We held our next tinyML Talks webcast. Martino Sorbaro from SynSense has presented Always-on visual classification below 1 mW with spiking convolutional networks on Dynap™-CNN on February 2, 2021.
Neuromorphic hardware enables real-world applications in computer vision, audition and other sensory modalities to be deployed with very low power consumption (<1 mW). Commercial neuromorphic solutions are beginning to emerge, based on inference in spiking neural networks. In these systems, computation is performed using asynchronous 1-bit binary signals, and sensory input is processed in real-time. In this talk we will present our approach to training spiking convolutional neural networks for practical applications, and showcase our results on real-world data, presenting our novel Dynap™-CNN convolutional neuromorphic chip. We will illustrate the pipeline from data collection to training, simulation and on-chip classification of visual scenes at an average power consumption below 1 mW.
Martino Sorbaro is a research and development scientist at SynSense AG, Zürich, Switzerland, and a postdoc at the institute of Neuroinformatics of the University of Zürich and ETH. He obtained a MSc in physics at the university of Pavia, Italy, and a PhD in neuroinformatics at the university of Edinburgh, Scotland. His current work focuses on learning in spiking neural networks, both for theoretical research and technological applications.
Watch on YouTube:
Download presentation slide:
Feel free to ask your questions on this thread and keep the conversation going!