TIE: A Training-Inversion-Exclusion Framework for Visually Interpretable and Uncertainty-Guided Out-of-Distribution Detection

2025-11-28

AI Summary

Deep neural networks often yield overconfident yet incorrect predictions on out-of-distribution (OOD) inputs, necessitating a unified framework for uncertainty quantification and OOD detection. This paper proposes TIE—a “Train-Invert-Exclude” framework that explicitly models OOD samples as a learnable junk class. By iteratively performing gradient-based inversion on in-distribution (ID) data, TIE generates uncertain samples that evolve into interpretable visual prototypes, rendering decision-making transparent. Crucially, TIE requires no external OOD data and performs end-to-end training solely on ID data. It integrates classifier extension, uncertainty-driven sample exclusion, and threshold-free evaluation (AUROC, AUPR, FPR@95%TPR). On MNIST and FashionMNIST, TIE achieves near-zero FPR@95%TPR—significantly outperforming state-of-the-art methods—while simultaneously delivering well-calibrated uncertainty estimates and visual interpretability.

Technology Category

• Machine Learning: Calibration & Uncertainty Quantification
• Reasoning under Uncertainty: Uncertainty Representations
• Computer Vision: Adversarial Attacks & Robustness

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Security and Privacy: Data transparency and provenance
• Graph Algorithms and Modeling for the Web: Graph neural networks and deep learning approaches for Web-related graphs

Abstract

Deep neural networks often struggle to recognize when an input lies outside their training experience, leading to unreliable and overconfident predictions. Building dependable machine learning systems therefore requires methods that can both estimate predictive extit{uncertainty} and detect extit{out-of-distribution (OOD)} samples in a unified manner. In this paper, we propose extbf{TIE: a Training--Inversion--Exclusion} framework for visually interpretable and uncertainty-guided anomaly detection that jointly addresses these challenges through iterative refinement. TIE extends a standard $n$-class classifier to an $(n+1)$-class model by introducing a garbage class initialized with Gaussian noise to represent outlier inputs. Within each epoch, TIE performs a closed-loop process of extit{training, inversion, and exclusion}, where highly uncertain inverted samples reconstructed from the just-trained classifier are excluded into the garbage class. Over successive iterations, the inverted samples transition from noisy artifacts into visually coherent class prototypes, providing transparent insight into how the model organizes its learned manifolds. During inference, TIE rejects OOD inputs by either directly mapping them to the garbage class or producing low-confidence, uncertain misclassifications within the in-distribution classes that are easily separable, all without relying on external OOD datasets. A comprehensive threshold-based evaluation using multiple OOD metrics and performance measures such as extit{AUROC}, extit{AUPR}, and extit{FPR@95%TPR} demonstrates that TIE offers a unified and interpretable framework for robust anomaly detection and calibrated uncertainty estimation (UE) achieving near-perfect OOD detection with extbf{(!approx!) 0 FPR@95%TPR} when trained on MNIST or FashionMNIST and tested against diverse unseen datasets.

Problem

• Develops a unified framework for out-of-distribution detection and uncertainty estimation
• Introduces an interpretable method using iterative training, inversion, and exclusion
• Achieves robust anomaly detection without relying on external OOD datasets

Innovation

• Extends classifier with garbage class for outlier representation
• Iteratively refines model via training inversion and exclusion
• Enables interpretable OOD detection without external datasets