Method for Hybrid Precision Convolutional Neural Network Representation
This addresses the problem of optimizing CNN implementations for hardware efficiency in integrated circuits, but it appears incremental as it builds on existing quantization techniques.
The paper tackles the trade-off between precision and accuracy in fixed-point CNN representations for integrated circuits, proposing a hybrid precision method to improve power efficiency and throughput while maintaining accuracy.
This invention addresses fixed-point representations of convolutional neural networks (CNN) in integrated circuits. When quantizing a CNN for a practical implementation there is a trade-off between the precision used for operations between coefficients and data and the accuracy of the system. A homogenous representation may not be sufficient to achieve the best level of performance at a reasonable cost in implementation complexity or power consumption. Parsimonious ways of representing data and coefficients are needed to improve power efficiency and throughput while maintaining accuracy of a CNN.