CORTEXA
← Browse
arxivcs.LGcs.CV2026-07-22

Detecting Neural Network Failures through Spectral Analysis of Internal Activations

Arunan J

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.

View free PDFSource page

Related papers

arxivcs.LGcs.CV2026-07-03

Observable- and Positional-Encoding-Dependent Symmetry Readout from Neural Network Weights

Naoya Chiba, Satoshi Sugiyama, Yuki Uranishi

Post-hoc analysis of trained neural network weights often seeks to recover geometric structure directly from the parameters. We show that, for positional-encoding-equipped neural fields, the symmetry visible from weights is not the true symmetry group itself, but an observable sy…

View free PDFSource page
arxivcs.CVcs.AIcs.LG2026-06-26

FedLAS: Feature-Modulated Bidirectional Label Smoothing for Neural Network Calibration

Thiru Thillai Nadarasar Bahavan, Sachith Seneviratne, Saman Halgamuge

Deep Neural Network (DNN) classifiers suffer from poor calibration when their softmax outputs (predictive confidence) deviate from the empirical likelihoods. This manifests itself as either overconfident incorrect predictions or under-confident correct predictions. Label smoothin…

View free PDFSource page
arxivcs.LGcs.CVeess.IV2026-07-08

Prior-matched evaluation of operational Earth-observation classifiers: a three-number reporting method demonstrated on Sentinel-1 internal-wave detection

Joao Pinelo, Joao Goncalves, Arun Shukla, Adriana Santos-Ferreira

The Internal Waves Service screens the Sentinel-1 Wave-mode archive for internal solitary waves, routing detections to experts whose adjudication time is the resource the effort exists to conserve. Because attention is the cost of error, precision leads. Its classifier was traine…

View free PDFSource page