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

Download:

Current browse context:

cs.LG

Change to browse by:

References & Citations

DBLP - CS Bibliography

Bookmark

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

Computer Science > Machine Learning

Title: Saddlepoints in Unsupervised Least Squares

Authors: Samuel Gerber
Abstract: This paper sheds light on the risk landscape of unsupervised least squares in the context of deep auto-encoding neural nets. We formally establish an equivalence between unsupervised least squares and principal manifolds. This link provides insight into the risk landscape of auto--encoding under the mean squared error, in particular all non-trivial critical points are saddlepoints. Finding saddlepoints is in itself difficult, overcomplete auto-encoding poses the additional challenge that the saddlepoints are degenerate. Within this context we discuss regularization of auto-encoders, in particular bottleneck, denoising and contraction auto-encoding and propose a new optimization strategy that can be framed as particular form of contractive regularization.
Subjects: Machine Learning (cs.LG); Machine Learning (stat.ML)
Cite as: arXiv:2104.05000 [cs.LG]
  (or arXiv:2104.05000v1 [cs.LG] for this version)

Submission history

From: Samuel Gerber [view email]
[v1] Sun, 11 Apr 2021 12:15:36 GMT (921kb,D)

Link back to: arXiv, form interface, contact.