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: Learning Rates for Multi-task Regularization Networks

Abstract: Multi-task learning is an important trend of machine learning in facing the era of artificial intelligence and big data. Despite a large amount of researches on learning rate estimates of various single-task machine learning algorithms, there is little parallel work for multi-task learning. We present mathematical analysis on the learning rate estimate of multi-task learning based on the theory of vector-valued reproducing kernel Hilbert spaces and matrix-valued reproducing kernels. For the typical multi-task regularization networks, an explicit learning rate dependent both on the number of sample data and the number of tasks is obtained. It reveals that the generalization ability of multi-task learning algorithms is indeed affected as the number of tasks increases.
Subjects: Machine Learning (cs.LG); Functional Analysis (math.FA)
Cite as: arXiv:2104.00453 [cs.LG]
  (or arXiv:2104.00453v2 [cs.LG] for this version)

Submission history

From: Haizhang Zhang [view email]
[v1] Thu, 1 Apr 2021 13:10:29 GMT (15kb)
[v2] Tue, 20 Apr 2021 10:39:04 GMT (15kb)

Link back to: arXiv, form interface, contact.