Abstract
Metric-learning-based methods, which attempt to learn a deep embedding space on extremely large episodes, have been successfully applied to few-shot classification problems. In this paper, we propose the adoption of large margin nearest center (LMNC) loss during episodic training to enhance metric-learning-based few-shot classification methods. Loss functions (such as cross-entropy and mean square error) commonly used in episodic training strive to achieve the strict goal that differently labeled examples in the embedding space are separated by an infinite distance. However, the learned embedding space cannot guarantee that this goal will be achieved for every episode sampled from a large number of classes. Instead of an infinite distance, LMNC loss requires only that differently labeled examples be separated by a large margin, which can well relax the strict constraint of the traditional loss functions, easily leading to a discriminative embedding space. Moreover, considering the multilevel similarity between various classes, we alleviate the constraint of a fixed large margin and extend LMNC loss to weighted LMNC (WLMNC) loss, which can effectively take advantage of interclass information, achieving a more separable embedding space with adaptive interclass margins. Experiments on state-of-the-art benchmarks demonstrate that the adoption of LMNC and WLMNC losses can strongly improve the embedding learning performance and classification accuracy of metric-based few-shot classification methods for various few-shot scenarios. In particular, LMNC and WLMNC losses can obtain 1.86% and 2.46% gains in prototypical network on miniImageNet for 5-way 1-shot scenario, respectively.
Highlights
Benefiting from massive data consumption, deep learning has made great progress in various computer vision tasks [1], [2] but has struggled in tackling problems where only limited data are available
To reach the full potential of metric learning methods, this paper proposes the adoption of large margin nearest center (LMNC) loss [15], [26] in episodic training, which requires differently labeled examples from the query set and support set to be separated by a large margin instead of a maximum distance
We start with a brief explanation of episodic training and a summary of two benchmarks: the prototypical network [19] (P-net) and the relation network [20] (Rnet) followed by descriptions of LMNC and weighted LMNC (WLMNC) lossaugmented few-shot classification methods
Summary
Benefiting from massive data consumption, deep learning has made great progress in various computer vision tasks [1], [2] but has struggled in tackling problems where only limited data are available. The significant gap between human and deep learning prompts the strong research interests in few-shot learning [3], which attempts to emulate the ability of humans to learn new concepts from limited supervised information. Recent advances in few-shot learning resort to two types of approaches: meta-learning-based methods and metriclearning-based methods to avoid the danger of overfitting on few labeled examples The former type is trained in an auxiliary meta-learning [4] phase based on optimization methods [5]–[9] and modeling methods [10]–[14], where task-generic information is learned from various few-shot tasks as prior knowledge to contend with the target task. The latter type adopts a relatively simple architecture and a distance function to learn a discriminant embedding space to transfer knowledge
Sign-in/Register to view highlights & summary Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a callDisclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.
