2020/05/16 by Riccardo Massa, Massa, Riccardo, Alberto Marchisio +5 · 2 citations
Engineering · Neuroscience · #Advanced Memory and Neural Computing #Ferroelectric and Negative Capacitance Devices #Neural dynamics and brain function
paper · pdf · doi:10.48550/arxiv.2006.09985
Spiking Neural Networks (SNNs), the third generation NNs, have come under the\nspotlight for machine learning based applications due to their biological\nplausibility and reduced complexity compared to traditional artificial Deep\nNeural Networks (DNNs). These SNNs can be implemented with extreme energy\nefficiency on neuromorphic processors like the Intel Loihi research chip, and\nfed by event-based sensors, such as DVS cameras. However, DNNs with many layers\ncan achieve relatively high accuracy on image classification and recognition\ntasks, as the research on learning rules for SNNs for real-world applications\nis still not mature. The accuracy results for SNNs are typically obtained\neither by converting the trained DNNs into SNNs, or by directly designing and\ntraining SNNs in the spiking domain. Towards the conversion from a DNN to an\nSNN, we perform a comprehensive analysis of such process, specifically designed\nfor Intel Loihi, showing our methodology for the design of an SNN that achieves\nnearly the same accuracy results as its corresponding DNN. Towards the usage of\nthe event-based sensors, we design a pre-processing method, evaluated for the\nDvsGesture dataset, which makes it possible to be used in the DNN domain.\nHence, based on the outcome of the first analysis, we train a DNN for the\npre-processed DvsGesture dataset, and convert it into the spike domain for its\ndeployment on Intel Loihi, which enables real-time gesture recognition. The\nresults show that our SNN achieves 89.64% classification accuracy and occupies\nonly 37 Loihi cores. The source code for generating our experiments is\navailable online at https://github.com/albertomarchisio/EfficientSNN.\n