Mitigating over-exploration in latent space optimization using LES

TL;DR

This paper introduces LES, a method to reduce over-exploration in latent space optimization by leveraging the decoder's data distribution, improving solution quality across various tasks without additional training.

Contribution

LES is a novel, decoder-based score that mitigates over-exploration in LSO, applicable to pretrained VAEs without retraining or architecture changes.

Findings

LES improves solution quality in all tested benchmarks.

LES maintains high objective values while reducing unrealistic solutions.

The method is computationally efficient and widely applicable.

Abstract

We develop Latent Exploration Score (LES) to mitigate over-exploration in Latent Space Optimization (LSO), a popular method for solving black-box discrete optimization problems. LSO utilizes continuous optimization within the latent space of a Variational Autoencoder (VAE) and is known to be susceptible to over-exploration, which manifests in unrealistic solutions that reduce its practicality. LES leverages the trained decoder's approximation of the data distribution, and can be employed with any VAE decoder - including pretrained ones - without additional training, architectural changes or access to the training data. Our evaluation across five LSO benchmark tasks and twenty-two VAE models demonstrates that LES always enhances the quality of the solutions while maintaining high objective values, leading to improvements over existing solutions in most cases. We believe that new avenues to LSO will be opened by LES' ability to identify out of distribution areas, differentiability, and computational tractability. Open source code for LES is available at https://github.com/OmerRonen/les.