RADENN: A Domain-Specific Language for the Rapid Development of Neural Networks

Abstract

RADENN is a domain-specific language designed to rapidly develop fully connected neural networks for classification and regression problems. The primary objective of this language is to make neural network algorithms more accessible to a broader audience. RADENN is built on top of Keras API with Tensorflow as its back-end. This language follows the imperative paradigm; it uses dynamic scoping, is weakly typed, and utilizes type inference. The contribution of RADENN is to incorporate specific data types and built-in functions to facilitate the creation, training, and evaluation of neural networks. All these features make RADENN an ideal tool for Data Scientists, Data Analysts, Big Data Engineers, Software Enginers, and anyone who needs a fast and efficient way to create prototypes and models without extensive programming or deep learning knowledge. This work provides a detailed overview of the features of RADENN and compares it to Keras and PyTorch, which are currently among the most widely used libraries in industry and research.

Keywords

• Neural networks
• domain-specific language
• deep learning