Research Profile

Independent researcher specialising in AI safety, multi-agent systems, active inference, and formal mathematical frameworks for autonomous agency. Author of four foundational preprints developing a unified theory of autonomous AI: from precision-weighted inference (Belief-Precision Geometry) to hierarchical identity dynamics (Hierarchical FEMI) to multi-agent safety and power equity (PVF-U) to autonomous identity transformation (Born Again). PVF-U is publicly available; the remaining three preprints are currently private to preserve intellectual property.

Research Publications

PVF-U: The Unified Primitive Valence Foundation (2026) — Public

• Introduces three-tiered architecture solving the “Emptiness Asymptote” (agent collapse in isolation).
• Formal guarantees: Bounded-Input Bounded-Output (BIBO) stability, gate convergence, veto stabilisation.
• Achieves Power Equity Index (PEI) > 0.83 in multi-agent systems.
• Validated across 4 benchmarks with 200 replications, outperforming 6 Reinforcement Learning (RL) baselines.

Belief-Precision Geometry (2026) — Private Preprint

A Riemannian Framework for Affective Predictive Processing and Multi-Sensor Fusion.

• Unifies information geometry, predictive processing, and affective computing.
• Proves convergence of natural gradient belief updates.
• Identifies critical stability failure in hierarchical extensions and provides a five-part fix.
• Derives Bayesian belief fusion that jointly combines precision and mean.

Hierarchical FEMI (2026) — Private Preprint

Multi-Level Identity Dynamics with Deep Temporal Prior and Identity-Preserving Action Selection.

• Extends identity theory to L coupled levels with closed-form variance cascade.
• Derives a two-phase step response and a taxonomy of 4L + 5 failure modes.
• Provides coupling learning dynamics with convergence guarantees.

Born Again (2026) — Private Preprint

Hierarchical Belief States with Structural Plasticity for Autonomous Identity Transformation.

• Introduces valence-driven structural plasticity for autonomous model restructuring.
• Three-phase reset protocol: severance, dissolution, re-formation.
• Formal convergence guarantees and permanent structural scarring mechanism.
• Domain-general: robotics, financial systems, power grids, personal AI, fleets, drug discovery.

Software & Engineering

PVF-U Simulation Pipeline (Python/JAX) — Full implementation of three-tiered architecture. Four benchmark environments. n = 200 replication harness with BCa bootstrap confidence intervals. Open-source: github.com/preciousadeniyi/Pvf_u.

Technical Skills: Active Inference, Variational Free Energy, Predictive Processing; Natural Gradient Descent, Information Geometry; Stochastic Differential Equations, Lyapunov Stability Analysis; Python, JAX, NumPy; Reinforcement Learning (PPO, Actor-Critic); Multi-Agent Systems.

Grants & Funding Applications (Pending)

• EA Funds — Long-Term Future Fund ($150,000, under final review)
• Longview Philanthropy — Power Concentration RFP ($150,000, submitted)
• Sentient Foundation — Open Source AGI Grant ($200,000, submitted)
• Survival & Flourishing Fund — Main + Fairness Track ($200,000, submitted)

Education

Self-directed research in mathematical AI, active inference, and formal systems (2024–present). Independent study: stochastic processes, linear systems theory, information geometry, variational inference.

Languages

• English (Native/Professional fluency)
• Yoruba (Native)

Confidentiality Note

Preprints for BPG, HFEMI, and Born Again are currently private and are not publicly available. They are included here to demonstrate the breadth of my research program. Full manuscripts are available upon request under confidentiality.