N-Omniglot: a Large-scale Neuromorphic Dataset for Spatio-Temporal Sparse Few-shot Learning

Few-shot learning (learning with a few samples) is one of the most important capacities of the human brain. However, the current artificial intelligence systems meet difficulties in achieving this ability, so as the biologically plausible spiking neural networks (SNNs). Datasets for traditional few-shot learning domains provide few amounts of temporal information. And the absence of the neuromorphic datasets has hindered the development of few-shot learning for SNNs. Here, we provide the first neuromorphic dataset: N-Omniglot, using the Dynamic Vision Sensor (DVS). It contains 1623 categories of handwritten characters, with only 20 samples per class. N-Omniglot eliminates the need for a neuromorphic dataset for SNNs with high spareness and tremendous temporal coherence. Additionally, the dataset provides a powerful challenge and a suitable benchmark for developing SNNs algorithm in the few-shot learning domain due to the chronological information of strokes. We also provide the improved nearest neighbor, convolutional network, SiameseNet, and meta-learning algorithm in spiking version for verification.