MDFM: Multi-Decision Fusing Model for Few-Shot Learning

In recent years, researchers pay growing attention to the few-shot learning (FSL) task address data-scarce problem. A standard FSL framework is composed of two components: i) Pre-train. Employ base data generate a CNN-based feature extraction model (FEM). ii) Meta-test. Apply trained FEM novel (category different from data) acquire embeddings and recognize them. Although have made remarkable breakthroughs in FSL, there still exists fundamental Since with usually cannot adapt class flawlessly, data’s may lead distribution shift To this challenge, we hypothesize that even if most decisions based on FEMs are viewed as weak decisions , which not available for all classes, they perform decent some specific categories. Inspired by assumption, propose method Multi-Decision Fusing Model (MDFM), comprehensively considers multiple enhance efficacy robustness model. MDFM simple, flexible, non-parametric can directly apply existing FEMs. Besides, extend proposed settings (e.g., supervised semi-supervised settings). We evaluate five benchmark datasets achieve significant improvements 3.4%-7.3% compared state-of-the-arts.