Neural Network

This module combines the given layers to a Neural Network. The Neural Network can be used, for example, as a function approximator in a Supervised Learning problem.

The user can choose between different backends to perform the computations. Besides the Korali lightweight implementation, we support

• Intel’s oneDNN library.

• Nvidia’s cuDNN library.

This module enables two Operation types:

• Training: In this operation, the weights and biases are optimized to minimize a given loss function. The initial guess is chosen according to the specified Weight Initialization.

• Inference: In this operation, weights and biases are either provided by the user, or obtained by the training operation. This configuration is only used to compute the output for a given input.

Sub-Categories:

• Layers
  • Activation Layer
  • Linear Layer
  • Recurrent Layer
    • GRU Layer
    • GRU Layer
  • Input Layer
  • Output Layer