Towards Robust Explanations for Deep Neural Networks
arXiv:2012.10425
Abstract
Explanation methods shed light on the decision process of black-box classifiers such as deep neural networks. But their usefulness can be compromised because they are susceptible to manipulations. With this work, we aim to enhance the resilience of explanations. We develop a unified theoretical framework for deriving bounds on the maximal manipulability of a model. Based on these theoretical insights, we present three different techniques to boost robustness against manipulation: training with weight decay, smoothing activation functions, and minimizing the Hessian of the network. Our experimental results confirm the effectiveness of these approaches.
References in corpus (19)
- Axiomatic Attribution for Deep Networks
- Methods for Interpreting and Understanding Deep Neural Networks
- Learning Important Features Through Propagating Activation Differences
- Quantum-Chemical Insights from Deep Tensor Neural Networks
- Explaining NonLinear Classification Decisions with Deep Taylor Decomposition
- SmoothGrad: removing noise by adding noise
- How to Explain Individual Classification Decisions
- Sanity Checks for Saliency Maps
- A Benchmark for Interpretability Methods in Deep Neural Networks
- Learning how to explain neural networks: PatternNet and PatternAttribution
- "What is Relevant in a Text Document?": An Interpretable Machine Learning Approach
- Visualizing and Understanding Atari Agents
- From Clustering to Cluster Explanations via Neural Networks
- Towards Explaining Anomalies: A Deep Taylor Decomposition of One-Class Models
- Building and Interpreting Deep Similarity Models
- Graying the black box: Understanding DQNs
- Fairwashing Explanations with Off-Manifold Detergent
- Smoothed Geometry for Robust Attribution
- Gradient-based Analysis of NLP Models is Manipulable