Presentation Information
[2Yin-A-01]An Information-Dynamical Model of Nonequilibrium Memory Updating via Projection Operator Sequences
〇Takeaki Makimo1, Satoru Makino (1. INFORMATION DEVELOPMENT CO., LTD.)
Keywords:
Nonequilibrium Memory,Projection Operators,Unitary Recurrent Neural Networks,Effective Degree of Freedom Scaling,Information Dynamical Metrics
We propose an information-dynamical model of action selection in preference formation, where memory updates follow unitary-like dynamics interleaved with recall/write projections. Our objective is to relate internal order metrics to external task success and to identify when observation-interface design causes their decoupling. Our results show an “output failure under information overload,” in which no final action is produced despite stable internal order. A stroboscopic view of the operator sequence suggests a Floquet-like handle on nonequilibrium order and the χ-scaling functional fluctuations in an effective dimension (bond dimension χ). We conclude that robust intelligence may require explicit observation operators and dynamically expandable information spaces to preserve coherence while enabling adaptive variability.