Finding Task-Relevant Features for Few-Shot Learning by Category Traversal

Abstract

Few-shot learning is an important area of research. Conceptually, humans are readily able to understand new concepts given just a few examples, while in more pragmatic terms, limited-example training situations are common in practice. Recent effective approaches to few-shot learning employ a metric-learning framework to learn a feature similarity comparison between a query (test) example, and the few support (training) examples. However, these approaches treat each support class independently from one another, never looking at the entire task as a whole. Because of this, they are constrained to use a single set of features for all possible test-time tasks, which hinders the ability to distinguish the most relevant dimensions for the task at hand. In this work, we introduce a Category Traversal Module that can be inserted as a plug-and-play module into most metric-learning based few-shot learners. This component traverses across the entire support set at once, identifying task-relevant features based on both intra-class commonality and inter-class uniqueness in the feature space. Incorporating our module improves performance considerably (5%-10% relative) over baseline systems on both mini-ImageNet and tieredImageNet benchmarks, with overall performance competitive with recent state-of-the-art systems.

Moving Indoor: Unsupervised Video Depth Learning in Challenging Environments

“…Supervised Depth Estimation Estimating the depth from a single image has been long studied. Recently due to the success of deep learning [27,28], many networks for depth estimation have been proposed [11,10,29,31,13,24]. Eigen et al [11] applied multi-scale networks which first estimate coarse depth by a coarse-scale network and refine it by another network.…”

Revisiting metric learning for few-shot image classification

“…Notably, the prototypical networks [2], siamese networks [31], and relation net [33] all adopt the episodebased training strategy, where each episode is designed to mimic few-shot learning. More recently, Li et al [46] proposed Figure 1. The framework of deep K-tuplet Network for few-shot learning.…”

Robust Compare Network for Few-Shot Learning

“…The goal of FSL [11]- [13] is to simulate the learning mechanism of the human brain, which can learn novel concepts from very few labeled samples. Earlier works on FSL tended to explore a variational Bayesian method on the basis of prior knowledge [11] or introduce a generative model to simulate human learning [27].…”

“…FSL is actually a weakly supervised learning task [13], which aims to achieve the image classification with a data set D = {D train , D support , D test }. Formally, given a labeled dataset D train with a large amount of images in each class, the goal of FSL is to learn concepts in novel classes D new = {D support , D test } with a few samples in each class.…”

“…Few-shot learning (FSL) [11]- [13] attempts to simulate the learning mechanism of the human brain, which can learn novel concepts from very few labeled samples. Due to the lack of sufficient labeled training samples, FSL is often formulated as transfer learning problems [14], [15] or meta-learning problems [16], [17].…”