We gratefully acknowledge support from
the Simons Foundation and member institutions.
Full-text links:


Current browse context:


Change to browse by:

References & Citations


(what is this?)
CiteULike logo BibSonomy logo Mendeley logo del.icio.us logo Digg logo Reddit logo ScienceWISE logo

Statistics > Machine Learning

Title: A Generative Model for Sampling High-Performance and Diverse Weights for Neural Networks

Abstract: Recent work on mode connectivity in the loss landscape of deep neural networks has demonstrated that the locus of (sub-)optimal weight vectors lies on continuous paths. In this work, we train a neural network that serves as a hypernetwork, mapping a latent vector into high-performance (low-loss) weight vectors, generalizing recent findings of mode connectivity to higher dimensional manifolds. We formulate the training objective as a compromise between accuracy and diversity, where the diversity takes into account trivial symmetry transformations of the target network. We demonstrate how to reduce the number of parameters in the hypernetwork by parameter sharing. Once learned, the hypernetwork allows for a computationally efficient, ancestral sampling of neural network weights, which we recruit to form large ensembles. The improvement in classification accuracy obtained by this ensembling indicates that the generated manifold extends in dimensions other than directions implied by trivial symmetries. For computational efficiency, we distill an ensemble into a single classifier while retaining generalization.
Comments: arXiv admin note: substantial text overlap with arXiv:1801.01952
Subjects: Machine Learning (stat.ML); Machine Learning (cs.LG)
Cite as: arXiv:1905.02898 [stat.ML]
  (or arXiv:1905.02898v1 [stat.ML] for this version)

Submission history

From: Lior Deutsch [view email]
[v1] Tue, 7 May 2019 04:28:46 GMT (9060kb,D)

Link back to: arXiv, form interface, contact.