Overview / Usage

The need for using DPC++

In general cases for neural net we see that we know the output if given input is given but we exactly don’t know formula to achieve it.Hence in this case we need lot of training data and parallelism is there so DPC++ comes into picture.

Methodology / Approach

Neural net simulation using DPC++

The need for using DPC++

In general cases for the neural net, we see that we know the output if given input is given but we exactly don’t know the formula to achieve it. Hence, in this case, we need a lot of training data and parallelism is there so DPC++ comes into the picture.

To bring in DPC++ in the picture we have to think of these patterns

We will need to create functions

We will have to see that it supports dependencies such as CMake etc.

We also have to make sure the kernels work perfectly so that we can target it to different inference capabilities such as XGPU’s ,FPGA’s and CPU’s .

We have to initiate the neural network as a topology.

For training, we need to have some other members functions

Feedforward.

To hold the values we need a data structure such as an array. We can work with Buffers for DPC++ over here.

We also have to track how the outputs are behaving and also the learning process for that we need another member function

Backprop

For deriving the results we need another function that acts as a container we call it as

getResults

For training we need the following functions

Feedforward

Backprop

​

Technologies Used

AI

Intel One API

DPC++

Dev Cloud

Concepts of neural net

Documents and Presentations