ScaLES: Scalable Latent Exploration Score for Pre-Trained Generative Networks

2024-06-14

arXiv.org

AI Summary

Latent Space Optimization (LSO) methods often yield impractical solutions due to excessive exploration of the latent space, compromising real-world applicability. To address this, we propose ScaLES—a scalable Latent-space Exploration Scoring framework—that theoretically approximates the data distribution using only a pretrained VAE decoder, enabling out-of-distribution (OOD) region identification without retraining, external training data, or architectural modifications. Its core innovation lies in jointly estimating the gradient of the latent-space probability density and approximating the decoder’s Jacobian determinant, thereby enabling differentiable, principled quantification of latent-space feasibility. Evaluated across five LSO benchmark tasks and three distinct VAE architectures, ScaLES consistently improves the trade-off between solution realism and objective optimization performance. The implementation is publicly available.

Technology Category

• Search and Optimization: Learning to Search
• Constraint Satisfaction and Optimization: Distributed CSP/Optimization
• Computer Vision: Learning & Optimization for CV

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Economics, Online Markets and Human Computation: Cost models of using LLMs in production systems
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs

Abstract

We develop Scalable Latent Exploration Score (ScaLES) 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. ScaLES is an exact and theoretically motivated method leveraging the trained decoder's approximation of the data distribution. ScaLES can be calculated with any existing decoder, e.g. from a VAE, without additional training, architectural changes, or access to the training data. Our evaluation across five LSO benchmark tasks and three VAE architectures demonstrates that ScaLES enhances the quality of the solutions while maintaining high objective values, leading to improvements over existing solutions. We believe that new avenues to LSO will be opened by ScaLES ability to identify out of distribution areas, differentiability, and computational tractability. Open source code for ScaLES is available at https://github.com/OmerRonen/scales.

Problem

• Mitigates over-exploration in latent space optimization
• Enhances solution quality in black-box discrete optimization
• Leverages Variational Autoencoder without additional training

Innovation

• Latent Exploration Score mitigates over-exploration
• LES works with any VAE decoder
• LES enhances solution quality effectively