Prototypical Networks for Few-shot Learning

NeurIPS 2017  ยท  Jake Snell, Kevin Swersky, Richard S. Zemel ยท

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.

PDF Abstract NeurIPS 2017 PDF NeurIPS 2017 Abstract
Task Dataset Model Metric Name Metric Value Global Rank Result Benchmark
Few-Shot Image Classification CUB 200 50-way (0-shot) Prototypical Networks Accuracy 54.6 # 1
Few-Shot Image Classification Dirichlet Mini-Imagenet (5-way, 1-shot) ProtoNet 1:1 Accuracy 53.6 # 10
Few-Shot Image Classification Dirichlet Mini-Imagenet (5-way, 5-shot) ProtoNet 1:1 Accuracy 74.2 # 8
Few-Shot Image Classification Meta-Dataset Prototypical Networks Accuracy 60.573 # 15
Few-Shot Image Classification Meta-Dataset Rank Prototypical Networks Mean Rank 8.5 # 8
Few-Shot Image Classification Mini-Imagenet 10-way (1-shot) Prototypical Networks (Higher Way) Accuracy 34.6 # 9
Few-Shot Image Classification Mini-Imagenet 10-way (1-shot) Prototypical Networks Accuracy 32.9 # 10
Few-Shot Image Classification Mini-Imagenet 10-way (5-shot) Prototypical Networks Accuracy 49.3 # 9
Few-Shot Image Classification Mini-Imagenet 10-way (5-shot) Prototypical Networks (Higher Way) Accuracy 50.1 # 8
Few-Shot Image Classification Mini-Imagenet 5-way (10-shot) Prototypical Networks Accuracy 74.3 # 5
Few-Shot Image Classification Mini-Imagenet 5-way (1-shot) Prototypical Networks Accuracy 49.42 # 94
Few-Shot Image Classification Mini-Imagenet 5-way (5-shot) Prototypical Networks Accuracy 68.20 # 79
Few-Shot Image Classification OMNIGLOT - 1-Shot, 20-way Prototypical Networks Accuracy 96% # 12
Few-Shot Image Classification OMNIGLOT - 1-Shot, 5-way Prototypical Networks Accuracy 98.8 # 9
Few-Shot Image Classification OMNIGLOT - 5-Shot, 20-way Prototypical Networks Accuracy 98.9% # 10
Few-Shot Image Classification OMNIGLOT - 5-Shot, 5-way Prototypical Networks Accuracy 99.7 # 9
Few-Shot Image Classification Stanford Cars 5-way (1-shot) Prototypical Nets++ Accuracy 40.90 # 5
Few-Shot Image Classification Stanford Cars 5-way (5-shot) Prototypical Nets++ Accuracy 52.93 # 5
Few-Shot Image Classification Tiered ImageNet 10-way (1-shot) Prototypical Networks (Higher Way) Accuracy 38.6 # 7
Few-Shot Image Classification Tiered ImageNet 10-way (1-shot) Prototypical Networks Accuracy 37.3 # 8
Few-Shot Image Classification Tiered ImageNet 10-way (5-shot) Prototypical Networks (Higher Way) Accuracy 58.3 # 6
Few-Shot Image Classification Tiered ImageNet 10-way (5-shot) Prototypical Networks Accuracy 57.8 # 9

Results from Other Papers


Task Dataset Model Metric Name Metric Value Rank Source Paper Compare
Few-Shot Image Classification Mini-ImageNet-CUB 5-way (1-shot) ProtoNet (Snell et al., 2017) Accuracy 45.31 # 6
Few-Shot Image Classification Stanford Dogs 5-way (5-shot) Prototypical Nets++ Accuracy 48.19 # 5
Image Classification Tiered ImageNet 5-way (5-shot) Prototypical Net Accuracy 69.57 # 6

Methods


No methods listed for this paper. Add relevant methods here