Input perturbation methods occlude parts of an input to a function and measure the change in the function's output. Recently, input perturbation methods have been applied to generate and evaluate saliency maps from convolutional neural networks. In practice, neutral baseline images are used for the occlusion, such that the baseline image's impact on the classification probability is minimal. However, in this paper we show that arguably neutral baseline images still impact the generated saliency maps and their evaluation with input perturbations. We also demonstrate that many choices of hyperparameters lead to the divergence of saliency maps generated by input perturbations. We experimentally reveal inconsistencies among a selection of input perturbation methods and find that they lack robustness for generating saliency maps and for evaluating saliency maps as saliency metrics.
Convolutional neural networks (CNNs) are commonly used for image classification. Saliency methods are examples of approaches that can be used to interpret CNNs post hoc, identifying the most relevant pixels for a prediction following the gradients flow. Even though CNNs can correctly classify images, the underlying saliency maps could be erroneous in many cases. This can result in skepticism as to the validity of the model or its interpretation. We propose a novel approach for training trustworthy CNNs by penalizing parameter choices that result in inaccurate saliency maps generated during tra…
A fundamental bottleneck in utilising complex machine learning systems for critical applications has been not knowing why they do and what they do, thus preventing the development of any crucial safety protocols. To date, no method exist that can provide full insight into the granularity of the neural network's decision process. In the past, saliency maps were an early attempt at resolving this problem through sensitivity calculations, whereby dimensions of a data point are selected based on how sensitive the output of the system is to them. However, the success of saliency maps has been at be…
Visualisation, Faculty of Computer Science, Technische Universit\" a t Dresden, N\" o thnitzer Stra e 46, 01187 Dresden, Germany
Image fusion helps in merging two or more images to construct a more informative single fused image. Recently, unsupervised learning based convolutional neural networks (CNN) have been utilized for different types of image fusion tasks such as medical image fusion, infrared-visible image fusion for autonomous driving as well as multi-focus and multi-exposure image fusion for satellite imagery. However, it is challenging to analyze the reliability of these CNNs for the image fusion tasks since no groundtruth is available. This led to the use of a wide variety of model architectures and optimiza…
Peking University（北京大学）；Beijing Institute of Big Data Research（北京大数据研究院）
We attempt to interpret how adversarially trained convolutional neural networks (AT-CNNs) recognize objects. We design systematic approaches to interpret AT-CNNs in both qualitative and quantitative ways and compare them with normally trained models. Surprisingly, we find that adversarial training alleviates the texture bias of standard CNNs when trained on object recognition tasks, and helps CNNs learn a more shape-biased representation. We validate our hypothesis from two aspects. First, we compare the salience maps of AT-CNNs and standard CNNs on clean images and images under different tran…
Saliency Map, the gradient of the score function with respect to the input, is the most basic technique for interpreting deep neural network decisions. However, saliency maps are often visually noisy. Although several hypotheses were proposed to account for this phenomenon, there are few works that provide rigorous analyses of noisy saliency maps. In this paper, we firstly propose a new hypothesis that noise may occur in saliency maps when irrelevant features pass through ReLU activation functions. Then, we propose Rectified Gradient, a method that alleviates this problem through layer-wise th…
Understanding the predictions made by Artificial Intelligence (AI) systems is becoming more and more important as deep learning models are used for increasingly complex and high-stakes tasks. Saliency mapping -- a popular visual attribution method -- is one important tool for this, but existing formulations are limited by either computational cost or architectural constraints. We therefore propose Hierarchical Perturbation, a very fast and completely model-agnostic method for interpreting model predictions with robust saliency maps. Using standard benchmarks and datasets, we show that our sali…
Convolutional Neural Networks (CNNs) compression is crucial to deploying these models in edge devices with limited resources. Existing channel pruning algorithms for CNNs have achieved plenty of success on complex models. They approach the pruning problem from various perspectives and use different metrics to guide the pruning process. However, these metrics mainly focus on the model's `outputs' or `weights' and neglect its `interpretations' information. To fill in this gap, we propose to address the channel pruning problem from a novel perspective by leveraging the interpretations of a model …
Clinical applicability of automated decision support systems depends on a robust, well-understood classification interpretation. Artificial neural networks while achieving class-leading scores fall short in this regard. Therefore, numerous approaches have been proposed that map a salient region of an image to a diagnostic classification. Utilizing heuristic methodology, like blurring and noise, they tend to produce diffuse, sometimes misleading results, hindering their general adoption. In this work we overcome these issues by presenting a model agnostic saliency mapping framework tailored to …
Image classification based on machine learning is being commonly used. However, a classification result given by an advanced method, including deep learning, is often hard to interpret. This problem of interpretability is one of the major obstacles in deploying a trained model in safety-critical systems. Several techniques have been proposed to address this problem; one of which is RISE, which explains a classification result by a heatmap, called a saliency map, which explains the significance of each pixel. We propose MC-RISE (Multi-Color RISE), which is an enhancement of RISE to take color i…
University of Cambridge（剑桥大学）；The Alan Turing Institute（艾伦·图灵研究所）
We investigate the influence of adversarial training on the interpretability of convolutional neural networks (CNNs), specifically applied to diagnosing skin cancer. We show that gradient-based saliency maps of adversarially trained CNNs are significantly sharper and more visually coherent than those of standardly trained CNNs. Furthermore, we show that adversarially trained networks highlight regions with significant color variation within the lesion, a common characteristic of melanoma. We find that fine-tuning a robust network with a small learning rate further improves saliency maps' sharp…
A particular class of Explainable AI (XAI) methods provide saliency maps to highlight part of the image a Convolutional Neural Network (CNN) model looks at to classify the image as a way to explain its working. These methods provide an intuitive way for users to understand predictions made by CNNs. Other than quantitative computational tests, the vast majority of evidence to highlight that the methods are valuable is anecdotal. Given that humans would be the end-users of such methods, we devise three human subject experiments through which we gauge the effectiveness of these saliency-based exp…
Convolutional neural networks (CNNs) achieve state-of-the-art performance in a wide variety of tasks in computer vision. However, interpreting CNNs still remains a challenge. This is mainly due to the large number of parameters in these networks. Here, we investigate the role of compression and particularly pruning filters in the interpretation of CNNs. We exploit our recently-proposed greedy structural compression scheme that prunes filters in a trained CNN. In our compression, the filter importance index is defined as the classification accuracy reduction (CAR) of the network after pruning t…
Brigham and Women's Hospital（布莱根妇女医院）；Harvard Medical School（哈佛医学院）；Massachusetts General Hospital（麻省总医院）
Background: MRI is the modality of choice for cartilage imaging; however, its diagnostic performance is variable and significantly lower than the gold standard diagnostic knee arthroscopy. In recent years, deep learning has been used to automatically interpret medical images to improve diagnostic accuracy and speed. Purpose: The primary purpose of this study was to evaluate whether deep learning applied to the interpretation of knee MRI images can be utilized to identify cartilage defects accurately. Methods: We analyzed data from patients who underwent knee MRI evaluation and consequently had…
Healthcare Technology and Innovation Center (HTIC), Indian Institute of Technology (IIT-M)（印度理工学院马德拉斯分校医疗技术与创新中心）；Indian Institute of Technology, Madras (IITM)（印度理工学院马德拉斯分校）
Cardiac arrhythmia is a prevalent and significant cause of morbidity and mortality among cardiac ailments. Early diagnosis is crucial in providing intervention for patients suffering from cardiac arrhythmia. Traditionally, diagnosis is performed by examination of the Electrocardiogram (ECG) by a cardiologist. This method of diagnosis is hampered by the lack of accessibility to expert cardiologists. For quite some time, signal processing methods had been used to automate arrhythmia diagnosis. However, these traditional methods require expert knowledge and are unable to model a wide range of arr…
Interpretability is a critical factor in applying complex deep learning models to advance the understanding of brain disorders in neuroimaging studies. To interpret the decision process of a trained classifier, existing techniques typically rely on saliency maps to quantify the voxel-wise or feature-level importance for classification through partial derivatives. Despite providing some level of localization, these maps are not human-understandable from the neuroscience perspective as they do not inform the specific meaning of the alteration linked to the brain disorder. Inspired by the image-t…
Deep learning architectures are an extremely powerful tool for recognizing and classifying images. However, they require supervised learning and normally work on vectors the size of image pixels and produce the best results when trained on millions of object images. To help mitigate these issues, we propose the fusion of bottom-up saliency and top-down attention employing only unsupervised learning techniques, which helps the object recognition module to focus on relevant data and learn important features that can later be fine-tuned for a specific task. In addition, by utilizing only relevant…
Nagoya Institute of Technology（名古屋工业大学）；RIKEN（理化学研究所）；Nagoya University（名古屋大学）
Various saliency map methods have been proposed to interpret and explain predictions of deep learning models. Saliency maps allow us to interpret which parts of the input signals have a strong influence on the prediction results. However, since a saliency map is obtained by complex computations in deep learning models, it is often difficult to know how reliable the saliency map itself is. In this study, we propose a method to quantify the reliability of a salient region in the form of p-values. Our idea is to consider a salient region as a selected hypothesis by the trained deep learning model…
Modern deep learning models for NLP are notoriously opaque. This has motivated the development of methods for interpreting such models, e.g., via gradient-based saliency maps or the visualization of attention weights. Such approaches aim to provide explanations for a particular model prediction by highlighting important words in the corresponding input text. While this might be useful for tasks where decisions are explicitly influenced by individual tokens in the input, we suspect that such highlighting is not suitable for tasks where model decisions should be driven by more complex reasoning.…
TU Berlin（柏林工业大学）；Fraunhofer HHI（弗劳恩霍夫海因里希·赫兹研究所）；Leibniz Institute of Agricultural Engineering and Bioeconomy e.V. (ATB)（莱布尼茨农业工程与生物经济研究所）；BIFOLD – Berlin Institute for the Foundations of Learning and Data（柏林学习与数据基础研究所）；University of Potsdam（波茨坦大学）
University of Bremen（不来梅大学）；German Research Center for Artificial Intelligence（德国人工智能研究中心）
Deep Reinforcement Learning (DRL) connects the classic Reinforcement Learning algorithms with Deep Neural Networks. A problem in DRL is that CNNs are black-boxes and it is hard to understand the decision-making process of agents. In order to be able to use RL agents in highly dangerous environments for humans and machines, the developer needs a debugging tool to assure that the agent does what is expected. Currently, rewards are primarily used to interpret how well an agent is learning. However, this can lead to deceptive conclusions if the agent receives more rewards by memorizing a policy an…
MIT CSAIL（麻省理工学院计算机科学与人工智能实验室）；IBM Research（IBM研究院）
Saliency methods -- techniques to identify the importance of input features on a model's output -- are a common step in understanding neural network behavior. However, interpreting saliency requires tedious manual inspection to identify and aggregate patterns in model behavior, resulting in ad hoc or cherry-picked analysis. To address these concerns, we present Shared Interest: metrics for comparing model reasoning (via saliency) to human reasoning (via ground truth annotations). By providing quantitative descriptors, Shared Interest enables ranking, sorting, and aggregating inputs, thereby fa…
University of California, Los Angeles（加州大学洛杉矶分校）；Jet Propulsion Laboratory, California Institute of Technology（加州理工学院喷气推进实验室）
Deep Convolutional Neural Networks (CNNs) have demonstrated excellent performance in image classification, but still show room for improvement in object-detection tasks with many categories, in particular for cluttered scenes and occlusion. Modern detection algorithms like Regions with CNNs (Girshick et al., 2014) rely on Selective Search (Uijlings et al., 2013) to propose regions which with high probability represent objects, where in turn CNNs are deployed for classification. Selective Search represents a family of sophisticated algorithms that are engineered with multiple segmentation, appe…
Learning and Interpreting Gravitational-Wave Features from CNNs with a Random Forest Approach
School of Physics and Astronomy, China West Normal University（物理与天文学院，中国西南师范大学）；International Centre for Theoretical Physics Asia-Pacific (ICTP-AP), University of Chinese Academy of Sciences (UCAS)（亚洲太平洋理论物理中心（ICTP-AP），中国科学院大学）；Taiji Laboratory for Gravitational Wave Universe (Beijing/Hangzhou), University of Chinese Academy of Sciences (UCAS)（引力波宇宙 Taiji 实验室（北京/杭州），中国科学院大学）；Hangzhou Institute for Advanced Study, UCAS（杭州先进研究所，中国科学院大学）；School of Electronic Science and Engineering, Chongqing University of Posts and Telecommunications（电子科学与工程学院，重庆邮电大学）；Institute for Frontiers in Astronomy and Astrophysics, Beijing Normal University（天文与天体物理学前沿研究所，北京师范大学）；School of Physics and Astronomy, Beijing Normal University（物理与天文学院，北京师范大学）
Gradient-based analysis methods, such as saliency map visualizations and adversarial input perturbations, have found widespread use in interpreting neural NLP models due to their simplicity, flexibility, and most importantly, their faithfulness. In this paper, however, we demonstrate that the gradients of a model are easily manipulable, and thus bring into question the reliability of gradient-based analyses. In particular, we merge the layers of a target model with a Facade that overwhelms the gradients without affecting the predictions. This Facade can be trained to have gradients that are mi…
Towards Scalable Insect Monitoring: Ultra-Lightweight CNNs as On-Device Triggers for Insect Camera Traps
University of Exeter（埃克塞特大学）；Institute for Data Science and AI, University of Exeter（埃克塞特大学数据科学与人工智能学院）；Centre for Ecology and Conservation, University of Exeter（埃克塞特大学生态与保护中心）
针对 PIR 传感器难以触发昆虫相机陷阱的问题，本文提出在低功耗微控制器上运行超轻量级 CNN 检测连续图像中的昆虫，实现零延迟、高 AUC、低功耗与更长部署。
