Spectral Dictionary Learning for Generative Image Modeling
Andrew Kiruluta · 2025
Read Spectral Dictionary Learning for Generative Image Modeling by Andrew Kiruluta (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2504-17804.txt.
Description
Andrew Kiruluta presents a 2025 arXiv paper on a spectral route to generating images. Instead of the usual variational, adversarial, or diffusion methods, the model flattens each image into a signal and rebuilds it from learned wave-like basis pieces. Each piece carries its own frequency, phase, and size, and each image gets mixing weights over the shared dictionary.
A simple probability model over the weights then allows new images to be sampled directly. Frequency-domain losses keep both broad structure and fine texture honest. On the CIFAR-10 test set, the method matched rivals on quality while training more stably. The dictionary also gives direct control over image frequency content.
Catalog record
| CREATOR | Andrew Kiruluta |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 32 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2504-17804