An Explainable Gaussian Process Auto-encoder for Tabular Data

2025-08-31

AI Summary

This work addresses the limited interpretability of black-box models for tabular data by proposing a lightweight Gaussian process (GP)-based autoencoder framework for high-quality counterfactual generation. Methodologically, it embeds a GP into the autoencoder’s latent space to jointly model the underlying data manifold and predictive uncertainty; introduces a novel density-aware loss function; and incorporates an automatic regularization strength selection mechanism to ensure counterfactuals reside within the true data distribution while maintaining sufficient diversity. Experiments on multiple large-scale tabular benchmarks demonstrate substantial improvements over state-of-the-art baselines: +12.6% in distributional fidelity, +19.3% in diversity metrics, and ~40% reduction in parameter count—effectively mitigating overfitting. The core contributions lie in (i) GP-driven latent-space modeling that captures both geometry and uncertainty of tabular data manifolds, and (ii) an adaptive regularization strategy balancing faithfulness and diversity of counterfactuals.

Technology Category

• Machine Learning: Deep Generative Models & Autoencoders
• Reasoning under Uncertainty: Relational Probabilistic Models
• Natural Language Processing: Generation

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• User Modeling, Personalization and Recommendation: Attacks and countermeasures in recommendation systems
• Economics, Online Markets and Human Computation: Trust and reliance of crowd workers and data experts on GenAI

Abstract

Explainable machine learning has attracted much interest in the community where the stakes are high. Counterfactual explanations methods have become an important tool in explaining a black-box model. The recent advances have leveraged the power of generative models such as an autoencoder. In this paper, we propose a novel method using a Gaussian process to construct the auto-encoder architecture for generating counterfactual samples. The resulting model requires fewer learnable parameters and thus is less prone to overfitting. We also introduce a novel density estimator that allows for searching for in-distribution samples. Furthermore, we introduce an algorithm for selecting the optimal regularization rate on density estimator while searching for counterfactuals. We experiment with our method in several large-scale tabular datasets and compare with other auto-encoder-based methods. The results show that our method is capable of generating diversified and in-distribution counterfactual samples.

Problem

• Generating counterfactual explanations for black-box models
• Reducing overfitting in auto-encoder architectures
• Ensuring counterfactual samples remain in-distribution

Innovation

• Gaussian process auto-encoder architecture
• Novel density estimator for in-distribution samples
• Optimal regularization rate selection algorithm