Despite widespread adoption, machine learning models remain mostly black boxes. Understanding the reasons behind predictions is, however, quite important in assessing trust, which is fundamental if one plans to take action based on a prediction, or when choosing whether to deploy a new model. Such understanding also provides insights into the model, which can be used to transform an untrustworthy model or prediction into a trustworthy one. In this work, we propose LIME, a novel explanation technique that explains the predictions of any classifier in an interpretable and faithful manner, by learning an interpretable model locally varound the prediction. We also propose a method to explain models by presenting representative individual predictions and their explanations in a non-redundant way, framing the task as a submodular optimization problem. We demonstrate the flexibility of these methods by explaining different models for text (e.g. random forests) and image classification (e.g. neural networks). We show the utility of explanations via novel experiments, both simulated and with human subjects, on various scenarios that require trust: deciding if one should trust a prediction, choosing between models, improving an untrustworthy classifier, and identifying why a classifier should not be trusted.

/pdf/why-should-i-trust-you-explaining-the-predictions-of-any-ejtlrtykxr.pdf

"Why Should I Trust You?": Explaining the Predictions of Any Classifier

Deep neural networks are highly expressive models that have recently achieved state of the art performance on speech and visual recognition tasks. While their expressiveness is the reason they succeed, it also causes them to learn uninterpretable solutions that could have counter-intuitive properties. In this paper we report two such properties. 
First, we find that there is no distinction between individual high level units and random linear combinations of high level units, according to various methods of unit analysis. It suggests that it is the space, rather than the individual units, that contains of the semantic information in the high layers of neural networks. 
Second, we find that deep neural networks learn input-output mappings that are fairly discontinuous to a significant extend. We can cause the network to misclassify an image by applying a certain imperceptible perturbation, which is found by maximizing the network's prediction error. In addition, the specific nature of these perturbations is not a random artifact of learning: the same perturbation can cause a different network, that was trained on a different subset of the dataset, to misclassify the same input.

Intriguing properties of neural networks

This paper addresses the visualisation of image classification models, learnt using deep Convolutional Networks (ConvNets). We consider two visualisation techniques, based on computing the gradient of the class score with respect to the input image. The first one generates an image, which maximises the class score [5], thus visualising the notion of the class, captured by a ConvNet. The second technique computes a class saliency map, specific to a given image and class. We show that such maps can be employed for weakly supervised object segmentation using classification ConvNets. Finally, we establish the connection between the gradient-based ConvNet visualisation methods and deconvolutional networks [13].

/pdf/deep-inside-convolutional-networks-visualising-image-3a83mk6d8a.pdf

Deep Inside Convolutional Networks: Visualising Image Classification Models and Saliency Maps

Understanding and interpreting classification decisions of automated image classification systems is of high value in many applications, as it allows to verify the reasoning of the system and provides additional information to the human expert. Although machine learning methods are solving very successfully a plethora of tasks, they have in most cases the disadvantage of acting as a black box, not providing any information about what made them arrive at a particular decision. This work proposes a general solution to the problem of understanding classification decisions by pixel-wise decomposition of nonlinear classifiers. We introduce a methodology that allows to visualize the contributions of single pixels to predictions for kernel-based classifiers over Bag of Words features and for multilayered neural networks. These pixel contributions can be visualized as heatmaps and are provided to a human expert who can intuitively not only verify the validity of the classification decision, but also focus further analysis on regions of potential interest. We evaluate our method for classifiers trained on PASCAL VOC 2009 images, synthetic image data containing geometric shapes, the MNIST handwritten digits data set and for the pre-trained ImageNet model available as part of the Caffe open source package.

/pdf/on-pixel-wise-explanations-for-non-linear-classifier-4j59r9tlxm.pdf

On Pixel-Wise Explanations for Non-Linear Classifier Decisions by Layer-Wise Relevance Propagation.

This paper addresses the visualisation of image classification models, learnt using deep Convolutional Networks (ConvNets). We consider two visualisation techniques, based on computing the gradient of the class score with respect to the input image. The first one generates an image, which maximises the class score [Erhan et al., 2009], thus visualising the notion of the class, captured by a ConvNet. The second technique computes a class saliency map, specific to a given image and class. We show that such maps can be employed for weakly supervised object segmentation using classification ConvNets. Finally, we establish the connection between the gradient-based ConvNet visualisation methods and deconvolutional networks [Zeiler et al., 2013].

After building a classifier with modern tools of machine learning we typically have a black box at hand that is able to predict well for unseen data. Thus, we get an answer to the question what is the most likely label of a given unseen data point. However, most methods will provide no answer why the model predicted a particular label for a single instance and what features were most influential for that particular instance. The only method that is currently able to provide such explanations are decision trees. This paper proposes a procedure which (based on a set of assumptions) allows to explain the decisions of any classification method.

/pdf/how-to-explain-individual-classification-decisions-1diyu9ob2u.pdf

How to Explain Individual Classification Decisions

Statistical models are frequently used to estimate molecular properties, e.g., to establish quantitative structure-activity and structure-property relationships. For such models, interpretability, knowledge of the domain of applicability, and an estimate of confidence in the predictions are essential. We develop and validate a method for the interpretation of kernel-based prediction models. As a consequence of interpretability, the method helps to assess the domain of applicability of a model, to judge the reliability of a prediction, and to determine relevant molecular features. Increased interpretability also facilitates the acceptance of such models. Our method is based on visualization: For each prediction, the most contributing training samples are computed and visualized. We quantitatively show the effectiveness of our approach by conducting a questionnaire study with 71 participants, resulting in significant improvements of the participants' ability to distinguish between correct and incorrect predictions of a Gaussian process model for Ames mutagenicity.

Visual Interpretation of Kernel-Based Prediction Models.

After building a classifier with modern tools of machine learning we typically have a black box at hand that is able to predict well for unseen data. Thus, we get an answer to the question what is the most likely label of a given unseen data point. However, most methods will provide no answer why the model predicted the particular label for a single instance and what features were most influential for that particular instance. The only method that is currently able to provide such explanations are decision trees. This paper proposes a procedure which (based on a set of assumptions) allows to explain the decisions of any classification method.

Monitoring the occurrence of adverse events in the scientific literature is a mandatory process in drug marketing surveillance. This is a very time‐consuming and complex task to fulfill the compliance and, most importantly, to ensure patient safety. Therefore, a machine learning (ML) algorithm has been trained to support this manual intellectual review process, by automatically providing a classification of the literature articles into two types. An algorithm has been designed to automatically classify “relevant articles” which are reporting any kind of drug safety relevant information, and those which are not reporting an adverse drug reaction as “not relevant.” The review process is consisted of many rules and aspects which needed to be taken into consideration. Therefore, for the training of the algorithm, thousands of documents from previous screenings have been used. After several iterations of adjustments and fine tuning, the ML approach is definitively a great achievement in pre‐sorting the articles into “relevant” and “non‐relevant” and supporting the intellectual review process.

David Baehrens

Papers

How to Explain Individual Classification Decisions

Visual Interpretation of Kernel-Based Prediction Models.

How to Explain Individual Classification Decisions

Machine learning approach to identify adverse events in scientific biomedical literature

POSA318 Automation of Title and Abstract Screening: CAN Robots Replace Humans?