FRINK Linked Data Fragments server

Matches in SemOpenAlex for { <https://semopenalex.org/work/W3191738967> ?p ?o ?g. }

• W3191738967 abstract "Multi-task learning (MTL) aims to improve the generalization of several related tasks by learning them jointly. As a comparison, in addition to the joint training scheme, modern meta-learning allows unseen tasks with limited labels during the test phase, in the hope of fast adaptation over them. Despite the subtle difference between MTL and meta-learning in the problem formulation, both learning paradigms share the same insight that the shared structure between existing training tasks could lead to better generalization and adaptation. In this paper, we take one important step further to understand the close connection between these two learning paradigms, through both theoretical analysis and empirical investigation. Theoretically, we first demonstrate that MTL shares the same optimization formulation with a class of gradient-based meta-learning (GBML) algorithms. We then prove that for over-parameterized neural networks with sufficient depth, the learned predictive functions of MTL and GBML are close. In particular, this result implies that the predictions given by these two models are similar over the same unseen task. Empirically, we corroborate our theoretical findings by showing that, with proper implementation, MTL is competitive against state-of-the-art GBML algorithms on a set of few-shot image classification benchmarks. Since existing GBML algorithms often involve costly second-order bi-level optimization, our first-order MTL method is an order of magnitude faster on large-scale datasets such as mini-ImageNet. We believe this work could help bridge the gap between these two learning paradigms, and provide a computationally efficient alternative to GBML that also supports fast task adaptation." @default.
• W3191738967 created "2021-08-16" @default.
• W3191738967 creator A5026062552 @default.
• W3191738967 creator A5026419076 @default.
• W3191738967 creator A5042417097 @default.
• W3191738967 date "2021-06-16" @default.
• W3191738967 modified "2023-09-27" @default.
• W3191738967 title "Bridging Multi-Task Learning and Meta-Learning: Towards Efficient Training and Effective Adaptation" @default.
• W3191738967 cites W1542791059 @default.
• W3191738967 cites W1621791442 @default.
• W3191738967 cites W1896424170 @default.
• W3191738967 cites W2101234009 @default.
• W3191738967 cites W2104849640 @default.
• W3191738967 cites W2108598243 @default.
• W3191738967 cites W2165644552 @default.
• W3191738967 cites W2194775991 @default.
• W3191738967 cites W2237537322 @default.
• W3191738967 cites W2251743902 @default.
• W3191738967 cites W2601450892 @default.
• W3191738967 cites W2624871570 @default.
• W3191738967 cites W2753709519 @default.
• W3191738967 cites W2794363191 @default.
• W3191738967 cites W2795900505 @default.
• W3191738967 cites W2809090039 @default.
• W3191738967 cites W2910655610 @default.
• W3191738967 cites W2913340405 @default.
• W3191738967 cites W2925446857 @default.
• W3191738967 cites W2926246612 @default.
• W3191738967 cites W2952116677 @default.
• W3191738967 cites W2952689429 @default.
• W3191738967 cites W2962698540 @default.
• W3191738967 cites W2963303956 @default.
• W3191738967 cites W2963310665 @default.
• W3191738967 cites W2963341924 @default.
• W3191738967 cites W2963547174 @default.
• W3191738967 cites W2963677766 @default.
• W3191738967 cites W2964052793 @default.
• W3191738967 cites W2964098911 @default.
• W3191738967 cites W2964121744 @default.
• W3191738967 cites W2964121793 @default.
• W3191738967 cites W2964206659 @default.
• W3191738967 cites W2965435631 @default.
• W3191738967 cites W2970217468 @default.
• W3191738967 cites W2970389519 @default.
• W3191738967 cites W2970697704 @default.
• W3191738967 cites W2970814018 @default.
• W3191738967 cites W2970971581 @default.
• W3191738967 cites W2971043187 @default.
• W3191738967 cites W2994689640 @default.
• W3191738967 cites W2994747787 @default.
• W3191738967 cites W2995049146 @default.
• W3191738967 cites W2995443162 @default.
• W3191738967 cites W2998314487 @default.
• W3191738967 cites W3006505669 @default.
• W3191738967 cites W3007891689 @default.
• W3191738967 cites W3008360155 @default.
• W3191738967 cites W3008372792 @default.
• W3191738967 cites W3015585292 @default.
• W3191738967 cites W3015606043 @default.
• W3191738967 cites W3035151691 @default.
• W3191738967 cites W3035990676 @default.
• W3191738967 cites W3037059960 @default.
• W3191738967 cites W3042103612 @default.
• W3191738967 cites W3080894165 @default.
• W3191738967 cites W3108975329 @default.
• W3191738967 cites W3118062200 @default.
• W3191738967 cites W3118608800 @default.
• W3191738967 cites W3120402410 @default.
• W3191738967 cites W3121902971 @default.
• W3191738967 cites W3157430111 @default.
• W3191738967 cites W3169366835 @default.
• W3191738967 doi "https://doi.org/10.48550/arxiv.2106.09017" @default.
• W3191738967 hasPublicationYear "2021" @default.
• W3191738967 type Work @default.
• W3191738967 sameAs 3191738967 @default.
• W3191738967 citedByCount "0" @default.
• W3191738967 crossrefType "posted-content" @default.
• W3191738967 hasAuthorship W3191738967A5026062552 @default.
• W3191738967 hasAuthorship W3191738967A5026419076 @default.
• W3191738967 hasAuthorship W3191738967A5042417097 @default.
• W3191738967 hasBestOaLocation W31917389671 @default.
• W3191738967 hasConcept C11413529 @default.
• W3191738967 hasConcept C119857082 @default.
• W3191738967 hasConcept C120665830 @default.
• W3191738967 hasConcept C121332964 @default.
• W3191738967 hasConcept C134306372 @default.
• W3191738967 hasConcept C139807058 @default.
• W3191738967 hasConcept C154945302 @default.
• W3191738967 hasConcept C162324750 @default.
• W3191738967 hasConcept C165464430 @default.
• W3191738967 hasConcept C174348530 @default.
• W3191738967 hasConcept C177148314 @default.
• W3191738967 hasConcept C177264268 @default.
• W3191738967 hasConcept C187736073 @default.
• W3191738967 hasConcept C199360897 @default.
• W3191738967 hasConcept C2780451532 @default.
• W3191738967 hasConcept C2781002164 @default.
• W3191738967 hasConcept C28006648 @default.
• W3191738967 hasConcept C31258907 @default.
• W3191738967 hasConcept C33923547 @default.

• next