Abstract: Systems and methods are disclosed for selecting parameters for use by a system. The parameters can describe a behavior of the system, which can be represented by a model having an input and an output. The model can include an operation representable by a matrix. The parameters can include the input and output ranges of the operation, the dimensions of the matrix, a noise value for the system, an overflow probability, a regularization parameter, and a desired number of accurate digits. A design environment can be configured to determine values or ranges of values for one or more of the parameters based on values or ranges of values of the remaining parameters. In some embodiments, the design environment can select, recommend, or validate a choice of datatype, minimum system noise, or the dimensions of the matrix. The model can be used to generate code, which can be used to configure the system to perform the operation.

Abstract: Systems and methods quantize an application having a trained Deep Neural Network (DNN) for deployment on target hardware. The application may be instrumented to observe data values generated during execution of the application. Statistics may be generated for the observed data values and presented in a visualization tool. The application may be quantized through a rules based approach. The quantization may be based on the statistics and on constraints imposed by resources available at the target hardware. The systems and methods may present the proposed data types resulting from the quantization and may create a quantized version of the application incorporating the proposed data types. The systems and methods may generate performance data to validate the quantized version of the application. Changes to the rules may be made and the quantization process repeated if the performance is not satisfactory.