SimpleShot: Revisiting Nearest-Neighbor Classification for Few-Shot Learning
arXiv:1911.04623
Abstract
Few-shot learners aim to recognize new object classes based on a small number of labeled training examples. To prevent overfitting, state-of-the-art few-shot learners use meta-learning on convolutional-network features and perform classification using a nearest-neighbor classifier. This paper studies the accuracy of nearest-neighbor baselines without meta-learning. Surprisingly, we find simple feature transformations suffice to obtain competitive few-shot learning accuracies. For example, we find that a nearest-neighbor classifier used in combination with mean-subtraction and L2-normalization outperforms prior results in three out of five settings on the miniImageNet dataset.
References in corpus (4)
Cited by in corpus (16)
- Representation Learning for Natural Language Processing
- Multi-level Second-order Few-shot Learning
- Simple and Effective Few-Shot Named Entity Recognition with Structured Nearest Neighbor Learning
- Machine learning with limited data
- Aligning Visual Prototypes with BERT Embeddings for Few-Shot Learning
- Few-shot Intent Classification and Slot Filling with Retrieved Examples
- Transductive Few-Shot Learning: Clustering is All You Need?
- Predicting the Accuracy of a Few-Shot Classifier
- Gaussian Process Meta Few-shot Classifier Learning via Linear Discriminant Laplace Approximation
- Model-Agnostic Graph Regularization for Few-Shot Learning
- Fuzzy Simplicial Networks: A Topology-Inspired Model to Improve Task Generalization in Few-shot Learning
- Learning by Examples Based on Multi-level Optimization
- Reinforced Attention for Few-Shot Learning and Beyond
- DecisiveNets: Training Deep Associative Memories to Solve Complex Machine Learning Problems
- How Sensitive are Meta-Learners to Dataset Imbalance?
- Mutual-Information Based Few-Shot Classification