Hebbian Fast Weights enhance Vision Transformers for few-shot character recognition

Researchers have developed a new approach to few-shot character recognition by integrating Hebbian Fast-Weight (HFW) modules into Vision Transformer architectures. This method aims to mimic biological neural systems' ability to form transient associative memories during inference, unlike standard transformers that rely on fixed representations. When applied to a Swin-Tiny model, this strategy achieved a 96.2% accuracy in 5-way 1-shot classification and 99.2% in 5-way 5-shot classification on the Omniglot benchmark, slightly outperforming its non-Hebbian counterpart. AI

Summary written by gemini-2.5-flash-lite from 1 source. How we write summaries →

IMPACT Introduces a novel method for few-shot learning that could improve model adaptability in low-data scenarios.

RANK_REASON This is a research paper presenting a novel method for few-shot learning in computer vision. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.CV →

• Hebbian Fast-Weight
• Vision Transformer
• Swin-Tiny
• Omniglot
• ViT-Small
• DeiT-Small

• paper
• other