Detecting Neural Network Failures through Spectral Analysis of Internal Activations

A new study from arXiv identifies Spectral Drift in neural network internal activations as a reliable indicator of misclassifications, with failures showing a 1.9% increase in drift (p<0.001). The researchers introduce Self-Detecting Neural Networks (SDNN), which monitor spectral dynamics across network depth and achieve 79.0 ± 25.3% AUROC on CIFAR-10, outperforming confidence-based methods like MaxSoftmax (50.5%) and Energy Score (52.9%) by 25-30 percentage points.

arXiv:2607.20590v1 Announce Type: new Abstract: Neural network misclassifications exhibit characteristic spectral instability in internal activations that is invisible at the output layer. This phenomenon is identified and formalized as Spectral Drift -- the frequency-domain distance between consecutive layer activations -- with empirical validation showing that failures exhibit significantly higher drift than correct predictions 1.9% increase, p<0.001 . This spectral signature emerges during internal processing but becomes masked in final outputs, explaining why confidence-based detection methods struggle. This work introduces Self-Detecting Neural Networks SDNN , a framework that monitors spectral dynamics across network depth using Short-Time Fourier Transform, wavelet decomposition, and statistical moments to capture multi-scale spectral features. A lightweight detector network 5% parameter overhead learns to identify failure-indicative patterns via curriculum learning on progressively challenging distributions: natural misclassifications, distribution shifts, and adversarial perturbations. Experiments on CIFAR-10 demonstrate that SDNN achieves 79.0 +/- 25.3% AUROC across three seeds, substantially outperforming confidence-based baselines including MaxSoftmax 50.5% and Energy Score 52.9% by approximately 25-30 percentage points. Ablation studies reveal that wavelet decomposition and statistical features make consistent contributions, while STFT's role remains unclear. This work establishes spectral analysis of internal activations as a promising direction for neural network reliability, revealing diagnostic information inaccessible to output-based approaches.