MetaNODE: Prototype Optimization as a Neural ODE for Few-Shot Learning

Zhang, Baoquan; Li, Xutao; Feng, Shouhua; Ye, Yunming; Rui, Yong

doi:10.48550/arxiv.2103.14341

Cited by 1 publication

(1 citation statement)

References 0 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The meta-learning based approaches focus on learning a task-agnostic meta-learner by constructing a large number of few-shot tasks from base classes, and then leverage the meta-learner to quickly learn/infer an FSL classifier for recognizing novel classes. The meta-learner can be a good initial model [13,32], optimization algorithm [3,53], embedding network [21,36,55], metric strategy [23,45,56], or label propagation strategy [31,34,37,50], etc. The representation learning-based approaches aim to design a good feature extractor [47,52] or training strategy [8,11,25,38] to learn transferable representations from abundant base classes, so that the novel class samples can be recognized by a simple cosine classifier [9] or logistic regression classifier [30].…”

Section: Few-shot Learningmentioning

confidence: 99%

MetaDT: Meta Decision Tree for Interpretable Few-Shot Learning

Zhang¹,

Jiang²,

Xutao³

et al. 2022

Preprint

Self Cite

View full text Add to dashboard Cite

Few-Shot Learning (FSL) is a challenging task, which aims to recognize novel classes with few examples. Recently, lots of methods have been proposed from the perspective of meta-learning and representation learning for improving FSL performance. However, few works focus on the interpretability of FSL decision process. In this paper, we take a step towards the interpretable FSL by proposing a novel decision tree-based meta-learning framework, namely, MetaDT. Our insight is replacing the last black-box FSL classifier of the existing representation learning methods by an interpretable decision tree with meta-learning.The key challenge is how to effectively learn the decision tree (i.e., the tree structure and the parameters of each node) in the FSL setting. To address the challenge, we introduce a tree-like class hierarchy as our prior: 1) the hierarchy is directly employed as the tree structure; 2) by regarding the class hierarchy as an undirected graph, a graph convolution-based decision tree inference network is designed as our meta-learner to learn to infer the parameters of each node. At last, a two-loop optimization mechanism is incorporated into our framework for a fast adaptation of the decision tree with few examples. Extensive experiments on performance comparison and interpretability analysis show the effectiveness and superiority of our MetaDT. Our code will be publicly available upon acceptance.

show abstract