A Closer Look at Few-shot Classification

Abstract

Few-shot classification aims to learn a classifier to recognize unseen classes during training with limited labeled examples. While significant progress has been made, the growing complexity of network designs, meta-learning algorithms, and differences in implementation details make a fair comparison difficult. In this paper, we present 1) a consistent comparative analysis of several representative few-shot classification algorithms, with results showing that deeper backbones significantly reduce the performance differences among methods on datasets with limited domain differences, 2) a modified baseline method that surprisingly achieves competitive performance when compared with the state-of-the-art on both the \miniI and the CUB datasets, and 3) a new experimental setting for evaluating the cross-domain generalization ability for few-shot classification algorithms. Our results reveal that reducing intra-class variation is an important factor when the feature backbone is shallow, but not as critical when using deeper backbones. In a realistic cross-domain evaluation setting, we show that a baseline method with a standard fine-tuning practice compares favorably against other state-of-the-art few-shot learning algorithms.

ICLR 2019. Code: https://github.com/wyharveychen/CloserLookFewShot . Project: https://sites.google.com/view/a-closer-look-at-few-shot/

References in corpus (7)

• Model-Agnostic Meta-Learning for Fast Adaptation of Deep Networks
• Prototypical Networks for Few-shot Learning
• Data Augmentation Generative Adversarial Networks
• EMNIST: an extension of MNIST to handwritten letters
• Few-Shot Learning with Graph Neural Networks
• Few-Shot Learning with Metric-Agnostic Conditional Embeddings
• Learning to cluster in order to transfer across domains and tasks

Cited by in corpus (63)

• Frustratingly Simple Few-Shot Object Detection
• Federated Meta-Learning with Fast Convergence and Efficient Communication
• Out-of-Distribution Generalization via Risk Extrapolation (REx)
• Tip-Adapter: Training-free CLIP-Adapter for Better Vision-Language Modeling
• Rapid Learning or Feature Reuse? Towards Understanding the Effectiveness of MAML
• Empirical Bayes Transductive Meta-Learning with Synthetic Gradients
• Semi-supervised Domain Adaptation via Minimax Entropy
• BOIL: Towards Representation Change for Few-shot Learning
• Self-Augmentation: Generalizing Deep Networks to Unseen Classes for Few-Shot Learning
• FewCLUE: A Chinese Few-shot Learning Evaluation Benchmark
• Data Augmentation for Meta-Learning
• Scalable Transfer Learning with Expert Models
• A Baseline for Few-Shot Image Classification
• Tackling the Unannotated: Scene Graph Generation with Bias-Reduced Models
• Cross-Domain Few-Shot Learning with Meta Fine-Tuning
• Learning to Learn Variational Semantic Memory
• Adaptive Deep Kernel Learning
• Unraveling Meta-Learning: Understanding Feature Representations for Few-Shot Tasks
• Multi-scale Adaptive Task Attention Network for Few-Shot Learning
• Auxiliary Task Reweighting for Minimum-data Learning
• Are Few-Shot Learning Benchmarks too Simple ? Solving them without Task Supervision at Test-Time
• Dynamic Distillation Network for Cross-Domain Few-Shot Recognition with Unlabeled Data
• Cross-domain few-shot learning with unlabelled data
• Graph-based Interpolation of Feature Vectors for Accurate Few-Shot Classification
• Is Support Set Diversity Necessary for Meta-Learning?
• Meta-learning algorithms for Few-Shot Computer Vision
• Dizygotic Conditional Variational AutoEncoder for Multi-Modal and Partial Modality Absent Few-Shot Learning
• Revisiting Classical Bagging with Modern Transfer Learning for On-the-fly Disaster Damage Detector
• Non-Gaussian Gaussian Processes for Few-Shot Regression
• Image-Level or Object-Level? A Tale of Two Resampling Strategies for Long-Tailed Detection
• Machine learning with limited data
• Object-aware Long-short-range Spatial Alignment for Few-Shot Fine-Grained Image Classification
• Hierarchical Representation based Query-Specific Prototypical Network for Few-Shot Image Classification
• SB-MTL: Score-based Meta Transfer-Learning for Cross-Domain Few-Shot Learning
• FLAT: Few-Shot Learning via Autoencoding Transformation Regularizers
• Orthogonal Over-Parameterized Training
• MetaSelector: Meta-Learning for Recommendation with User-Level Adaptive Model Selection
• ACDC: Weight Sharing in Atom-Coefficient Decomposed Convolution
• A Foliated View of Transfer Learning
• AFD-Net: Adaptive Fully-Dual Network for Few-Shot Object Detection
• Crossover Learning for Fast Online Video Instance Segmentation
• How Does the Task Landscape Affect MAML Performance?
• Model-Agnostic Learning to Meta-Learn
• Addressing the Real-world Class Imbalance Problem in Dermatology
• ST-MAML: A Stochastic-Task based Method for Task-Heterogeneous Meta-Learning
• Multi-source Few-shot Domain Adaptation
• Few-shot Continual Learning: a Brain-inspired Approach
• One Representation to Rule Them All: Identifying Out-of-Support Examples in Few-shot Learning with Generic Representations
• Task Affinity with Maximum Bipartite Matching in Few-Shot Learning
• Continual Mixed-Language Pre-Training for Extremely Low-Resource Neural Machine Translation
• FDNAS: Improving Data Privacy and Model Diversity in AutoML
• Semi Supervised Learning For Few-shot Audio Classification By Episodic Triplet Mining
• Analysing Wideband Absorbance Immittance in Normal and Ears with Otitis Media with Effusion Using Machine Learning
• Fuzzy Simplicial Networks: A Topology-Inspired Model to Improve Task Generalization in Few-shot Learning
• Discrete Few-Shot Learning for Pan Privacy
• Improving the Generalization of Meta-learning on Unseen Domains via Adversarial Shift
• Domain Agnostic Few-Shot Learning For Document Intelligence
• Learning Clusterable Visual Features for Zero-Shot Recognition
• Meta Learning as Bayes Risk Minimization
• Calibrating Class Activation Maps for Long-Tailed Visual Recognition
• CLTA: Contents and Length-based Temporal Attention for Few-shot Action Recognition
• Semi-Supervised Domain Adaptation via Selective Pseudo Labeling and Progressive Self-Training
• On Data Efficiency of Meta-learning