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.

The goal of this work is to improve machine learning methods for drug discovery. A widely used strategy for generating new molecules involves performing Bayesian Optimization in the continuous latent space of a Variational Autoencoder (VAE). However, this approach often suffers from a well-known failure mode: it tends to over-explore the latent space, resulting in unrealistic or invalid molecules. To address this issue, we introduce a novel constraint called the Latent Exploration Score (LES), which encourages the optimization process to remain close to the training data—i.e., known, valid molecules. Through extensive experiments, we demonstrate that incorporating LES leads to higher-quality molecular candidates and more efficient optimization.