Rethinking the Role of Gradient-Based Attribution Methods for Model Interpretability

Abstract

Current methods for the interpretability of discriminative deep neural networks commonly rely on the model's input-gradients, i.e., the gradients of the output logits w.r.t. the inputs. The common assumption is that these input-gradients contain information regarding $p_{\theta }(y\mid x)$, the model's discriminative capabilities, thus justifying their use for interpretability. However, in this work we show that these input-gradients can be arbitrarily manipulated as a consequence of the shift-invariance of softmax without changing the discriminative function. This leaves an open question: if input-gradients can be arbitrary, why are they highly structured and explanatory in standard models? We investigate this by re-interpreting the logits of standard softmax-based classifiers as unnormalized log-densities of the data distribution and show that input-gradients can be viewed as gradients of a class-conditional density model $p_{\theta }(x\mid y)$ implicit within the discriminative model. This leads us to hypothesize that the highly structured and explanatory nature of input-gradients may be due to the alignment of this class-conditional model $p_{\theta }(x\mid y)$ with that of the ground truth data distribution $p_{\text{data}}(x\mid y)$. We test this hypothesis by studying the effect of density alignment on gradient explanations. To achieve this alignment we use score-matching, and propose novel approximations to this algorithm to enable training large-scale models. Our experiments show that improving the alignment of the implicit density model with the data distribution enhances gradient structure and explanatory power while reducing this alignment has the opposite effect. Overall, our finding that input-gradients capture information regarding an implicit generative model implies that we need to re-think their use for interpreting discriminative models.

arXiv commentOral Presentation at ICLR 2021

References in corpus (12)

• Learning Important Features Through Propagating Activation Differences
• Generative Modeling by Estimating Gradients of the Data Distribution
• SmoothGrad: removing noise by adding noise
• Sanity Checks for Saliency Maps
• Visualizing Deep Convolutional Neural Networks Using Natural Pre-Images
• Improving the Adversarial Robustness and Interpretability of Deep Neural Networks by Regularizing their Input Gradients
• Fooling Neural Network Interpretations via Adversarial Model Manipulation
• Image Synthesis with a Single (Robust) Classifier
• Full-Gradient Representation for Neural Network Visualization
• Adversarial Robustness as a Prior for Learned Representations
• Sliced Score Matching: A Scalable Approach to Density and Score Estimation
• Are Perceptually-Aligned Gradients a General Property of Robust Classifiers?

Cited by in corpus (1)

• Physics-Inspired Interpretability Of Machine Learning Models