| ai.viXra.org > Artificial Intelligence > 2607 |
|
 |
| ai.viXra.org > Artificial Intelligence > 2607 |
|
|
[3] ai.viXra.org:2607.0087 [pdf] submitted on 2026-07-30 00:32:22
Authors: Peilin Wen
Comments: 38 Pages.
We present the Modular Neural Network (MNN), a brain-inspired computational architecture that fundamentally transcends the limitations of current large-scale Transformer and Mixture-of-Experts (MoE) models. MNN employs five core mechanisms working in concert: (1) dynamic semantic routing with persistent memory, enabling adaptive expert allocation based on input complexity; (2) independent experts withinternal recursion and dual-role local dynamic penalty for self-stabilization; (3) bijectionflow adapters with compression projection and mutual information maximization forefficient cross-modal communication; (4) directional precision-weighted permeation forasymmetric, selective inter-expert information flow; and (5) global energy functional-driven consensus with posterior verification and self-correction. We provide complete mathematical formalization of all mechanisms and propose a training framework based on implicit differentiation that resolves the computational challenges of dynamic iteration depth. An engineering deployment scheme, encompassingtime-box scheduling, stateless horizontal scaling, elastic degradation, and hardware-level isolation, provides a practical blueprint for implementation on existing GPU infrastructure. The MNN architecture simulates the hierarchical compression, consensus optimization, and metacognitive arbitration mechanisms of biological intelligence, providing a theoretically self-consistent and engineeringly executable blueprint for building scalable, interpretable, and continuously evolving generalartificial intelligence systems.
Category: Artificial Intelligence
[2] ai.viXra.org:2607.0072 [pdf] submitted on 2026-07-27 08:08:11
Authors: Veaceslav Molodiuc
Comments: 8 pages. AI-assisted preprint, Version 1.1.
Verified distributed kernels require more than authenticated messages: they require formal identity, recoverable operational states, and measurable consensus risk. This paper connects formal identity, discrete stabilization, the LoculoSoft kernel vocabulary, and PRISM probabilistic model checking. LoculoSoft is treated cautiously as a proposed structural framework and research direction. A reproducible PRISM 4.10.1 check on a three-process Herman-style token ring shows stable reachability with probability 1 and maximum expected stabilization time near 4/3 steps.Keywords: formal identity, discrete stabilization, consensus risk, distributed kernels, PRISM, LoculoSoft.
Category: Artificial Intelligence
[1] ai.viXra.org:2607.0010 [pdf] submitted on 2026-07-06 19:47:16
Authors: P. H. Antom
Comments: 11 Pages.
We present the Feedback Epistemic Equilibrium Law (FEEL), a formal stability theorem for artificial reasoning systems operating under corrective oversight. Standard control theory stabilises a plant using a controller assumed reliable by construction. FEEL removes this assumption: the feedback mechanism is itself a dynamical system with its own persistence, susceptibility to epistemic contamination, and potential for instability. We model the joint system as the affine recurrence Z_(t+1) = MZ_t + c where the block matrix M (A, -C; D, E)encodes intrinsic reasoning evolution (A) , corrective intervention (C), epistemic leak from reasoning into feedback (D), and feedback persistence (E), We prove that ρ(M) < 1 is necessary and sufficient for (i) existence and uniqueness of an equilibrium, (ii) global convergence from any initial condition, (iii) exponential stability with an explicit rate, and (iv) bounded epistemic-feedback disagreement. We characterise the failure regime ρ(M) >= 1, cataloguing divergence, oscillation, feedback amplification, and runaway correction cascades. We illustrate the criterion on a toy coupled system in which increasing the epistemic-leak strength alone — holding each subsystem’s standalone spectral radius fixed — drives the joint system through the predicted ρ(M) = 1 transition, confirmed across 200 random initial conditions. The result supplies a computable certificate for safe feedback in AI systems and identifies a structural gap in existing alignment frameworks: local stability of reasoning and feedback components does not imply global stability of their coupling.
Category: Artificial Intelligence