eduzhai > Applied Sciences > Engineering >

In-Memory Resistive RAM Implementation of Binarized Neural Networks for Medical Applications

  • Save

... pages left unread,continue reading

Document pages: 6 pages

Abstract: The advent of deep learning has considerably accelerated machine learningdevelopment. The deployment of deep neural networks at the edge is howeverlimited by their high memory and energy consumption requirements. With newmemory technology available, emerging Binarized Neural Networks (BNNs) arepromising to reduce the energy impact of the forthcoming machine learninghardware generation, enabling machine learning on the edge devices and avoidingdata transfer over the network. In this work, after presenting ourimplementation employing a hybrid CMOS - hafnium oxide resistive memorytechnology, we suggest strategies to apply BNNs to biomedical signals such aselectrocardiography and electroencephalography, keeping accuracy level andreducing memory requirements. We investigate the memory-accuracy trade-off whenbinarizing whole network and binarizing solely the classifier part. We alsodiscuss how these results translate to the edge-oriented Mobilenet~V1 neuralnetwork on the Imagenet task. The final goal of this research is to enablesmart autonomous healthcare devices.

Please select stars to rate!


0 comments Sign in to leave a comment.

    Data loading, please wait...