Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning

Efficient Heterogeneous Meta-Learning via Channel Shuffling Modulation

ICLR 2024 · Label-efficient and robust learning › Zero-/few-shot and meta-learning · approach: Meta-augmentation and meta-objectives

Read the paperCheck your idea against 108,967 papers

AI-extracted summary. Written by AI from the paper's abstract only. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Proposes a channel-shuffling modulation method for efficient meta-learning across heterogeneous, multi-modal task distributions.

Problem

Meta-learning across heterogeneous task distributions, where most existing methods assume homogeneous tasks and thus generalize poorly to new tasks from a multi-modal task distribution using few labeled samples.

Approach

Uses channel shuffling as a modulation mechanism to extract and generalize transferable meta-knowledge from streaming multi-modal task sets; the abstract is truncated so the precise mechanism is not fully stated.

Closest papers by meaning

• Improving Generalization in Meta-learning via Task Augmentation (ICML 2021 · similarity 0.90)
• Task Groupings Regularization: Data-Free Meta-Learning with Heterogeneous Pre-trained Models (ICML 2024 · similarity 0.89)
• Large-Scale Meta-Learning with Continual Trajectory Shifting (ICML 2021 · similarity 0.88)
• A Channel Coding Benchmark for Meta-Learning (NeurIPS 2021 · similarity 0.88)
• Bridging Multi-Task Learning and Meta-Learning: Towards Efficient Training and Effective Adaptation (ICML 2021 · similarity 0.88)
• The Advantage of Conditional Meta-Learning for Biased Regularization and Fine Tuning (NeurIPS 2020 · similarity 0.88)