Adaptive Compositional Continual Meta-Learning

Bin Wu, Jinyuan Fang, Xiangxiang Zeng, Shangsong Liang, Qiang Zhang
Proceedings of the 40th International Conference on Machine Learning, PMLR 202:37358-37378, 2023.

Abstract

This paper focuses on continual meta-learning, where few-shot tasks are heterogeneous and sequentially available. Recent works use a mixture model for meta-knowledge to deal with the heterogeneity. However, these methods suffer from parameter inefficiency caused by two reasons: (1) the underlying assumption of mutual exclusiveness among mixture components hinders sharing meta-knowledge across heterogeneous tasks. (2) they only allow increasing mixture components and cannot adaptively filter out redundant components. In this paper, we propose an Adaptive Compositional Continual Meta-Learning (ACML) algorithm, which employs a compositional premise to associate a task with a subset of mixture components, allowing meta-knowledge sharing among heterogeneous tasks. Moreover, to adaptively adjust the number of mixture components, we propose a component sparsification method based on evidential theory to filter out redundant components. Experimental results show ACML outperforms strong baselines, showing the effectiveness of our compositional meta-knowledge, and confirming that ACML can adaptively learn meta-knowledge.

Cite this Paper

BibTeX
@InProceedings{pmlr-v202-wu23d, title = {Adaptive Compositional Continual Meta-Learning}, author = {Wu, Bin and Fang, Jinyuan and Zeng, Xiangxiang and Liang, Shangsong and Zhang, Qiang}, booktitle = {Proceedings of the 40th International Conference on Machine Learning}, pages = {37358--37378}, year = {2023}, editor = {Krause, Andreas and Brunskill, Emma and Cho, Kyunghyun and Engelhardt, Barbara and Sabato, Sivan and Scarlett, Jonathan}, volume = {202}, series = {Proceedings of Machine Learning Research}, month = {23--29 Jul}, publisher = {PMLR}, pdf = {https://proceedings.mlr.press/v202/wu23d/wu23d.pdf}, url = {https://proceedings.mlr.press/v202/wu23d.html}, abstract = {This paper focuses on continual meta-learning, where few-shot tasks are heterogeneous and sequentially available. Recent works use a mixture model for meta-knowledge to deal with the heterogeneity. However, these methods suffer from parameter inefficiency caused by two reasons: (1) the underlying assumption of mutual exclusiveness among mixture components hinders sharing meta-knowledge across heterogeneous tasks. (2) they only allow increasing mixture components and cannot adaptively filter out redundant components. In this paper, we propose an Adaptive Compositional Continual Meta-Learning (ACML) algorithm, which employs a compositional premise to associate a task with a subset of mixture components, allowing meta-knowledge sharing among heterogeneous tasks. Moreover, to adaptively adjust the number of mixture components, we propose a component sparsification method based on evidential theory to filter out redundant components. Experimental results show ACML outperforms strong baselines, showing the effectiveness of our compositional meta-knowledge, and confirming that ACML can adaptively learn meta-knowledge.} }
Endnote
%0 Conference Paper %T Adaptive Compositional Continual Meta-Learning %A Bin Wu %A Jinyuan Fang %A Xiangxiang Zeng %A Shangsong Liang %A Qiang Zhang %B Proceedings of the 40th International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2023 %E Andreas Krause %E Emma Brunskill %E Kyunghyun Cho %E Barbara Engelhardt %E Sivan Sabato %E Jonathan Scarlett %F pmlr-v202-wu23d %I PMLR %P 37358--37378 %U https://proceedings.mlr.press/v202/wu23d.html %V 202 %X This paper focuses on continual meta-learning, where few-shot tasks are heterogeneous and sequentially available. Recent works use a mixture model for meta-knowledge to deal with the heterogeneity. However, these methods suffer from parameter inefficiency caused by two reasons: (1) the underlying assumption of mutual exclusiveness among mixture components hinders sharing meta-knowledge across heterogeneous tasks. (2) they only allow increasing mixture components and cannot adaptively filter out redundant components. In this paper, we propose an Adaptive Compositional Continual Meta-Learning (ACML) algorithm, which employs a compositional premise to associate a task with a subset of mixture components, allowing meta-knowledge sharing among heterogeneous tasks. Moreover, to adaptively adjust the number of mixture components, we propose a component sparsification method based on evidential theory to filter out redundant components. Experimental results show ACML outperforms strong baselines, showing the effectiveness of our compositional meta-knowledge, and confirming that ACML can adaptively learn meta-knowledge.
APA
Wu, B., Fang, J., Zeng, X., Liang, S. & Zhang, Q.. (2023). Adaptive Compositional Continual Meta-Learning. Proceedings of the 40th International Conference on Machine Learning, in Proceedings of Machine Learning Research 202:37358-37378 Available from https://proceedings.mlr.press/v202/wu23d.html.

Related Material