible Repair Layers for Neural Networks with Hard Constraints

TL;DR

SnareNet introduces a flexible, differentiable repair layer with adaptive relaxation to ensure neural network outputs satisfy complex, input-dependent constraints reliably and with high precision.

Contribution

It presents a novel architecture with a repair layer and adaptive relaxation for training neural networks that strictly adhere to non-convex constraints.

Findings

SnareNet achieves higher objective quality while reliably satisfying constraints.

It enforces non-convex constraints at medium-to-high precision across instances.

The method improves constraint satisfaction in optimization and trajectory planning benchmarks.

Abstract

Neural networks are increasingly used as fast surrogate models across various domains, but unconstrained predictions can violate physical, operational, or safety requirements. We propose SnareNet, a feasibility-controlled architecture to learn mappings whose outputs must satisfy input-dependent constraints. SnareNet appends a differentiable repair layer that navigates in the constraint map's range space, steering iterates toward feasibility and producing a repaired output that satisfies constraints to a user-specified tolerance. We stabilize end-to-end training by adaptive relaxation, a new training paradigm that snares the neural network at initialization and shrinks it into the feasible set, enabling early exploration and strict feasibility later in training. On optimization learning and trajectory planning benchmarks, SnareNet consistently attains improved objective quality while satisfying constraints more reliably than prior work, and it is the first to enforce non-convex constraints at medium-to-high precision robustly across instances.