Laplacian Regularized Few-Shot Learning

28 Jun 2020  ·  Imtiaz Masud Ziko, Jose Dolz, Eric Granger, Ismail Ben Ayed ·

We propose a transductive Laplacian-regularized inference for few-shot tasks. Given any feature embedding learned from the base classes, we minimize a quadratic binary-assignment function containing two terms: (1) a unary term assigning query samples to the nearest class prototype, and (2) a pairwise Laplacian term encouraging nearby query samples to have consistent label assignments. Our transductive inference does not re-train the base model, and can be viewed as a graph clustering of the query set, subject to supervision constraints from the support set. We derive a computationally efficient bound optimizer of a relaxation of our function, which computes independent (parallel) updates for each query sample, while guaranteeing convergence. Following a simple cross-entropy training on the base classes, and without complex meta-learning strategies, we conducted comprehensive experiments over five few-shot learning benchmarks. Our LaplacianShot consistently outperforms state-of-the-art methods by significant margins across different models, settings, and data sets. Furthermore, our transductive inference is very fast, with computational times that are close to inductive inference, and can be used for large-scale few-shot tasks.

PDF Abstract

Datasets


Task Dataset Model Metric Name Metric Value Global Rank Result Benchmark
Few-Shot Image Classification Dirichlet CUB-200 (5-way, 1-shot) Laplacian-Shot 1:1 Accuracy 73.7 # 3
Few-Shot Image Classification Dirichlet CUB-200 (5-way, 5-shot) Laplacian-Shot 1:1 Accuracy 87.7 # 2
Few-Shot Image Classification Dirichlet Mini-Imagenet (5-way, 1-shot) Laplacian-Shot 1:1 Accuracy 65.4 # 3
Few-Shot Image Classification Dirichlet Mini-Imagenet (5-way, 5-shot) Laplacian-Shot 1:1 Accuracy 81.6 # 2
Few-Shot Image Classification Dirichlet Tiered-Imagenet (5-way, 1-shot) Laplacian-Shot 1:1 Accuracy 72.3 # 4
Few-Shot Image Classification Dirichlet Tiered-Imagenet (5-way, 5-shot) Laplacian-Shot 1:1 Accuracy 85.7 # 2

Methods


No methods listed for this paper. Add relevant methods here