In this paper, we propose a methodology to incrementally compute a convolutional layer of a neural network based on events, and an architecture that can efficiently implement it. In order to illustrate the approach, we present an application example, where we train a traditional DNN based on a LeNet architecture using a traffic signs dataset and then, we replace the first convolutional layer with our event based approach.
|Name||2021 Argentine Conference on Electronics - Congreso Argentino de Electronica 2021, CAE 2021|
|Konferenz||2021 Argentine Conference on Electronics (CAE)|
|Zeitraum||11/03/21 → 12/03/21|