English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

Exponential generalization bounds with near-tight rates have recently beenestablished for uniformly stable learning algorithms. The notion of uniformstability, however, is stringent in the sense that it is invariant to thedata-generating distribution. Under the weaker and distribution dependentnotions of stability such as hypothesis stability and $L_2$-stability, theliterature suggests that only polynomial generalization bounds are possible ingeneral cases. The present paper addresses this long standing tension betweenthese two regimes of results and makes progress towards relaxing it inside aclassic framework of confidence-boosting. To this end, we first establish anin-expectation first moment generalization error bound for potentiallyrandomized learning algorithms with $L_2$-stability, based on which we thenshow that a properly designed subbagging process leads to near-tightexponential generalization bounds over the randomness of both data andalgorithm. We further substantialize these generic results to stochasticgradient descent (SGD) to derive improved high-probability generalizationbounds for convex or non-convex optimization problems with natural timedecaying learning rates, which have not been possible to prove with theexisting hypothesis stability or uniform stability based results.

Boosting the Confidence of Generalization for $L_2$-Stable Randomized Learning Algorithms