<ul><li>A new PyTorch-based library called 'Synaptogen' has been proposed to simulate neural network execution with accurately captured memristor hardware properties.</li><li>The library aims to showcase how a machine learning system with millions of parameters would perform on memristor hardware using a Conformer model trained on TED-LIUMv2 speech recognition task as an example.</li><li>Through simulated analog computation using 3-bit weight precision, the research limits the relative degradation in word error rate to 25% for linear operations executed on memristor hardware.</li></ul>

Running Conventional Automatic Speech Recognition on Memristor Hardware: A Simulated Approach

Discover more