all AI news
Feature Learning and Generalization in Deep Networks with Orthogonal Weights
June 13, 2024, 4:49 a.m. | Hannah Day, Yonatan Kahn, Daniel A. Roberts
stat.ML updates on arXiv.org arxiv.org
Abstract: Fully-connected deep neural networks with weights initialized from independent Gaussian distributions can be tuned to criticality, which prevents the exponential growth or decay of signals propagating through the network. However, such networks still exhibit fluctuations that grow linearly with the depth of the network, which may impair the training of networks with width comparable to depth. We show analytically that rectangular networks with tanh activations and weights initialized from the ensemble of orthogonal matrices have …
abstract arxiv cs.lg feature growth hep-ph hep-th however independent network networks neural networks replace stat.ml through type
More from arxiv.org / stat.ML updates on arXiv.org
Proximal Interacting Particle Langevin Algorithms
3 days, 8 hours ago |
arxiv.org
Cluster Quilting: Spectral Clustering for Patchwork Learning
3 days, 8 hours ago |
arxiv.org
Jobs in AI, ML, Big Data
Senior Data Engineer
@ Displate | Warsaw
Content Designer
@ Glean | Palo Alto, CA
IT&D Data Solution Architect
@ Reckitt | Hyderabad, Telangana, IN, N/A
Python Developer
@ Riskinsight Consulting | Hyderabad, Telangana, India
Technical Lead (Java/Node.js)
@ LivePerson | Hyderabad, Telangana, India (Remote)
Backend Engineer - Senior and Mid-Level - Sydney Hybrid or AU remote
@ Displayr | Sydney, New South Wales, Australia