Welcome to nnodely’s documentation!
nnodely is a framework designed to facilitate the creation and deployment of Model-Structured Neural Networks (MSNNs). Modeling, control, and estimation of physical systems impose constraints that differ fundamentally from typical deep-learning tasks (e.g., images or text). In engineering applications, models often need to respect known physical laws or constraints, operate in real time, remain interpretable, and generalize reliably even when only limited experimental data are available. MS-NNs combine the learning capabilities of neural networks with structural priors grounded in physics, control and estimation theory, enabling:
Data Efficiency: By embedding structural priors, MS-NNs can learn effectively from limited data, reducing the need for extensive datasets.
Generalization: The incorporation of domain knowledge allows MS-NNs to generalize better to unseen scenarios.
Interpretability: The structured nature of MS-NNs enhances interpretability, allowing practitioners to understand and trust the model’s predictions.
Real-time: MS-NNs can be designed for real-time applications, making them suitable for control and estimation tasks in dynamic environments.
The main objective of the nnodely framework is to allow fast prototyping of MS-NNs for modeling, estimation and control of physical systems by embedding structural priors knowledge into the networks’ architecture.
In this documentation you will find a comprehensive guide for getting started with nnodely, illustrating the main blocks that constitute the framework.
Repository Getting Started Applications
Overview
