Deep Learning Block-Set A Simulink Native Deep Learning Framework

This paper introduces the Deep Learning Block-Set, a Simulink-native block set library, enabling seamless integration of deep learning models in Simulink environments and empowering researchers and engineers to efficiently prototype data-driven systems. The Deep Learning Block-Set provides a selection of building blocks for composing deep neural networks, including various types of layers and activation functions. Distinguishing itself from the Deep Learning Toolbox provided by MathWorks, the presented library is designed to model the backward path and provide the calculation of the gradient natively within the Simulink environment. This unlocks optimization algorithms such as gradient descent for code generation, hence enabling online parameter optimization (e.g., real-time tuning of controller parameters), even after the Simulink model is translated to embedded targets. Furthermore, it opens up deep learning to a new group of users who are familiar with Simulink, while being licensed under the MIT License and available online.