COOL: Chain-Oriented Objective Logic with Neural Networks Feedback Control for Multi-DSL Regulation

Jipeng Han

Keywords: Chain-of-Logic, Neural Network Feedback Control, Multi-DSL Regulation

TL;DR: A groundbreaking multi-DSL regulation framework combining control theory and neural networks, with theoretical and experimental proofs.

Abstract: Multi-DSL regulation requires dynamic coordination of modular domain logic, yet existing frameworks lack mechanisms for cross-DSL state management. We address these challenges with \textbf{COOL (Chain-Oriented Objective Logic)}, a neural-symbolic framework that introduces: (1) \textbf{Chain-of-Logic (CoL)}: Structures the reasoning process into hierarchical, expert-guided multiple sub-DSLs with heuristic vectors and runtime keywords; and (2)~\textbf{Neural Network Feedback Control (NNFC)}: A self-correcting mechanism that isolates neural components into reusable libraries, filtering erroneous predictions via sequential network coupling. Through rigorous theoretical analysis, we formally establish the efficacy of CoL and NNFC components. Ablation studies on relational and symbolic tasks validate that: CoL achieves \textbf{70\% higher accuracy} than non-CoL DSLs while reducing computational overhead by \textbf{91\% fewer tree operations} and \textbf{95\% faster reasoning}. Under adversarial conditions—insufficient training data, increased complexity, and multi-library requirements—NNFC further improves accuracy by \textbf{6\%} and reduces tree operations by \textbf{64\%} compared to the CoL-only variant. Both theoretical analysis and experimental validation confirm COOL as a highly efficient and reliable framework for multi-DSL regulation.

Primary Area: neurosymbolic & hybrid AI systems (physics-informed, logic & formal reasoning, etc.)

Submission Number: 9002