Classifying neuromorphic data using a deep learning framework for image\n classification

In the field of artificial intelligence, neuromorphic computing has been\naround for several decades. Deep learning has however made much recent progress\nsuch that it consistently outperforms neuromorphic learning algorithms in\nclassification tasks in terms of accuracy. Specifically in the field of image\nclassification, neuromorphic computing has been traditionally using either the\ntemporal or rate code for encoding static images in datasets into spike trains.\nIt is only till recently, that neuromorphic vision sensors are widely used by\nthe neuromorphic research community, and provides an alternative to such\nencoding methods. Since then, several neuromorphic datasets as obtained by\napplying such sensors on image datasets (e.g. the neuromorphic CALTECH 101)\nhave been introduced. These data are encoded in spike trains and hence seem\nideal for benchmarking of neuromorphic learning algorithms. Specifically, we\ntrain a deep learning framework used for image classification on the CALTECH\n101 and a collapsed version of the neuromorphic CALTECH 101 datasets. We\nobtained an accuracy of 91.66% and 78.01% for the CALTECH 101 and neuromorphic\nCALTECH 101 datasets respectively. For CALTECH 101, our accuracy is close to\nthe best reported accuracy, while for neuromorphic CALTECH 101, it outperforms\nthe last best reported accuracy by over 10%. This raises the question of the\nsuitability of such datasets as benchmarks for neuromorphic learning\nalgorithms.\n