Prime-Fork Neural Network (PFNN): A Dynamic Neural Architecture Driven by Prime Number Transitions
Fuente:
Zenodo
Saved in:
| Main Authors: | , |
|---|---|
| Format: | Recurso digital |
| Language: | English |
| Published: |
Zenodo
2025
|
| Subjects: | |
| Online Access: | |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866901773541703680 |
|---|---|
| author | Rezapour, Majid Rezapour, Ramin |
| author_facet | Rezapour, Majid Rezapour, Ramin |
| contents | <p>This article introduces the Prime-Fork Neural Network (PFNN), a novel dynamic neural architecture inspired by the Prime-Induced Function Switching Model (PIFSM). The PFNN design features four distinct functional sub-networks—linear, quadratic, logarithmic, and square-root—each activated by the prime-indexed modular value π(n) mod 4. The switching mechanism is deterministic and non-continuous, offering a new computational paradigm for hybrid neural logic and adaptive learning.</p> <p>The paper presents:</p> <p>- Mathematical formulation of switching logic based on prime-counting functions.</p> <p>- Complete architecture and algorithmic design of PFNN.</p> <p>- Simulations on datasets of size n = 1000 with input vectors of dimension 10.</p> <p>- Numerical outputs including gradient variance, entropy shifts, and sparsity metrics.</p> <p>- Potential applications in hybrid AI systems, low-power encoding, anomaly detection, and decision modeling.</p> <p>- Comparison with static neural networks and capsule architectures.</p> <p>The proposed model demonstrates promising improvements in generalization, sparsity control, and computational adaptability, positioning PFNN as a candidate architecture for future intelligent systems.</p> |
| format | Recurso digital |
| id | zenodo_https___doi_org_10_5281_zenodo_16621748 |
| institution | Zenodo |
| language | eng |
| publishDate | 2025 |
| publisher | Zenodo |
| record_format | zenodo |
| spellingShingle | Prime-Fork Neural Network (PFNN): A Dynamic Neural Architecture Driven by Prime Number Transitions Rezapour, Majid Rezapour, Ramin Prime Numbers Neural Architecture Function Switching Modular Logic Prime-Indexed Transition Dynamic Neural Network Adaptive Computation Nonlinear Activation Prime-Based Encoding Hybrid AI Models Discrete Learning Systems Mathematical Discontinuity Logic-Based Switching Forked Neural Logic Entropy-Guided Adaptation Phase-Switch Neural Dynamics Intelligent Signal Routing Deterministic Function Selection Gradient-Sensitive Architecture Prime-Fork Neural Framework <p>This article introduces the Prime-Fork Neural Network (PFNN), a novel dynamic neural architecture inspired by the Prime-Induced Function Switching Model (PIFSM). The PFNN design features four distinct functional sub-networks—linear, quadratic, logarithmic, and square-root—each activated by the prime-indexed modular value π(n) mod 4. The switching mechanism is deterministic and non-continuous, offering a new computational paradigm for hybrid neural logic and adaptive learning.</p> <p>The paper presents:</p> <p>- Mathematical formulation of switching logic based on prime-counting functions.</p> <p>- Complete architecture and algorithmic design of PFNN.</p> <p>- Simulations on datasets of size n = 1000 with input vectors of dimension 10.</p> <p>- Numerical outputs including gradient variance, entropy shifts, and sparsity metrics.</p> <p>- Potential applications in hybrid AI systems, low-power encoding, anomaly detection, and decision modeling.</p> <p>- Comparison with static neural networks and capsule architectures.</p> <p>The proposed model demonstrates promising improvements in generalization, sparsity control, and computational adaptability, positioning PFNN as a candidate architecture for future intelligent systems.</p> |
| title | Prime-Fork Neural Network (PFNN): A Dynamic Neural Architecture Driven by Prime Number Transitions |
| topic | Prime Numbers Neural Architecture Function Switching Modular Logic Prime-Indexed Transition Dynamic Neural Network Adaptive Computation Nonlinear Activation Prime-Based Encoding Hybrid AI Models Discrete Learning Systems Mathematical Discontinuity Logic-Based Switching Forked Neural Logic Entropy-Guided Adaptation Phase-Switch Neural Dynamics Intelligent Signal Routing Deterministic Function Selection Gradient-Sensitive Architecture Prime-Fork Neural Framework |
| url | https://doi.org/10.5281/zenodo.16621748 |