Multikernel Passive Stochastic Gradient Algorithms and Transfer Learning
- 11 May 2021
- journal article
- research article
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Automatic Control
- Vol. 67 (4) , 1792-1805
Abstract
This article develops a novel passive stochastic gradient algorithm. In passive stochastic approximation, the stochastic gradient algorithm does not have control over the location where noisy gradients of the cost function are evaluated. Classical passive stochastic gradient algorithms use a kernel that approximates a Dirac delta to weigh the gradients based on how far they are evaluated from the desired point. In this article, we construct a multikernel passive stochastic gradient algorithm. The algorithm performs substantially better in high dimensional problems and incorporates variance reduction. We analyze the weak convergence of the multikernel algorithm and its rate of convergence. In numerical examples, we study the multikernel version of the passive least mean squares algorithm for transfer learning to compare the performance with the classical passive version.Keywords
All Related Versions
Funding Information
- Army Research Office (W911NF-19-1-0365, W911NF-21-1-0093)
- National Science Foundation (CCF-1714180)
- Air Force Office of Scientific Research (FA9550-18-1-0268)
This publication has 14 references indexed in Scilit:
- Dual-Domain Adaptive Beamformer Under Linearly and Quadratically Constrained Minimum VarianceIEEE Transactions on Signal Processing, 2013
- The Bernstein-Von-Mises theorem under misspecificationElectronic Journal of Statistics, 2012
- A Survey on Transfer LearningIEEE Transactions on Knowledge and Data Engineering, 2009
- Inference in Hidden Markov Models2005
- Passive stochastic approximation with constant step size and window widthIEEE Transactions on Automatic Control, 1996
- Nonparametric sequential estimation of zeros and extrema of regression functionsIEEE Transactions on Information Theory, 1987
- Analysis of constrained LMS algorithm with application to adaptive beamforming using perturbation sequencesIEEE Transactions on Antennas and Propagation, 1986
- How to apply the method of stochastic approximation in the non-parametric estimation of a regression function1Series Statistics, 1977
- On the Bernstein-v. Mises Approximation of Posterior DistributionsThe Annals of Statistics, 1976
- An algorithm for linearly constrained adaptive array processingProceedings of the IEEE, 1972