Towards Explaining Anomalies: A Deep Taylor Decomposition of One-Class Models

Abstract

A common machine learning task is to discriminate between normal and anomalous data points. In practice, it is not always sufficient to reach high accuracy at this task, one also would like to understand why a given data point has been predicted in a certain way. We present a new principled approach for one-class SVMs that decomposes outlier predictions in terms of input variables. The method first recomposes the one-class model as a neural network with distance functions and min-pooling, and then performs a deep Taylor decomposition (DTD) of the model output. The proposed One-Class DTD is applicable to a number of common distance-based SVM kernels and is able to reliably explain a wide set of data anomalies. Furthermore, it outperforms baselines such as sensitivity analysis, nearest neighbor, or simple edge detection.

References in corpus (21)

• A Unified Approach to Interpreting Model Predictions
• Axiomatic Attribution for Deep Networks
• Striving for Simplicity: The All Convolutional Net
• Learning Important Features Through Propagating Activation Differences
• Quantum-Chemical Insights from Deep Tensor Neural Networks
• Explaining NonLinear Classification Decisions with Deep Taylor Decomposition
• Explainable Artificial Intelligence: Understanding, Visualizing and Interpreting Deep Learning Models
• Unmasking Clever Hans Predictors and Assessing What Machines Really Learn
• How to Explain Individual Classification Decisions
• PatternNet: A Benchmark Dataset for Performance Evaluation of Remote Sensing Image Retrieval
• Classifying and Segmenting Microscopy Images Using Convolutional Multiple Instance Learning
• Visualizing Deep Neural Network Decisions: Prediction Difference Analysis
• Toward Supervised Anomaly Detection
• "What is Relevant in a Text Document?": An Interpretable Machine Learning Approach
• Deep Structured Energy Based Models for Anomaly Detection
• Explaining How a Deep Neural Network Trained with End-to-End Learning Steers a Car
• Unsupervised Anomaly Detection with Generative Adversarial Networks to Guide Marker Discovery
• Convolutional Gaussian Processes
• The Doctor Just Won't Accept That!
• Exploring text datasets by visualizing relevant words
• Detecting Behavioral and Structural Anomalies in MediaCloud Applications

Cited by in corpus (24)

• Explaining Deep Neural Networks and Beyond: A Review of Methods and Applications
• A Unifying Review of Deep and Shallow Anomaly Detection
• Towards Explainable Artificial Intelligence
• Toward Explainable AI for Regression Models
• From Clustering to Cluster Explanations via Neural Networks
• DeepAID: Interpreting and Improving Deep Learning-based Anomaly Detection in Security Applications
• Explainable Deep One-Class Classification
• Explaining Anomalies Detected by Autoencoders Using SHAP
• Rule Extraction in Unsupervised Anomaly Detection for Model Explainability: Application to OneClass SVM
• Building and Interpreting Deep Similarity Models
• The Clever Hans Effect in Unsupervised Learning
• Counterfactual Explanation Based on Gradual Construction for Deep Networks
• How Much Can I Trust You? -- Quantifying Uncertainties in Explaining Neural Networks
• The Clever Hans Effect in Anomaly Detection
• Analyzing Atomic Interactions in Molecules as Learned by Neural Networks
• Insightful analysis of historical sources at scales beyond human capabilities using unsupervised Machine Learning and XAI
• Towards interpreting ML-based automated malware detection models: a survey
• Satellite galaxies' drag on field stars in the Milky Way
• Towards Robust Explanations for Deep Neural Networks
• Towards Interpretable Ensemble Learning for Image-based Malware Detection
• Framing Algorithmic Recourse for Anomaly Detection
• Fast and Accurate Explanations of Distance-Based Classifiers by Uncovering Latent Explanatory Structures
• Detection of Dataset Shifts in Learning-Enabled Cyber-Physical Systems using Variational Autoencoder for Regression
• Unsupervised Detection and Explanation of Latent-class Contextual Anomalies