Improving Feasibility via Fast Autoencoder-Based Projections

TL;DR

This paper introduces a fast, data-driven autoencoder-based projection method to efficiently enforce complex constraints in learning and control systems, reducing computational costs.

Contribution

It proposes a novel autoencoder approach trained with an adversarial objective to quickly project infeasible predictions onto feasible sets.

Findings

Effective enforcement of nonconvex constraints in various problems

Low computational cost compared to traditional solvers

Applicable to reinforcement learning and optimization tasks

Abstract

Enforcing complex (e.g., nonconvex) operational constraints is a critical challenge in real-world learning and control systems. However, existing methods struggle to efficiently enforce general classes of constraints. To address this, we propose a novel data-driven amortized approach that uses a trained autoencoder as an approximate projector to provide fast corrections to infeasible predictions. Specifically, we train an autoencoder using an adversarial objective to learn a structured, convex latent representation of the feasible set. This enables rapid correction of neural network outputs by projecting their associated latent representations onto a simple convex shape before decoding into the original feasible set. We test our approach on a diverse suite of constrained optimization and reinforcement learning problems with challenging nonconvex constraints. Results show that our method effectively enforces constraints at a low computational cost, offering a practical alternative to expensive feasibility correction techniques based on traditional solvers.