Meta-Tasks: Improving Robustness in Few-Shot Classification with Unsupervised and Semi-Supervised Learning
Primary Area: transfer learning, meta learning, and lifelong learning
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.
Keywords: few-shot classification, meta-learning, machine learning, semi-supervised learning, unsupervised learning.
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2024/AuthorGuide.
TL;DR: Using tasks as regularization for few shot learning
Abstract: Few-shot learning (FSL) is a challenging problem in machine learning due to the limited availability of labeled data. A major obstacle to FSL is the ability to generalize well on both novel tasks and training tasks. In this paper, we propose a new branch of unsupervised and semi-supervised regularization tasks to combat this problem. Our approach leverages both labeled and unlabelled data to improve the robustness and generalization performance of FSL models. Experimental results demonstrate the effectiveness of our proposed method by showing faster and better convergence, lower generalization, and standard deviation error both on novel tasks and training tasks, highlighting its potential for practical applications in FSL. Our proposed approach offers a promising solution to address the challenge of regularization in FSL, paving the way for future research in this area.
Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors' identity.
No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.
Submission Number: 1413
Loading