{"about":{"site":"https://codewithpapers.app","non_affiliation":"Code with Papers and Syntology are not affiliated with, endorsed by, or sponsored by Papers with Code, Meta, or the pwc-archive mirror.","licence":"CC BY-SA 4.0","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","attribution":"https://codewithpapers.app/attribution","modified":"archive material modified by Syntology; see the attribution page"},"url":"/paper/diagnostic-visualization-for-deep-neural","title":"Diagnostic Visualization for Deep Neural Networks Using Stochastic Gradient Langevin Dynamics","arxiv_id":"1812.04604","date":"2018-12-11","proceeding":null,"authors":["Biye Jiang","David M. Chan","Tianhao Zhang","John F. Canny"],"abstract":"The internal states of most deep neural networks are difficult to interpret,\nwhich makes diagnosis and debugging during training challenging. Activation\nmaximization methods are widely used, but lead to multiple optima and are hard\nto interpret (appear noise-like) for complex neurons. Image-based methods use\nmaximally-activating image regions which are easier to interpret, but do not\nprovide pixel-level insight into why the neuron responds to them. In this work\nwe introduce an MCMC method: Langevin Dynamics Activation Maximization (LDAM),\nwhich is designed for diagnostic visualization. LDAM provides two affordances\nin combination: the ability to explore the set of maximally activating\npre-images, and the ability to trade-off interpretability and pixel-level\naccuracy using a GAN-style discriminator as a regularizer. We present case\nstudies on MNIST, CIFAR and ImageNet datasets exploring these trade-offs.\nFinally we show that diagnostic visualization using LDAM leads to a novel\ninsight into the parameter averaging method for deep net training.","url_abs":"http://arxiv.org/abs/1812.04604v1","url_pdf":"http://arxiv.org/pdf/1812.04604v1.pdf","source":{"archive":"pwc-archive (Hugging Face), CC BY-SA 4.0","snapshot":"2025-07-28","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","row_kind":"abstracts"},"code_links":[{"paper_slug":"diagnostic-visualization-for-deep-neural","repo_url":"https://github.com/BIDData/BIDMach","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":0,"framework":"none","reach":null}],"tasks":[{"task_slug":"diagnostic","task_name":"Diagnostic"}],"methods":[{"method_slug":"interpretability","method_name":"Interpretability"}],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":null,"mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}