🤖 AI Summary
This study addresses the absence of explicit energy models in tabular anomaly detection by revisiting Deep Boltzmann Machines (DBMs). We propose that mean-field energy and reconstruction scores exhibit complementarity, integrating them via a rank fusion strategy to provide a non-redundant perspective for anomaly detection. Experimental results demonstrate significant improvements in AU-ROC across multiple benchmarks, with statistical superiority over established baselines. These findings validate the effectiveness of classical energy-based models on tabular data and underscore their complementary value alongside reconstruction-based approaches. Consequently, this work establishes a novel research direction for tabular anomaly detection by bridging traditional energy modeling with modern evaluation frameworks, offering both theoretical insights and practical performance gains in identifying anomalous instances within structured datasets.
📝 Abstract
Tabular anomaly detection is dominated by classical density-proxy methods (Isolation Forest, OCSVM, LOF), reconstruction-based detectors (Autoencoders, VAEs), and modern non-parametric scorers (COPOD, ECOD, Deep SVDD), all of which approximate the inlier distribution only indirectly; explicit energy-based models are largely absent. Motivated by the recent revival of EBMs in deep learning (e.g., Energy-Based Transformers, JEPA), we revisit the classical Deep Boltzmann Machine (DBM) for this task and hypothesize that its mean-field energy combines more effectively with a reconstruction-based score than same-lineage pairs do. We evaluate a two-hidden-layer DBM on two tabular benchmarks spanning distinct domains (UCI Bank Marketing and NSL-KDD) against eight classical and modern baselines across twenty random seeds. The DBM mean-field energy matches the strongest baseline (the Autoencoder) on Bank Marketing and statistically beats it on NSL-KDD, while significantly outperforming the remaining seven on both datasets. When fused with the Autoencoder via rank fusion, the DBM energy yields a statistically significant improvement on both datasets (AUROC=+0.014, p<0.01 on Bank Marketing; +0.002, p<0.001 on NSL-KDD); every non-DBM-derived base model instead fails to improve or significantly degrades the AE-paired ensemble. Our position is that classical EBMs, exemplified by the DBM, deserve a place in the tabular anomaly detection toolbox as a non-redundant complementary view to the reconstruction-based scores that dominate current practice.