Interpretable machine learning tackles the important problem that humans cannot understand the behaviors of complex machine learning models and how these models arrive at a particular decision. Although many approaches have been proposed, a comprehensive understanding of the achievements and challenges is still lacking. We provide a survey covering existing techniques to increase the interpretability of machine learning models. We also discuss crucial issues that the community should consider in future work such as designing user-friendly explanations and developing comprehensive evaluation metrics to further push forward the area of interpretable machine learning.
Interpretability of the underlying AI representations is a key raison d'\^{e}tre for Open Learner Modelling (OLM) -- a branch of Intelligent Tutoring Systems (ITS) research. OLMs provide tools for 'opening' up the AI models of learners' cognition and emotions for the purpose of supporting human learning and teaching. Over thirty years of research in ITS (also known as AI in Education) produced important work, which informs about how AI can be used in Education to best effects and, through the OLM research, what are the necessary considerations to make it interpretable and explainable for the benefit of learning. We argue that this work can provide a valuable starting point for a framework of interpretable AI, and as such is of relevance to the application of both knowledge-based and machine learning systems in other high-stakes contexts, beyond education.
Several researchers have argued that a machine learning system's interpretability should be defined in relation to a specific agent or task: we should not ask if the system is interpretable, but to whom is it interpretable. We describe a model intended to help answer this question, by identifying different roles that agents can fulfill in relation to the machine learning system. We illustrate the use of our model in a variety of scenarios, exploring how an agent's role influences its goals, and the implications for defining interpretability. Finally, we make suggestions for how our model could be useful to interpretability researchers, system developers, and regulatory bodies auditing machine learning systems.
Recent advances in interpretable Machine Learning (iML) and eXplainable AI (XAI) construct explanations based on the importance of features in classification tasks. However, in a high-dimensional feature space this approach may become unfeasible without restraining the set of important features. We propose to utilize the human tendency to ask questions like "Why this output (the fact) instead of that output (the foil)?" to reduce the number of features to those that play a main role in the asked contrast. Our proposed method utilizes locally trained one-versus-all decision trees to identify the disjoint set of rules that causes the tree to classify data points as the foil and not as the fact. In this study we illustrate this approach on three benchmark classification tasks.
Deep neural networks are being used increasingly to automate data analysis and decision making, yet their decision-making process is largely unclear and is difficult to explain to the end users. In this paper, we address the problem of Explainable AI for deep neural networks that take images as input and output a class probability. We propose an approach called RISE that generates an importance map indicating how salient each pixel is for the model's prediction. In contrast to white-box approaches that estimate pixel importance using gradients or other internal network state, RISE works on black-box models. It estimates importance empirically by probing the model with randomly masked versions of the input image and obtaining the corresponding outputs. We compare our approach to state-of-the-art importance extraction methods using both an automatic deletion/insertion metric and a pointing metric based on human-annotated object segments. Extensive experiments on several benchmark datasets show that our approach matches or exceeds the performance of other methods, including white-box approaches. Project page: http://cs-people.bu.edu/vpetsiuk/rise/
Recent years have produced great advances in training large, deep neural networks (DNNs), including notable successes in training convolutional neural networks (convnets) to recognize natural images. However, our understanding of how these models work, especially what computations they perform at intermediate layers, has lagged behind. Progress in the field will be further accelerated by the development of better tools for visualizing and interpreting neural nets. We introduce two such tools here. The first is a tool that visualizes the activations produced on each layer of a trained convnet as it processes an image or video (e.g. a live webcam stream). We have found that looking at live activations that change in response to user input helps build valuable intuitions about how convnets work. The second tool enables visualizing features at each layer of a DNN via regularized optimization in image space. Because previous versions of this idea produced less recognizable images, here we introduce several new regularization methods that combine to produce qualitatively clearer, more interpretable visualizations. Both tools are open source and work on a pre-trained convnet with minimal setup.
Scoring systems are linear classification models that only require users to add, subtract and multiply a few small numbers in order to make a prediction. These models are in widespread use by the medical community, but are difficult to learn from data because they need to be accurate and sparse, have coprime integer coefficients, and satisfy multiple operational constraints. We present a new method for creating data-driven scoring systems called a Supersparse Linear Integer Model (SLIM). SLIM scoring systems are built by solving an integer program that directly encodes measures of accuracy (the 0-1 loss) and sparsity (the $\ell_0$-seminorm) while restricting coefficients to coprime integers. SLIM can seamlessly incorporate a wide range of operational constraints related to accuracy and sparsity, and can produce highly tailored models without parameter tuning. We provide bounds on the testing and training accuracy of SLIM scoring systems, and present a new data reduction technique that can improve scalability by eliminating a portion of the training data beforehand. Our paper includes results from a collaboration with the Massachusetts General Hospital Sleep Laboratory, where SLIM was used to create a highly tailored scoring system for sleep apnea screening
As a contribution to interpretable machine learning research, we develop a novel optimization framework for learning accurate and sparse two-level Boolean rules. We consider rules in both conjunctive normal form (AND-of-ORs) and disjunctive normal form (OR-of-ANDs). A principled objective function is proposed to trade classification accuracy and interpretability, where we use Hamming loss to characterize accuracy and sparsity to characterize interpretability. We propose efficient procedures to optimize these objectives based on linear programming (LP) relaxation, block coordinate descent, and alternating minimization. Experiments show that our new algorithms provide very good tradeoffs between accuracy and interpretability.
The Clock Drawing Test (CDT) is a rapid, inexpensive, and popular neuropsychological screening tool for cognitive conditions. The Digital Clock Drawing Test (dCDT) uses novel software to analyze data from a digitizing ballpoint pen that reports its position with considerable spatial and temporal precision, making possible the analysis of both the drawing process and final product. We developed methodology to analyze pen stroke data from these drawings, and computed a large collection of features which were then analyzed with a variety of machine learning techniques. The resulting scoring systems were designed to be more accurate than the systems currently used by clinicians, but just as interpretable and easy to use. The systems also allow us to quantify the tradeoff between accuracy and interpretability. We created automated versions of the CDT scoring systems currently used by clinicians, allowing us to benchmark our models, which indicated that our machine learning models substantially outperformed the existing scoring systems.
The last decade has seen huge progress in the development of advanced machine learning models; however, those models are powerless unless human users can interpret them. Here we show how the mind's construction of concepts and meaning can be used to create more interpretable machine learning models. By proposing a novel method of classifying concepts, in terms of 'form' and 'function', we elucidate the nature of meaning and offer proposals to improve model understandability. As machine learning begins to permeate daily life, interpretable models may serve as a bridge between domain-expert authors and non-expert users.
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 around 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.
At the core of interpretable machine learning is the question of whether humans are able to make accurate predictions about a model's behavior. Assumed in this question are three properties of the interpretable output: coverage, precision, and effort. Coverage refers to how often humans think they can predict the model's behavior, precision to how accurate humans are in those predictions, and effort is either the up-front effort required in interpreting the model, or the effort required to make predictions about a model's behavior. In this work, we propose anchor-LIME (aLIME), a model-agnostic technique that produces high-precision rule-based explanations for which the coverage boundaries are very clear. We compare aLIME to linear LIME with simulated experiments, and demonstrate the flexibility of aLIME with qualitative examples from a variety of domains and tasks.
Models obtained by decision tree induction techniques excel in being interpretable.However, they can be prone to overfitting, which results in a low predictive performance. Ensemble techniques are able to achieve a higher accuracy. However, this comes at a cost of losing interpretability of the resulting model. This makes ensemble techniques impractical in applications where decision support, instead of decision making, is crucial. To bridge this gap, we present the GENESIM algorithm that transforms an ensemble of decision trees to a single decision tree with an enhanced predictive performance by using a genetic algorithm. We compared GENESIM to prevalent decision tree induction and ensemble techniques using twelve publicly available data sets. The results show that GENESIM achieves a better predictive performance on most of these data sets than decision tree induction techniques and a predictive performance in the same order of magnitude as the ensemble techniques. Moreover, the resulting model of GENESIM has a very low complexity, making it very interpretable, in contrast to ensemble techniques.
Text documents can be described by a number of abstract concepts such as semantic category, writing style, or sentiment. Machine learning (ML) models have been trained to automatically map documents to these abstract concepts, allowing to annotate very large text collections, more than could be processed by a human in a lifetime. Besides predicting the text's category very accurately, it is also highly desirable to understand how and why the categorization process takes place. In this paper, we demonstrate that such understanding can be achieved by tracing the classification decision back to individual words using layer-wise relevance propagation (LRP), a recently developed technique for explaining predictions of complex non-linear classifiers. We train two word-based ML models, a convolutional neural network (CNN) and a bag-of-words SVM classifier, on a topic categorization task and adapt the LRP method to decompose the predictions of these models onto words. Resulting scores indicate how much individual words contribute to the overall classification decision. This enables one to distill relevant information from text documents without an explicit semantic information extraction step. We further use the word-wise relevance scores for generating novel vector-based document representations which capture semantic information. Based on these document vectors, we introduce a measure of model explanatory power and show that, although the SVM and CNN models perform similarly in terms of classification accuracy, the latter exhibits a higher level of explainability which makes it more comprehensible for humans and potentially more useful for other applications.
As machine learning systems become ubiquitous, there has been a surge of interest in interpretable machine learning: systems that provide explanation for their outputs. These explanations are often used to qualitatively assess other criteria such as safety or non-discrimination. However, despite the interest in interpretability, there is very little consensus on what interpretable machine learning is and how it should be measured. In this position paper, we first define interpretability and describe when interpretability is needed (and when it is not). Next, we suggest a taxonomy for rigorous evaluation and expose open questions towards a more rigorous science of interpretable machine learning.
We propose a technique for producing "visual explanations" for decisions from a large class of CNN-based models, making them more transparent. Our approach - Gradient-weighted Class Activation Mapping (Grad-CAM), uses the gradients of any target concept, flowing into the final convolutional layer to produce a coarse localization map highlighting important regions in the image for predicting the concept. Grad-CAM is applicable to a wide variety of CNN model-families: (1) CNNs with fully-connected layers, (2) CNNs used for structured outputs, (3) CNNs used in tasks with multimodal inputs or reinforcement learning, without any architectural changes or re-training. We combine Grad-CAM with fine-grained visualizations to create a high-resolution class-discriminative visualization and apply it to off-the-shelf image classification, captioning, and visual question answering (VQA) models, including ResNet-based architectures. In the context of image classification models, our visualizations (a) lend insights into their failure modes, (b) are robust to adversarial images, (c) outperform previous methods on localization, (d) are more faithful to the underlying model and (e) help achieve generalization by identifying dataset bias. For captioning and VQA, we show that even non-attention based models can localize inputs. We devise a way to identify important neurons through Grad-CAM and combine it with neuron names to provide textual explanations for model decisions. Finally, we design and conduct human studies to measure if Grad-CAM helps users establish appropriate trust in predictions from models and show that Grad-CAM helps untrained users successfully discern a 'stronger' nodel from a 'weaker' one even when both make identical predictions. Our code is available at https://github.com/ramprs/grad-cam/, along with a demo at http://gradcam.cloudcv.org, and a video at youtu.be/COjUB9Izk6E.
The purported "black box" nature of neural networks is a barrier to adoption in applications where interpretability is essential. Here we present DeepLIFT (Deep Learning Important FeaTures), a method for decomposing the output prediction of a neural network on a specific input by backpropagating the contributions of all neurons in the network to every feature of the input. DeepLIFT compares the activation of each neuron to its 'reference activation' and assigns contribution scores according to the difference. By optionally giving separate consideration to positive and negative contributions, DeepLIFT can also reveal dependencies which are missed by other approaches. Scores can be computed efficiently in a single backward pass. We apply DeepLIFT to models trained on MNIST and simulated genomic data, and show significant advantages over gradient-based methods. Video tutorial: http://goo.gl/qKb7pL, ICML slides: bit.ly/deeplifticmlslides, ICML talk: https://vimeo.com/238275076, code: http://goo.gl/RM8jvH.
Explaining the output of a deep network remains a challenge. In the case of an image classifier, one type of explanation is to identify pixels that strongly influence the final decision. A starting point for this strategy is the gradient of the class score function with respect to the input image. This gradient can be interpreted as a sensitivity map, and there are several techniques that elaborate on this basic idea. This paper makes two contributions: it introduces SmoothGrad, a simple method that can help visually sharpen gradient-based sensitivity maps, and it discusses lessons in the visualization of these maps. We publish the code for our experiments and a website with our results.
We study the problem of attributing the prediction of a deep network to its input features, a problem previously studied by several other works. We identify two fundamental axioms---Sensitivity and Implementation Invariance that attribution methods ought to satisfy. We show that they are not satisfied by most known attribution methods, which we consider to be a fundamental weakness of those methods. We use the axioms to guide the design of a new attribution method called Integrated Gradients. Our method requires no modification to the original network and is extremely simple to implement; it just needs a few calls to the standard gradient operator. We apply this method to a couple of image models, a couple of text models and a chemistry model, demonstrating its ability to debug networks, to extract rules from a network, and to enable users to engage with models better.