Information Theoretical Study on Neural Learning Processes
Primary Area: learning theory
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.
Keywords: Neural Networks, Information Flow, Explainability
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2024/AuthorGuide.
Abstract: In this paper, we reveal how deep networks converge over data flows in their training process via an information theoretical analysis. By looking into individual neurons in different layers, we found that the whole learning process of neural networks are pertinent to the local flows per individual neurons in various layers.
Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors' identity.
No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.
Submission Number: 2640
Loading