DISCRETE DATA REPRESENTATION-SUPPORTING APPARATUS AND METHOD FOR BACK-TRAINING OF ARTIFICIAL NEURAL NETWORK

摘要

The present disclosure provides a device configured to perform reverse training of an artificial neural network supporting discrete data representation. The device includes an instruction caching unit, a controller unit, a data access unit, an interconnecting module, a primary operating module, a plurality of secondary operating modules, a discrete data operating module, and a converting module. The reverse training of the multilayer artificial neural network may be achieved by means of using the device. The device is characterized by support for discrete data, including storage, operation of the discrete data and conversion of the successive data into the discrete data. The data, such as weights and neurons, in the reverse training of the artificial neural network performed by the device may be discretely or successively represented. The discrete data representation refers to a storage manner of replacing data with specified numbers. For example, four numbers, 00, 01, 02, and 03, may represent four numbers, -1, -1/8, 1/8, and 1, respectively. This storage manner differs from that of using 00/01/10/11 in the decimal system to represent four numbers 0/1/2/3. The discrete data operating module replaces basic operations of the successive data, for example a multiplication operation and an addition operation, with different bitwise operations, for example an exclusive-OR operation and a NOT operation, according to the values of the discrete data. The converting module converts the successive data into the discrete data.