Research Talk (ITC'2020) FAT Fault Aware Training for Reliable Inference under Hardware Faults

Video presentation for the International Test Conference (ITC) of a technique developed in Xilinx Research Labs called FAT. In this video we present the general concepts and the main results of Fault-Aware Training (FAT), which includes error modeling during neural network (NN) training, to make convolutional neural networks (CNNs) resilient to specific fault models during inference on the hardware. Experiments show that by injecting faults in the convolutional layers during training, highly accurate CNNs can be
trained which exhibits much better error tolerance compared to the original.