Published on Mon Feb 22 2021

Explainers in the Wild: Making Surrogate Explainers Robust to Distortions through Perception

Alexander Hepburn, Raul Santos-Rodriguez

Explanation of models is becoming pervasive in the image processing domain. Disparities to images can affect not only the predictions but also the explanation. We propose a methodology to assess the effect of distortions in explanations.

3
0
0
Abstract

Explaining the decisions of models is becoming pervasive in the image processing domain, whether it is by using post-hoc methods or by creating inherently interpretable models. While the widespread use of surrogate explainers is a welcome addition to inspect and understand black-box models, assessing the robustness and reliability of the explanations is key for their success. Additionally, whilst existing work in the explainability field proposes various strategies to address this problem, the challenges of working with data in the wild is often overlooked. For instance, in image classification, distortions to images can not only affect the predictions assigned by the model, but also the explanation. Given a clean and a distorted version of an image, even if the prediction probabilities are similar, the explanation may still be different. In this paper we propose a methodology to evaluate the effect of distortions in explanations by embedding perceptual distances that tailor the neighbourhoods used to training surrogate explainers. We also show that by operating in this way, we can make the explanations more robust to distortions. We generate explanations for images in the Imagenet-C dataset and demonstrate how using a perceptual distances in the surrogate explainer creates more coherent explanations for the distorted and reference images.

Wed Oct 14 2020
Artificial Intelligence
Human-interpretable model explainability on high-dimensional data
The importance of explainability in machine learning continues to grow. Unique challenges arise when a model's input features become high-dimensional. This means principled model-agnostic approaches to explainability become too computationally expensive.
0
0
0
Sat Mar 14 2020
Machine Learning
Measuring and improving the quality of visual explanations
The ability of to explain neural network decisions goes hand in hand with their safe deployment. We propose a new procedure for evaluating explanations. We use it to investigate visual explanations extracted from a range of possible sources in a neural network.
0
0
0
Wed Jun 05 2019
Machine Learning
c-Eval: A Unified Metric to Evaluate Feature-based Explanations via Perturbation
C-Eval is the minimum-distortion perturbation that successfully alters a classifier's prediction while keeping the explanation's features unchanged. The metric can be computed using some modifications to existing adversarial generation libraries.
0
0
0
Wed Oct 09 2019
Machine Learning
Explaining image classifiers by removing input features using generative models
Perturbation-based explanation methods often measure the contribution of an input feature to an image classifier's outputs by heuristically removing it via blurring, adding noise, or graying out. Instead, we propose to integrate a generative inpainter into three representative attribution methods.
0
0
0
Thu Mar 18 2021
Machine Learning
Beyond Trivial Counterfactual Explanations with Diverse Valuable Explanations
In computer vision applications, generative counterfactual methods indicate how to perturb a model's input to change its prediction. Current counterfactuals make ambiguous interpretations as they combine multiple biases of the model and the data in a single interpretation.
1
2
6
Sat May 29 2021
Machine Learning
EDDA: Explanation-driven Data Augmentation to Improve Model and Explanation Alignment
Explanations may not always align perfectly with classifier predictions. We seek a methodology that can improve alignment between model predictions and explanation method. We achieve this through a novel explanation-driven data augmentation (EDDA) method.
4
0
0
Tue Feb 16 2016
Machine Learning
"Why Should I Trust You?": Explaining the Predictions of Any Classifier
Despite widespread adoption, machine learning models remain mostly black boxes. LIME is a novel explanation technique that explains the predictions of any classifier in an interpretable and faithful manner.
6
0
11
Wed Dec 02 2015
Computer Vision
Rethinking the Inception Architecture for Computer Vision
Convolutional networks are at the core of most state-of-the-art computer vision solutions. Here we explore ways to scale up networks in ways that aim at utilizing the added computation as efficiently as possible.
0
0
0
Wed Jul 04 2018
Neural Networks
Benchmarking Neural Network Robustness to Common Corruptions and Surface Variations
ImageNet-C standardizes and expands the corruption robustness topic, while showing which classifiers are preferable in safety-critical applications. We find that there are negligible changes in relative corruption robustness from AlexNet to ResNet classifiers.
0
0
0
Tue Oct 29 2019
Machine Learning
bLIMEy: Surrogate Prediction Explanations Beyond LIME
Surrogate explainers of black-box machine learning predictions are of paramount importance in the field of eXplainable Artificial Intelligence. The Local Interpretable Model-agnostic Explanations (LIME) algorithm is often mistakenly unified with a more general framework of surrogate explainers.
0
0
0
Wed Mar 20 2019
Machine Learning
Approximating CNNs with Bag-of-local-Features models works surprisingly well on ImageNet
Deep Neural Networks (DNNs) excel on many complex perceptual tasks but it has proven notoriously difficult to understand how they reach their decisions. We introduce a high-performance DNN architecture on ImageNet whose decisions are considerably easier to explain.
0
0
0
Mon Jan 23 2017
Artificial Intelligence
Perceptually Optimized Image Rendering
We develop a framework for rendering photographic images, taking into account display limitations. We minimize a measure of perceptual similarity, the Normalized Laplacian Pyramid Distance. When rendering images acquired with higher dynamic range than that of the display, we find that the optimized solution boosts the contrast of low-
0
0
0