Neural Network model#
The NN metamodel enables to represent neural networks. The key concepts in the NN metamodel are represented using meta-classes and their associations. Our design was heavily inspired by the two popular deep learning frameworks PyTorch and TensorFlow. Specifically, we compared concepts from the two frameworks to come up with a metamodel design that is general to represent neural network components and allows the definition of both sequential and non-sequential architectures. This approach ensures that the metamodel remains versatile and adaptable across different contexts within neural networks development.
Note
The class highlighted in green originates from the structural metamodel.
Supported notations#
To create an NN model, you can use the following notations: