Understanding the Role of Individual Units in a Deep Neural Network
Computer Science and Artificial Intelligence Laboratory, Massachusetts Institute of Technology（麻省理工学院计算机科学与人工智能实验室）；Adobe Research, Adobe Inc.（Adobe公司研究院）；Massachusetts Institute of Technology–International Business Machines (IBM) Watson Artificial Intelligence Laboratory（麻省理工学院-IBM沃森人工智能实验室）；Media Lab, Massachusetts Institute of Technology（麻省理工学院媒体实验室）；Universitat Oberta de Catalunya（加泰罗尼亚开放大学）；The Chinese University of Hong Kong（香港中文大学）
David Bau, Jun-Yan Zhu, Hendrik Strobelt, Agata Lapedriza, Bolei Zhou, Antonio Torralba
Deep neural networks excel at finding hierarchical representations that solve complex tasks over large data sets. How can we humans understand these learned representations? In this work, we present network dissection, an analytic framework to systematically identify the semantics of individual hidden units within image classification and image generation networks. First, we analyze a convolutional neural network (CNN) trained on scene classification and discover units that match a diverse set of object concepts. We find evidence that the network has learned many object classes that play crucial roles in classifying scene classes. Second, we use a similar analytic method to analyze a generative adversarial network (GAN) model trained to generate scenes. By analyzing changes made when small sets of units are activated or deactivated, we find that objects can be added and removed from the output scenes while adapting to the context. Finally, we apply our analytic framework to understanding adversarial attacks and to semantic image editing.
Massachusetts Institute of Technology（麻省理工学院）；MIT-IBM Watson AI Lab（MIT-IBM沃森人工智能实验室）；IBM Research（IBM研究院）；The Chinese University of Hong Kong（香港中文大学）
Generative Adversarial Networks (GANs) have recently achieved impressive results for many real-world applications, and many GAN variants have emerged with improvements in sample quality and training stability. However, they have not been well visualized or understood. How does a GAN represent our visual world internally? What causes the artifacts in GAN results? How do architectural choices affect GAN learning? Answering such questions could enable us to develop new insights and better models. In this work, we present an analytic framework to visualize and understand GANs at the unit-, object-…
Ulsan National Institute of Science and Technology (UNIST)（蔚山科学技术院）；Korea Advanced Institute of Science and Technology (KAIST)（韩国科学技术院）；INEEJI
Despite significant improvements on the image generation performance of Generative Adversarial Networks (GANs), generations with low visual fidelity still have been observed. As widely used metrics for GANs focus more on the overall performance of the model, evaluation on the quality of individual generations or detection of defective generations is challenging. While recent studies try to detect featuremap units that cause artifacts and evaluate individual samples, these approaches require additional resources such as external networks or a number of training data to approximate the real data…
We revisit the importance of the individual units in Convolutional Neural Networks (CNNs) for visual recognition. By conducting unit ablation experiments on CNNs trained on large scale image datasets, we demonstrate that, though ablating any individual unit does not hurt overall classification accuracy, it does lead to significant damage on the accuracy of specific classes. This result shows that an individual unit is specialized to encode information relevant to a subset of classes. We compute the correlation between the accuracy drop under unit ablation and various attributes of an individua…
School of Behavioral and Brain Sciences（行为与脑科学学院）；The University of Texas at Dallas（得克萨斯大学达拉斯分校）；UMIACS（马里兰大学信息与计算机科学学院）；University of Maryland（马里兰大学）
Deep convolutional neural networks (DCNNs) trained for face identification develop representations that generalize over variable images, while retaining subject (e.g., gender) and image (e.g., viewpoint) information. Identity, gender, and viewpoint codes were studied at the "neural unit" and ensemble levels of a face-identification network. At the unit level, identification, gender classification, and viewpoint estimation were measured by deleting units to create variably-sized, randomly-sampled subspaces at the top network layer. Identification of 3,531 identities remained high (area under th…
Understanding the internal representations of deep neural networks (DNNs) is crucal to explain their behavior. The interpretation of individual units, which are neurons in MLPs or convolution kernels in convolutional networks, has been paid much attention given their fundamental role. However, recent research (Morcos et al. 2018) presented a counterintuitive phenomenon, which suggests that an individual unit with high class selectivity, called interpretable units, has poor contributions to generalization of DNNs. In this work, we provide a new perspective to understand this counterintuitive ph…
Oculus VR；NEC Laboratories America（NEC美国实验室）；Enlitic；University of Michigan, Ann Arbor（密歇根大学安娜堡分校）
Recently, convolutional neural networks (CNNs) have been used as a powerful tool to solve many problems of machine learning and computer vision. In this paper, we aim to provide insight on the property of convolutional neural networks, as well as a generic method to improve the performance of many CNN architectures. Specifically, we first examine existing CNN models and observe an intriguing property that the filters in the lower layers form pairs (i.e., filters with opposite phase). Inspired by our observation, we propose a novel, simple yet effective activation scheme called concatenated ReL…
UC San Diego（加州大学圣地亚哥分校）；UC Santa Cruz（加州大学圣克鲁兹分校）；UC Santa Barbara（加州大学圣巴巴拉分校）
Recent advances in language and vision push forward the research of captioning a single image to describing visual differences between image pairs. Suppose there are two images, I_1 and I_2, and the task is to generate a description W_{1,2} comparing them, existing methods directly model { I_1, I_2 } -> W_{1,2} mapping without the semantic understanding of individuals. In this paper, we introduce a Learning-to-Compare (L2C) model, which learns to understand the semantic structures of these two images and compare them while learning to describe each one. We demonstrate that L2C benefits from a …
Understanding the internal representations of deep neural networks (DNNs) is crucal to explain their behavior. The interpretation of individual units, which are neurons in MLPs or convolution kernels in convolutional networks, has been paid much attention given their fundamental role. However, recent research (Morcos et al. 2018) presented a counterintuitive phenomenon, which suggests that an individual unit with high class selectivity, called interpretable units, has poor contributions to generalization of DNNs. In this work, we provide a new perspective to understand this counterintuitive ph…
Siemens AG（西门子股份公司）；The University of Munich（慕尼黑大学）
A number of backpropagation-based approaches such as DeConvNets, vanilla Gradient Visualization and Guided Backpropagation have been proposed to better understand individual decisions of deep convolutional neural networks. The saliency maps produced by them are proven to be non-discriminative. Recently, the Layer-wise Relevance Propagation (LRP) approach was proposed to explain the classification decisions of rectifier neural networks. In this work, we evaluate the discriminativeness of the generated explanations and analyze the theoretical foundation of LRP, i.e. Deep Taylor Decomposition. Th…
Tübingen AI Center（图宾根人工智能中心）；University of Tübingen（图宾根大学）；Technical University of Darmstadt（达姆施塔特工业大学）
A precise understanding of why units in an artificial network respond to certain stimuli would constitute a big step towards explainable artificial intelligence. One widely used approach towards this goal is to visualize unit responses via activation maximization. These synthetic feature visualizations are purported to provide humans with precise information about the image features that cause a unit to be activated - an advantage over other alternatives like strongly activating natural dataset samples. If humans indeed gain causal insight from visualizations, this should enable them to predic…
Counting Beyond Instances: A Benchmark for Group-Individual Object Counting
School of Computer Science and Technology, Huazhong University of Science and Technology（华中科技大学计算机科学与技术学院）；Shanghai Artificial Intelligence Laboratory（上海人工智能实验室）；University of Liverpool（利物浦大学）
ReUnit: Multi-Granularity Visual Unitization for Long Video Understanding
Shenzhen Campus of Sun Yat-sen University（中山大学深圳校区）；Harbin Institute of Technology, Shenzhen（哈尔滨工业大学（深圳））；Shenzhen Institute for Advanced Study, University of Electronic Science and Technology of China（电子科技大学深圳高等研究院）
As interpretability has been pointed out as the obstacle to the adoption of Deep Neural Networks (DNNs), there is an increasing interest in solving a transparency issue to guarantee the impressive performance. In this paper, we demonstrate the efficiency of recent attribution techniques to explain the diagnostic decision by visualizing the significant factors in the input image. By utilizing the characteristics of objectness that DNNs have learned, fully decomposing the network prediction visualizes clear localization of target lesion. To verify our work, we conduct our experiments on Chest X-…
Technical University of Munich（慕尼黑工业大学）；Amazon（亚马逊）；Huawei Germany（华为德国）；Know-Center GmbH（Know-Center有限公司）
In this work, we investigate the use of three information-theoretic quantities -- entropy, mutual information with the class variable, and a class selectivity measure based on Kullback-Leibler divergence -- to understand and study the behavior of already trained fully-connected feed-forward neural networks. We analyze the connection between these information-theoretic quantities and classification performance on the test set by cumulatively ablating neurons in networks trained on MNIST, FashionMNIST, and CIFAR-10. Our results parallel those recently published by Morcos et al., indicating that …
Most existing scene text detectors focus on detecting characters or words that only capture partial text messages due to missing contextual information. For a better understanding of text in scenes, it is more desired to detect contextual text blocks (CTBs) which consist of one or multiple integral text units (e.g., characters, words, or phrases) in natural reading order and transmit certain complete text messages. This paper presents contextual text detection, a new setup that detects CTBs for better understanding of texts in scenes. We formulate the new setup by a dual detection task which f…
Korea University（高丽大学）；Tel Aviv University（特拉维夫大学）；KAIST（韩国科学技术院）；Facebook AI Research（脸书人工智能研究院）
As Deep Neural Networks (DNNs) have demonstrated superhuman performance in a variety of fields, there is an increasing interest in understanding the complex internal mechanisms of DNNs. In this paper, we propose Relative Attributing Propagation (RAP), which decomposes the output predictions of DNNs with a new perspective of separating the relevant (positive) and irrelevant (negative) attributions according to the relative influence between the layers. The relevance of each neuron is identified with respect to its degree of contribution, separated into positive and negative, while preserving th…
École Polytechnique Fédérale de Lausanne (EPFL)（洛桑联邦理工学院）；Stanford University（斯坦福大学）；IDIAP Research Institute（IDIAP研究所）
We present a unified framework for understanding human social behaviors in raw image sequences. Our model jointly detects multiple individuals, infers their social actions, and estimates the collective actions with a single feed-forward pass through a neural network. We propose a single architecture that does not rely on external detection algorithms but rather is trained end-to-end to generate dense proposal maps that are refined via a novel inference scheme. The temporal consistency is handled via a person-level matching Recurrent Neural Network. The complete model takes as input a sequence …
