Prototypical Networks for Few-shot Learning
By Jake Snell, Kevin Swersky, Richard S. Zemel. 2017
10,206 citations recorded by the source corpus.
We propose Prototypical Networks for the problem of few-shot classification, where a classifier must generalize to new classes not seen in the training set, given only a small number of examples of each new class. Prototypical Networks learn a metric space in which classification can be performed by computing distances to prototype representations of each class. Compared to recent approaches for few-shot learning, they reflect a simpler inductive bias that is beneficial in this limited-data regime, and achieve excellent results. We provide an analysis showing that some simple design decisions can yield substantial improvements over recent approaches involving complicated architectural choices and meta-learning. We further extend Prototypical Networks to zero-shot learning and achieve state-of-the-art results on the CU-Birds dataset.
Subfields: Artificial Intelligence, Representation & Self-Supervised Learning
How researchers see this paper
The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.
In Artificial Intelligence
- Sergey Levine
- Yoshua Bengio
- Mohit Bansal
- Quoc V. Le
- Jianfeng Gao
- Trevor Darrell
- Dacheng Tao
- Percy Liang
In Representation & Self-Supervised Learning
- Trevor Darrell
- Yoshua Bengio
- Dacheng Tao
- Han-Jia Ye
- Andrea Vedaldi
- De-chuan Zhan
- Ishan Misra
- Andrew Zisserman
Citations in the corpus
- Meta-Learning for Low-Resource Neural Machine Translation (2018, background): “…of meta-learning: 1. learning a meta-policy for updating model parameters (see, e.g., Andrychowicz et al., 2016; Ha et al., 2016a; Mishra et al., 2017) 2. learning a good parameter initialization for fast adaptation (see, e.g., Finn et al., 2017; Vinyals et al., 2016; Snell et al., 2017).”
- Model-Agnostic Meta-Learning for Fast Adaptation of Deep Networks (2017, background): “Siamese networks (Koch, 2015) or recurrence with attention mechanisms (Vinyals et al., 2016; Shyam et al., 2017; Snell et al., 2017).”
- Do Better ImageNet Models Transfer Better? (2018, methodology): “This problem is closely related to few-shot learning [39, 71, 55, 62, 21, 62, 49], but these methods are typically evaluated with training and test classes from the same distribution.”
- Big Transfer (BiT): General Visual Representation Learning (2019, background): “tain good performance when faced with novel classes, and that use only a few examples per class. Meta-learning or metric-learning techniques have been proposed to learn with few or no labels, such as [48,39,41]. However, recent papers have shown that a simple linear classier on top of pre-trained re”
- Rethinking Few-Shot Image Classification: a Good Embedding Is All You Need? (2020, methodology): “ith 3 runs, where in each run the accuracy is the mean accuracy CIFAR-FS 5-way FC100 5-way model backbone 1-shot 5-shot 1-shot 5-shot MAML [12] 32-32-32-32 58.9 1.9 71.5 1.0 - - Prototypical Networks [46] 64-64-64-64 55.5 0.7 72.0 0.6 35.3 0.6 48.6 0.6 Relation Networks [48] 64-96-128-256 55.0 1.0 6”
- Shaping Visual Representations with Language for Few-Shot Classification (2019, positive): “[13] use METEOR scores between captions as a similarity metric for specializing embeddings for image retrieval, but do not directly [26], we propose few-shot classification models whose learned representations are constrained to predict natural language descriptions of the task during training, in…”
- Unsupervised Feature Learning via Non-parametric Instance Discrimination (2018, background): “Metric learning has been shown to be effective for few-shot learning [38, 41, 37].”
- Pre-train, Prompt, and Predict: A Systematic Survey of Prompting Methods in Natural Language Processing (2021, methodology): “These consist of either full-data learning, where a reasonably large number of training examples are used to train the model, or few-shot learning [126] where a very small number of examples are used to train the model.”
- Repurposing Pretrained Models for Robust Out-of-domain Few-Shot Learning (2021, methodology): “First, metric-based approaches solve few-shot tasks by training a feature extractor that maximizes inter-class similarity and intra-class dissimilarity (Vinyals et al., 2016; Snell et al., 2017; Sung et al., 2018).”
- Scaling Laws for the Few-Shot Adaptation of Pre-trained Image Classifiers (2021, methodology): “Few-shot performance is measured via the fine-tuning of the classification layer, all other weights remaining frozen, as well as the matching network (Vinyals et al., 2016) and prototypical network (Snell et al., 2017) method.”
- Flamingo: a Visual Language Model for Few-Shot Learning (2022, methodology): “In this paper, we take inspiration from the in-context [11] few-shot learning technique instead of more involved few-shot learning approaches based on metric learning [24, 103, 112, 117] or meta-learning [6, 7, 27, 31, 91, 155].”
- ORBIT: A Real-World Few-Shot Dataset for Teachable Object Recognition (2021, methodology): “In Figure 4 however, we show that ProtoNets’ personalization is not consistent across users, for some going as low as 25% in frame accuracy (for other metrics/models see Figure A.10).”
- CrossTransformers: spatially-aware few-shot transfer (2020, methodology): “Matching-based methods [28, 75, 78, 89] instead learn representations for similarity functions [11, 16, 17, 44, 82], in the hope that the similarities will generalize to new data.”
- Fast and Flexible Multi-Task Classification Using Conditional Neural Adaptive Processes (2019, methodology): “One approach adapts only the top, or head, of the classifier leaving the feature extractor fixed [4, 5].”
- Meta-Learning Probabilistic Inference for Prediction (2018, methodology): “, 2017; Ravi and Larochelle, 2017), metric based meta-learning (Snell et al., 2017), amortized MAP inference (Qiao et al.”
- Meta-learning with differentiable closed-form solvers (2018, methodology): “99, which within this binary setup works significantly better than originally described in [10].”
- Meta-Dataset: A Dataset of Datasets for Learning to Learn from Few Examples (2019, methodology): “As explained in Snell et al. (2017), Prototypical Networks can be re-interpreted as a linear classifier applied to a learned representation g ( x ) .”