Saved in:
| Main Author: | |
|---|---|
| Format: | Recurso digital |
| Language: | |
| Published: |
Zenodo
2025
|
| Online Access: | https://doi.org/10.5281/zenodo.17715490 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866901694206443520 |
|---|---|
| author | harihardiyan |
| author_facet | harihardiyan |
| contents | <h3> WaterFlow v1.0.0: Initial Thermodynamic Release</h3> <p>This is the first stable release of the <strong>WaterFlow</strong> project, a novel Normalizing Flow model trained using a custom <strong>Thermodynamic Loss function ($\mathcal{F}$)</strong> and <strong>Beta-Annealing</strong> schedule.</p> <h4>Key Features & Achievements:</h4> <ol> <li><strong>Thermodynamic Training:</strong> The model successfully converges by minimizing the Helmholtz Free Energy ($\mathcal{F}$), defined as $\mathcal{F} = \beta E(x) - \lambda S_{Shannon}$. This ensures the model seeks a state of optimal performance ($E$) combined with minimal structural complexity ($S$).</li> <li><strong>Perfect Reversibility:</strong> The core Affine Coupling architecture guarantees a bijective flow. As demonstrated by the generated images, the model can perfectly invert the transformation from the latent space back to the data space.</li> <li><strong>Capillary Action Visualization:</strong> The included script performs UMAP reduction to visualize the laminar flow of data points through all layers ($X \to Z \to X$), confirming the non-turbulent, non-disipative nature of the flow.</li> </ol> <h4>Contents of this Release:</h4> <ul> <li><code>waterflow_full.py</code>: The complete training, generation, and visualization script.</li> <li><code>requirements.txt</code>: Full list of dependencies.</li> <li><code>images/</code>: Contains visualization proof (e.g., Generated Samples and UMAP Flow).</li> <li><code>LICENSE-MIT.txt</code>: For non-commercial use (research/academic).</li> <li><code>LICENSE-COMMERCIAL.txt</code>: Placeholder for commercial licensing.</li> </ul> <p>We welcome academic researchers to utilize this framework under the MIT License.</p> |
| format | Recurso digital |
| id | zenodo_https___doi_org_10_5281_zenodo_17715490 |
| institution | Zenodo |
| language | |
| publishDate | 2025 |
| publisher | Zenodo |
| record_format | zenodo |
| spellingShingle | harihardiyan/WaterFlow-Thermodynamic-AI: WaterFlow: Initial Thermodynamic Release harihardiyan <h3> WaterFlow v1.0.0: Initial Thermodynamic Release</h3> <p>This is the first stable release of the <strong>WaterFlow</strong> project, a novel Normalizing Flow model trained using a custom <strong>Thermodynamic Loss function ($\mathcal{F}$)</strong> and <strong>Beta-Annealing</strong> schedule.</p> <h4>Key Features & Achievements:</h4> <ol> <li><strong>Thermodynamic Training:</strong> The model successfully converges by minimizing the Helmholtz Free Energy ($\mathcal{F}$), defined as $\mathcal{F} = \beta E(x) - \lambda S_{Shannon}$. This ensures the model seeks a state of optimal performance ($E$) combined with minimal structural complexity ($S$).</li> <li><strong>Perfect Reversibility:</strong> The core Affine Coupling architecture guarantees a bijective flow. As demonstrated by the generated images, the model can perfectly invert the transformation from the latent space back to the data space.</li> <li><strong>Capillary Action Visualization:</strong> The included script performs UMAP reduction to visualize the laminar flow of data points through all layers ($X \to Z \to X$), confirming the non-turbulent, non-disipative nature of the flow.</li> </ol> <h4>Contents of this Release:</h4> <ul> <li><code>waterflow_full.py</code>: The complete training, generation, and visualization script.</li> <li><code>requirements.txt</code>: Full list of dependencies.</li> <li><code>images/</code>: Contains visualization proof (e.g., Generated Samples and UMAP Flow).</li> <li><code>LICENSE-MIT.txt</code>: For non-commercial use (research/academic).</li> <li><code>LICENSE-COMMERCIAL.txt</code>: Placeholder for commercial licensing.</li> </ul> <p>We welcome academic researchers to utilize this framework under the MIT License.</p> |
| title | harihardiyan/WaterFlow-Thermodynamic-AI: WaterFlow: Initial Thermodynamic Release |
| url | https://doi.org/10.5281/zenodo.17715490 |