Pith. sign in

REVIEW 2 cited by

Efficient Dictionary Learning with Gradient Descent

Not yet reviewed by Pith; the record is open.

This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.

SPECIMEN: schema-true, not a live event

T0 review · schema-true

One-sentence machine reading of the paper's core claim.

pith:XXXXXXXX · record.json · timestamp

arxiv 1809.10313 v1 pith:QFCGWGGB submitted 2018-09-27 math.OC

classification math.OC
keywords descentgradientlearningnonconvexobjectivepointsproblemsconvergence
verification ladder T0 review T1 audit T2 compute T3 formal

Signed reviews

No signed human review yet.

0 comments
read the original abstract

Randomly initialized first-order optimization algorithms are the method of choice for solving many high-dimensional nonconvex problems in machine learning, yet general theoretical guarantees cannot rule out convergence to critical points of poor objective value. For some highly structured nonconvex problems however, the success of gradient descent can be understood by studying the geometry of the objective. We study one such problem -- complete orthogonal dictionary learning, and provide converge guarantees for randomly initialized gradient descent to the neighborhood of a global optimum. The resulting rates scale as low order polynomials in the dimension even though the objective possesses an exponential number of saddle points. This efficient convergence can be viewed as a consequence of negative curvature normal to the stable manifolds associated with saddle points, and we provide evidence that this feature is shared by other nonconvex problems of importance as well.

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 2 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. A Nonconvex Approach for Exact and Efficient Multichannel Sparse Blind Deconvolution

    eess.SP 2019-08 conditional novelty 6.0 of 10

    For multichannel sparse blind deconvolution, Huber-loss Riemannian gradient descent with random initialization plus an LP-rounding step provably recovers the kernel and sparse signals up to a signed shift, with sample...

  2. Short-and-Sparse Deconvolution -- A Geometric Approach

    eess.SP 2019-08 conditional novelty 5.0 of 10

    A practical alternating descent algorithm with data-driven initialization, momentum, homotopy continuation, and reweighting solves short-and-sparse blind deconvolution on synthetic and real imaging and neuroscience da...

Pith tools