Understanding the role of individual units in a deep neural network
Autor: | Bolei Zhou, Jun-Yan Zhu, Antonio Torralba, David Bau, Hendrik Strobelt, Agata Lapedriza |
---|---|
Rok vydání: | 2020 |
Předmět: |
I.2
FOS: Computer and information sciences Computer Science - Machine Learning I.4 Computer science Computer Vision and Pattern Recognition (cs.CV) Computer Science - Computer Vision and Pattern Recognition Context (language use) 02 engineering and technology Image editing 010501 environmental sciences computer.software_genre Machine learning Semantics 01 natural sciences Convolutional neural network Machine Learning (cs.LG) 0202 electrical engineering electronic engineering information engineering Neural and Evolutionary Computing (cs.NE) Set (psychology) Colloquium on the Science of Deep Learning 0105 earth and related environmental sciences Multidisciplinary Artificial neural network Contextual image classification business.industry Computer Science - Neural and Evolutionary Computing Object (computer science) 68T07 020201 artificial intelligence & image processing Artificial intelligence business computer |
Zdroj: | Proc Natl Acad Sci U S A |
ISSN: | 1091-6490 0027-8424 |
Popis: | 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. Comment: Proceedings of the National Academy of Sciences 2020. Code at https://github.com/davidbau/dissect/ and website at https://dissect.csail.mit.edu/ |
Databáze: | OpenAIRE |
Externí odkaz: |