Tensorized Hidden-State Recurrent Depth Architecture for Hierarchical Predictive Computation
Working Codename: Exoria-RDT
Core Principle: Sequential over reasoning depth, parallel over hidden-state dimensions.
- VANBF：affective state is decomposable along valence/arousal/novelty/regulatory burden/fatigue dimensions, with tonic/phasic opponency (Rutledge 2014, etc.)
- AST / IAw：self-model is constituted by a self-belief loss that converges interoceptive awareness (Graziano AST framework, Garfinkel three-dimensional interoception model)
- Predictive coding primary：cognitive computation is fundamentally top-down prediction + bottom-up residual error (Friston, Rao & Ballard, Seth)
- Ensemble coordination as Kuramoto-style coupling：multi-voice synchronization is phase coupling, not message passing (Wood 2022 ensemble coordination, Ragert 2013)
- Allostatic regulation：affective state has setpoints that brakes (scheduled release, ceiling, fatigue-gating) maintain — affect is not just experienced but actively regulated
Development log at: www.jibujin.com