UltraLIF: Fully Differentiable Spiking Neural Networks via Ultradiscretization and Max-Plus Algebra
Jose Marie Antonio Miñoza · 2026
Read UltraLIF: Fully Differentiable Spiking Neural Networks via Ultradiscretization and Max-Plus Algebra by Jose Marie Antonio Miñoza (2026) 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-2602-11206.txt.
Description
Spiking networks promise energy-fast, brain-like computing, but spikes do not differentiate. Jose Marie Antonio Miñoza, writing alone, offers UltraLIF, a frame that replaces surrogate gradients with ultradiscretization from tropical shape. A soft maximum converges to hard thresholding as a learnable heat falls.
Two neuron models follow: UltraLIF from leaky integrate-and-fire motion, and UltraDLIF from diffusion building gap-junction coupling. Both train by standard backpropagation with no forward-backward mismatch, and theory gives convergence to classic motion with bounded, non-vanishing gradients. Tests on six benchmarks beat surrogate-gradient rivals, most clearly in single-timestep settings on neuromorphic and time data.
Catalog record
| CREATOR | Jose Marie Antonio Miñoza |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 69 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2602-11206