Work

Neuro-Symbolic Cognitive Memory under Traumatic Disturbances (NS-CMTD)

Year: 2026
Type: preprint
Abstract: The article presents a mathematical neuro-symbolic model of cognitive memory under traumatic disturbances—NS-CMTD (Neuro-Symbolic Cognitive Memory under Traumatic Disturbances)—featuring a dual structure in which the agent's subjective memory (working memory) interacts with its objective memory (deep memory). Deep memory serves as a reference point, representing ethical rules, cultural code, or a domain-specific rule base in formal applications. The model accounts for the fact that memory traces are subject to continuous traumatic disturbances, understood as an internal entropy jump caused by a mismatch between incoming data and accepted assumptions. By integrating defense mechanisms with the mathematical quantification of partial entropies, the model describes the processes of illusion formation, depreciation, and memory deformation. The paper proposes a three-dimensional data (symbol) storage space and treats them as 3D objects known from computer graphics, which preserves high computational performance through parallel processing. The separation of semantic objects (clusters of interconnected events in the form of 3D objects) enables counterfactual reasoning and local corrections of trauma-distorted traces without violating the global consistency of the memory architecture. The proposed model shifts the paradigm from simple fact-checking to building disturbance-resilient systems capable of autonomously diagnosing computational costs, resolving internal inconsistencies, and simulating defense mechanisms in therapeutic applications and immersive environments. Consequently, this architecture mitigates the phenomenon of hallucination and semantic drift in language model-based systems, eliminates the 'black box' problem by increasing the interpretability of reasoning, and facilitates the creation of dedicated expert systems, serving primarily as a direct alternative to classical RAG (Retrieval-Augmented Generation) solutions.
Language: English
Open Access status: green
OpenAlex ID: W7234041099
Cites:
Cited by:
Related to: 10