Meta-Learning for Semi-Supervised Few-Shot Classification

Abstract

In few-shot classification, we are interested in learning algorithms that train a classifier from only a handful of labeled examples. Recent progress in few-shot classification has featured meta-learning, in which a parameterized model for a learning algorithm is defined and trained on episodes representing different classification problems, each with a small labeled training set and its corresponding test set. In this work, we advance this few-shot classification paradigm towards a scenario where unlabeled examples are also available within each episode. We consider two situations: one where all unlabeled examples are assumed to belong to the same set of classes as the labeled examples of the episode, as well as the more challenging situation where examples from other distractor classes are also provided. To address this paradigm, we propose novel extensions of Prototypical Networks (Snell et al., 2017) that are augmented with the ability to use unlabeled examples when producing prototypes. These models are trained in an end-to-end way on episodes, to learn to leverage the unlabeled examples successfully. We evaluate these methods on versions of the Omniglot and miniImageNet benchmarks, adapted to this new framework augmented with unlabeled examples. We also propose a new split of ImageNet, consisting of a large set of classes, with a hierarchical structure. Our experiments confirm that our Prototypical Networks can learn to improve their predictions due to unlabeled examples, much like a semi-supervised algorithm would.

Published as a conference paper at ICLR 2018. 15 pages

References in corpus (6)

• Adam: A Method for Stochastic Optimization
• Model-Agnostic Meta-Learning for Fast Adaptation of Deep Networks
• Prototypical Networks for Few-shot Learning
• One-shot Learning with Memory-Augmented Neural Networks
• Using Fast Weights to Attend to the Recent Past
• Learning Algorithms for Active Learning

Cited by in corpus (84)

• Generalizing from a Few Examples: A Survey on Few-Shot Learning
• TADAM: Task dependent adaptive metric for improved few-shot learning
• An Ensemble of Epoch-wise Empirical Bayes for Few-shot Learning
• Meta-Learning Update Rules for Unsupervised Representation Learning
• Dual Adaptive Representation Alignment for Cross-domain Few-shot Learning
• Multi-level Second-order Few-shot Learning
• Self-Augmentation: Generalizing Deep Networks to Unseen Classes for Few-Shot Learning
• Enhancing Few-Shot Image Classification through Learnable Multi-Scale Embedding and Attention Mechanisms
• MHFC: Multi-Head Feature Collaboration for Few-Shot Learning
• Learning from Few Samples: A Survey
• Sparse Spatial Transformers for Few-Shot Learning
• Leveraging the Feature Distribution in Transfer-based Few-Shot Learning
• Weight-Sharing Neural Architecture Search: A Battle to Shrink the Optimization Gap
• Defining Benchmarks for Continual Few-Shot Learning
• GCT: Graph Co-Training for Semi-Supervised Few-Shot Learning
• A Closer Look at Few-Shot 3D Point Cloud Classification
• DPGN: Distribution Propagation Graph Network for Few-shot Learning
• RP2K: A Large-Scale Retail Product Dataset for Fine-Grained Image Classification
• FS-BAN: Born-Again Networks for Domain Generalization Few-Shot Classification
• Adversarial Feature Hallucination Networks for Few-Shot Learning
• Lifelong Adaptive Machine Learning for Sensor-based Human Activity Recognition Using Prototypical Networks
• A Baseline for Few-Shot Image Classification
• Diversity Helps: Unsupervised Few-shot Learning via Distribution Shift-based Data Augmentation
• MetaSAug: Meta Semantic Augmentation for Long-Tailed Visual Recognition
• A Broader Study of Cross-Domain Few-Shot Learning
• Towards Cross-Granularity Few-Shot Learning: Coarse-to-Fine Pseudo-Labeling with Visual-Semantic Meta-Embedding
• Unsupervised Few-shot Learning via Self-supervised Training
• ReFine: Re-randomization before Fine-tuning for Cross-domain Few-shot Learning
• Towards Gradient-based Bilevel Optimization with Non-convex Followers and Beyond
• Deep Metric Transfer for Label Propagation with Limited Annotated Data
• Selecting Relevant Features from a Multi-domain Representation for Few-shot Classification
• p-Meta: Towards On-device Deep Model Adaptation
• How Important is the Train-Validation Split in Meta-Learning?
• Dynamic Distillation Network for Cross-Domain Few-Shot Recognition with Unlabeled Data
• Shaping Visual Representations with Attributes for Few-Shot Recognition
• Instance Credibility Inference for Few-Shot Learning
• Expert Training: Task Hardness Aware Meta-Learning for Few-Shot Classification
• ProtoTransformer: A Meta-Learning Approach to Providing Student Feedback
• Learning to Impute: A General Framework for Semi-supervised Learning
• Learning from Adversarial Features for Few-Shot Classification
• A comprehensive and easy-to-use multi-domain multi-task medical imaging meta-dataset
• Attentive Graph Neural Networks for Few-Shot Learning
• Looking back to lower-level information in few-shot learning
• Domain Generalization via Semi-supervised Meta Learning
• Hyperbolic Busemann Learning with Ideal Prototypes
• Adaptive Task Sampling for Meta-Learning
• MetaDelta: A Meta-Learning System for Few-shot Image Classification
• Graceful Degradation and Related Fields
• Select-ProtoNet: Learning to Select for Few-Shot Disease Subtype Prediction
• OnlineAugment: Online Data Augmentation with Less Domain Knowledge
• An empirical study of pretrained representations for few-shot classification
• Few-Shot Semantic Segmentation Augmented with Image-Level Weak Annotations
• Contextualizing Enhances Gradient Based Meta Learning
• Are Fewer Labels Possible for Few-shot Learning?
• Ontology-based n-ball Concept Embeddings Informing Few-shot Image Classification
• Aligning Visual Prototypes with BERT Embeddings for Few-Shot Learning
• ST-MAML: A Stochastic-Task based Method for Task-Heterogeneous Meta-Learning
• High-order structure preserving graph neural network for few-shot learning
• Automating Chapter-Level Classification for Electronic Theses and Dissertations
• Asymmetric Distribution Measure for Few-shot Learning
• Population-Based Evolution Optimizes a Meta-Learning Objective
• Few-shot Continual Learning: a Brain-inspired Approach
• Long-term Cross Adversarial Training: A Robust Meta-learning Method for Few-shot Classification Tasks
• Few Shot Activity Recognition Using Variational Inference
• How to trust unlabeled data? Instance Credibility Inference for Few-Shot Learning
• Learning to generate classifiers
• Meta-Learning with Network Pruning
• Medical Image Segmentation with Limited Supervision: A Review of Deep Network Models
• Improving traffic sign recognition by active search
• Uncertainty-Aware Few-Shot Image Classification
• Fine-Grained Few Shot Learning with Foreground Object Transformation
• Semi Supervised Learning For Few-shot Audio Classification By Episodic Triplet Mining
• Task Affinity with Maximum Bipartite Matching in Few-Shot Learning
• Subjectivity Learning Theory towards Artificial General Intelligence
• Memory-Augmented Relation Network for Few-Shot Learning
• Semi-supervised Learning with a Teacher-student Network for Generalized Attribute Prediction
• Cooperative Bi-path Metric for Few-shot Learning
• Augmented Bi-path Network for Few-shot Learning
• Edge-Labeling based Directed Gated Graph Network for Few-shot Learning
• Complex Momentum for Optimization in Games
• One-Class Meta-Learning: Towards Generalizable Few-Shot Open-Set Classification
• Training few-shot classification via the perspective of minibatch and pretraining
• Learning to Generalize to Unseen Tasks with Bilevel Optimization
• Unlocking the Full Potential of Small Data with Diverse Supervision