Before e i a hardware sings of an ANN, a number format (fixed, floating point etc.) must be considered for the inputs, weights and activation function function. And also the precision (number of bits) should be considered. The Incee of the precision of the design elements of the greatly ly s the resources used. Accuracy has a great impact in the learning phase; So the precision of the numbers must be high as a tha e e infy. However inthe propagation phase, lower precisions will are are . The resulting errors will be small enough to be neglected in in classification applications s1, 3, 4. In the XOR problem we applieded, the input space is is -1 and 1. The training resulted in weights - 2 and 2. We chose 8-bit precision for the system to cover the (-2,2) range, we chose in a precision of 1/64. Table 1 show al-various numbers in this this range and its 8-bit representation. To represent negative numbers, 2's complement method is used. ...
正在翻译中..