AMMD: Attentive Maximum Mean Discrepancy for Few-Shot Image Classification
Keywords: few-shot learning; metric learning; meta-learning
TL;DR: We propose an Attentive Maximum Mean Discrepancy (AMMD) metric for few-shot image classification.
Abstract: Metric-based methods have attained promising performance for the few-shot classification of images. Maximum Mean Discrepancy (MMD) is a typical distance between distributions, requiring to compute expectations w.r.t. data distributions. In this paper, we propose Attentive Maximum Mean Discrepancy (AMMD) to assist MMD with distributions adaptively estimated by an attentive distribution generation module. Based on AMMD, the few-shot learning is modeled as the AMMD metric learning problem. In implementation, we incorporate the part-based feature representation for modeling the AMMD between images. By meta-learning technique, the attentive distribution generation module of AMMD can be learned to generate feature distributions for computing MMD between images, with higher probability mass on the more discriminative features. In the meta-test phase, each query image is labeled as the support class with minimal AMMD to the query image. Extensive experiments show that our AMMD achieves competitive or state-of-the-art performance on few-shot classification benchmark datasets of miniImageNet, tieredImageNet, CIFAR-FS, and FC100.
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.
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2024/AuthorGuide.
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: 567
Loading