EntropyML is a compact machine‑learning framework that presents neural models through thermodynamic concepts such as entropy, equilibrium, and free‑energy‑like objectives. The framework provides small, readable implementations intended for experimentation and conceptual study.
EntropyML includes autoencoders, variational autoencoders, and a thermodynamic variant of the VAE. These models are implemented in plain C# and are designed to be structurally simple and transparent.
The project was previously named ThermoML. The name was changed to avoid collision with the existing NIST ThermoML standard.
A minimal autoencoder implementation for reconstruction experiments.
A standard VAE implementation with clear latent‑space structure.
A variant of the VAE that introduces equilibrium and entropy‑potential concepts for interpreting latent behavior.
A simple feedforward network used in examples.
Small utilities for synthetic and example datasets.
Runnable examples demonstrating AE, VAE, TVAE, NN, and data utilities.
EntropyML-Dev/
background/
doc/
solution/
EntropyML/
EntropyML.AE/
EntropyML.Data/
EntropyML.NN/
EntropyML.VAE/
EntropyML.TVAE/
Examples/
Documentation is currently being migrated to the EntropyML identity. The following documents have been renamed:
- EntropyML_API_Sheet.md
- EntropyML_FolderStructure.md
- EntropyML_HMD.md
- EntropyML_Landing.md
- EntropyML_QuickStart.md
- EntropyML_ReleaseBundle.md
- EntropyML_SpecLite.md
- EntropyML_Terminology.md
- EntropyML_Versioning.md
These documents will be updated after the implementation surface stabilizes.
Examples can be executed directly using the .NET command line:
dotnet run
Each example directory contains its own project file.
MIT License. See the LICENSE file for details.