Deep Learning Block-Set

The blockset provides a basic set of Simulink blocks for building deep learning models.

The Deep Learning Block-Set (DLBS) aims to make deep learning model design, training, and deployment possible directly in Simulink using modular subsystem blocks. DLBS combines tensor operations, layers, nonlinear activations, loss functions, optimizers, and routing utilities so models can be built graphically and trained in natively in Simnulink.

DLBS is intended for educational use, rapid prototyping, and workflows that benefit from Simulink-native model composition and code generation tooling. For a first hands-on model, start with the peak-classification example in the Getting Started page.

Citing

If you use this blockset in your research, please cite the following paper:

A. Oerder et al., "Deep Learning Block-Set - A Simulink Native Deep Learning Framework", 2026 International Power Electronics Conference (IPEC-Nagasaki 2026- ECCE Asia), to be published.

This entry will be updated, once the original paper is published in IEEE Xplore after the 2026 International Power Electronics Conference (IPEC 2026) held in Nagasaki, Japan

License

Copyright (c) Karlsruhe Institute of Technology (KIT).

This blockset is licensed under the MIT License. See the LICENSE.txt file for more details.

Development Repository

The GitHub repository KIT-ETI/Deep-Learning-Block-Set constitutes the development version of the blockset. If you are looking for the installable toolbox, visit Mathworks Exchange.

Structure of the Repository

deep-learning-block-set/
├── +dlbs/                      # Tooling for packaging and testing the toolbox
├── docs/                       # General documentation of the toolbox
├── toolbox/
│   ├── examples/
│   │   └── dlbsExamplePeakClassification
│   ├── library/
│   │   ├── **/                 # Implementaion of block of the library
│   │   ├── slblock.m
│   │   └── dlbs.slx
│   ├── dlbsAbstractHarnessTest.m
│   ├── dlbsHelpPageTemplate.m
│   └── info.xml
└── README.md

Each block implementation should adhere the the following folder structure, examplary for tanh:

tanh/
├── dlbsBlockTanh.slx           # Simulink Library implementing a single block, which is reference by dlbs.slx
├── dlbsHelpBlockTanh.m         # Documentation if this block, which returns an object of "dlbsHelpPageTemplate"
└── test
    ├── dlbsTestHarnessTanh.slx # Provide a harness applying constant inputs to the block to test
    └── dlbsTestTanh.m          # Testclass, inherting dlbsAbstractHarnessTest

Development

This block set was developed using MATLAB R2025b. It uses the python-integration for behavioural testing against PyTorch. MATLAB R2025b does not support python versions newer than 3.12.

To add the development version of the toolbox to the MATLAB path run:

dlbs.initPath();

For running tests you also need to create a virtual environment in the current folder with the given requirements. This can be done using

dlbs.createPyVenv();
dlbs.activatePyVenv();

Developing new blocks should be in line with the given folder structure. Then the steps needed to create a new block are:

1. Implement a new block, preferably by reusing other math blocks already provided in this library.
2. Optimize where appropriate
3. Write a delbsHelp*.m and add it to docs/helptoc.xml
4. Add a unit test.
5. Add the block to the main library toolbox/library/dlbs.slx

Packing the toolbox

To pack the toolbox run:

version = "0.0.0";
runTests = true";
buildDocuments = true;
package = true;
outputFolder = "";
dlbs.packageToolbox(version, ...
    runTests=runTests, ...
    buildDocumentation=buildDocumentation, ...
    package=package, ...
    outputFolder=outputFolder, ...
    );

Be aware that the toolbox UUID as well as the author infomation is hardcoded in +dlbs/packageToolbox.m and need to be changed when forking the repository.

Further Notes

The documentation consisting of dlbsHelp*.m files as well as scripts in +dlbs/+helpers/ were written using AI-tooling using GPT 5.3-Codex.

This project is not associated with MathWorks™.

Copyright notices

• MathWorks™, MATLAB®, Simulink®, Embedded Coder®, HDL Coder™ and Deep Learning Toolbox™ are registered trademarks of The MathWorks, Inc.
• PyTorch is a registered trademark of The Linux Foundation.