Preprint / scholarly article

A Metacognitive Perturbation Framework for Neuro-Inspired AI Optimization

K. Takahashi

Published
DOI
10.5281/zenodo.16860493

Full text PDF (Zenodo)

Abstract

Conventional AI optimization, particularly in deep learning, is often hampered by convergence to suboptimal local minima, limiting transformative performance gains.

Keywords

  • AI
  • large language models
  • neuro-inspired AI
  • bayesian optimization
  • metacognition
  • local optima
  • exploration-exploitation dilemma
  • predictive processing
  • free-energy principle
  • AI safety

Identifiers and source records