Open-source research software for Sensory Neural Network modelling and simulation
The NEURALSENS software framework explores how sensor-dependent electrical characteristics can be incorporated into neural-network models for low-level in-sensor data processing.
Developed within work package KPB2, the software provides a computational basis for studying the relationship between sensing-element responses, network parameters and predicted outputs.
Access the NEURALSENS software on GitHub
Deliverable information
- Deliverable: D2.3 — Open-source software for SeNN simulation with crossbar arrays of resistive sensing elements.
- Work package: KPB2 — SeNN algorithm.
- Responsible institution: Institute of Informatics, Slovak Academy of Sciences (II-SAS).
- Software implementation: Jaromír Klarák.
- Programming language: Python.
- Licence: MIT.
Software concept and capabilities
In the SeNN concept, sensing-element characteristics contribute to network functionality. The software investigates this approach by representing sensor-dependent behaviour through polynomial functions and incorporating these functions into network calculations.
The published implementation includes:
- Conventional neural-network models for comparison.
- Polynomial representations of sensor-dependent responses.
- Synthetic temperature-sensing examples with noise.
- Training procedures and prediction-error calculations.
- Visualisation of network predictions, errors and training loss.
These examples support exploration of how changes in sensor-response functions affect the behaviour of a network.
Repository contents
The public repository contains the following files:
| File | Content |
|---|---|
| SeNN.py | Sensor-dependent network modelling, temperature examples, comparison with a conventional neural network, and graphical evaluation. |
| SeNN_Theory.py | Exploratory neural-network calculations and training examples supporting methodology development. |
| LICENCE | MIT licence and software attribution. |
The software is distributed as Python research scripts. The included temperature examples use synthetic signals to illustrate model behaviour. Application to a particular experimental sensor requires suitable response data, model parameters and validation against measurements.
Role within the NEURALSENS project
The software forms part of the consortium’s research on linking sensing materials and devices with hardware-oriented neural-network design.
The project connects three complementary work packages:
- KPB1, led by Comenius University in Bratislava: hydrogen-sensing elements and their characterisation.
- KPB2, led by II-SAS: SeNN design and simulation methodology.
- KPB3, led by the Institute of Electrical Engineering SAS: sensing-array fabrication and laboratory-prototype integration.
All three institutions participate in the work packages. The overall project is coordinated by the Institute of Electrical Engineering, Slovak Academy of Sciences.
Within this research programme, temperature sensing provides a starting point for investigating the design methodology. Extension to hydrogen sensing involves connecting network models with the measured characteristics of hydrogen-sensitive elements.
Access and reuse
View source code on GitHub · Download source code (ZIP)
The source code is publicly available under the MIT licence. Please refer to the repository’s licence file for the applicable terms and attribution.
For the wider research context, see Project Overview and Results & Deliverables.
Funding
Development of the software was supported by the European Union – NextGenerationEU through the Recovery and Resilience Plan of the Slovak Republic under project No. 09I05-03-V02-00058 — NEURALSENS.