Lecture Description:

We have optimized the size of the machine learning inference engine by reducing the intra-engine connectivity, removing the need for floating-point data, and replacing the multiply-accumulate
operation with just accumulation. The resulting FPGA implementation is a good choice for batterypowered IoT endpoints. We present size and performance results on an anomaly detection/predictive maintenance IIoT use case for FPGA as well as MCU.