CUP (Computational Unified Protocol) is a compact, invertible, and modular neural architecture. It is designed to serve as a neural “building block” for symbolic, logical, and analytical systems.
What it does
- Learns and reproduces symbolic or logical functions.
- Enables inversion: from output, reconstruct original input.
- Supports chaining: stack multiple brains (e.g. CUP → CUP++ → CUP++).
- Each brain is a standalone module: save, load, share.
Architecture levels
- CUP: minimal, analytic, with 2
tanhlayers. - CUP++: adds contextual modulation masks (M₁, M₂) for adaptive processing.
- CUP++++: includes layer normalization, residuals, parametric activation functions (tanh, sigmoid, sinh).
dfairesearch.com
Use cases & advantages
- Neural calculators (addition, multiplication, division) with interpretability.
- Reverse engineering symbolic inputs from results.
- Embedding small brain modules inside simulators, agents, or AI systems.
- Connecting networks of brains with invertible logic.
dfairesearch.com
Licensing & availability
- CUP-Framework v1.0.0 available (core libraries, example code) under a noncommercial license.
- Commercial use requires explicit permission.
dfairesearch.com - Project is geared toward education, research, prototyping.