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

Multi-task learning (MTL) aims to improve the generalization of severalrelated tasks by learning them jointly. As a comparison, in addition to thejoint training scheme, modern meta-learning allows unseen tasks with limitedlabels during the test phase, in the hope of fast adaptation over them. Despitethe subtle difference between MTL and meta-learning in the problem formulation,both learning paradigms share the same insight that the shared structurebetween existing training tasks could lead to better generalization andadaptation. In this paper, we take one important step further to understand theclose connection between these two learning paradigms, through both theoreticalanalysis and empirical investigation. Theoretically, we first demonstrate thatMTL shares the same optimization formulation with a class of gradient-basedmeta-learning (GBML) algorithms. We then prove that for over-parameterizedneural networks with sufficient depth, the learned predictive functions of MTLand GBML are close. In particular, this result implies that the predictionsgiven by these two models are similar over the same unseen task. Empirically,we corroborate our theoretical findings by showing that, with properimplementation, MTL is competitive against state-of-the-art GBML algorithms ona set of few-shot image classification benchmarks. Since existing GBMLalgorithms often involve costly second-order bi-level optimization, ourfirst-order MTL method is an order of magnitude faster on large-scale datasetssuch as mini-ImageNet. We believe this work could help bridge the gap betweenthese two learning paradigms, and provide a computationally efficientalternative to GBML that also supports fast task adaptation.

Bridging Multi-Task Learning and Meta-Learning: Towards Efficient Training and Effective Adaptation