First-order non-convex Riemannian optimization algorithms have gained recent popularity in structured machine learning problems including principal component analysis and low-rank matrix completion. The current paper presents an efficient Riemannian Stochastic Path Integrated Differential EstimatoR (R-SPIDER) algorithm to solve the finite-sum and online Riemannian non-convex minimization problems. At the core of R-SPIDER is a recursive semi-stochastic gradient estimator that can accurately estimate Riemannian gradient under not only exponential mapping and parallel transport, but also general retraction and vector transport operations. Compared with prior Riemannian algorithms, such a recursive gradient estimation mechanism endows R-SPIDER with lower computational cost in first-order oracle complexity. Specifically, for finite-sum problems with n components, R-SPIDER is proved to converge to an ϵ-approximate stationary point within [Formula: see text] stochastic gradient evaluations, beating the best-known complexity [Formula: see text]; for online optimization, R-SPIDER is shown to converge with [Formula: see text] complexity which is, to the best of our knowledge, the first non-asymptotic result for online Riemannian optimization. For the special case of gradient dominated functions, we further develop a variant of R-SPIDER with improved linear rate of convergence. Extensive experimental results demonstrate the advantage of the proposed algorithms over the state-of-the-art Riemannian non-convex optimization methods.

Download full-text PDF

Source
http://dx.doi.org/10.1109/TPAMI.2019.2933841DOI Listing

Publication Analysis

Top Keywords

[formula text]
12
non-convex optimization
8
riemannian
8
riemannian optimization
8
online riemannian
8
riemannian non-convex
8
r-spider
6
optimization
5
gradient
5
faster first-order
4

Similar Publications

The accumulation of aging cells significantly contributes to chronic inflammatory diseases such as atherosclerosis. Human carotid artery single-cell sequencing has shown that large numbers of aging foam cells are present in the plaques of human patients. Berberine (BBR) has been shown to inhibit cell senescence, however, the mechanisms involved in its treatment of atherosclerotic senescence have not yet been determined.

View Article and Find Full Text PDF

We investigate the growth of amorphous MoSi thin films using magnetron co-sputtering and optimize the growth conditions with respect to crystal structure and superconducting properties (e.g., critical temperature [Formula: see text]).

View Article and Find Full Text PDF

Improved printed circuit board defect detection scheme.

Sci Rep

January 2025

School of Computer Engineering , Jiangsu Second Normal University, Nanjing, Jiangsu, 211200, China.

In this paper, an improved printed circuit board(PCB)defect detection scheme named PD-YOLOv8 is proposed, which is specialized in the common and challenging problem of small target recognition in PCB inspection. This improved scheme mainly relies on the basic framework of YOLOv8n, and effectively enhances the detection performance of PCB small defects through multiple innovative designs. First, we incorporate the Efficient Channel Attention Network (ECANet) attention mechanism into the backbone network of YOLOv8, which improves the performance of small-target detection by adaptively enhancing the expressiveness of key features, so that the network possesses higher sensitivity and focus on tiny details in PCB images.

View Article and Find Full Text PDF

The problem of protein structure determination is usually solved by X-ray crystallography. Several in silico deep learning methods have been developed to overcome the high attrition rate, cost of experiments and extensive trial-and-error settings, for predicting the crystallization propensities of proteins based on their sequences. In this work, we benchmark the power of open protein language models (PLMs) through the TRILL platform, a be-spoke framework democratizing the usage of PLMs for the task of predicting crystallization propensities of proteins.

View Article and Find Full Text PDF

We propose a general approach to quasi-deform the Korteweg-De Vries (KdV) equation by deforming its Hamiltonian. The standard abelianization process based on the inherent sl(2) loop algebra leads to an infinite number of anomalous conservation laws, that yield conserved charges for definite space-time parity of the solution. Judicious choice of the deformed Hamiltonian yields an integrable system with scaled parameters as well as a hierarchy of deformed systems, some of which possibly are quasi-integrable.

View Article and Find Full Text PDF

Want AI Summaries of new PubMed Abstracts delivered to your In-box?

Enter search terms and have AI summaries delivered each week - change queries or unsubscribe any time!