Adaptive computing structures for service-oriented multi-agent systems based on knowledge models

Abstract

This paper proposes an adaptive computing framework for service-oriented multiagent systems, based on knowledge models inspired by the hierarchical organization of the human brain. The approach integrates neurophysiological principles of conscious and subconscious processing with rigorous mathematical formalization and hardware-oriented architectural design. The conscious–subconscious interaction is modeled as a two-level computational hierarchy, in which subconscious processing is fast, parallel, adaptive, and high-dimensional, and conscious processing is deliberative, symbolic, and low-dimensional, being responsible for control, planning, and decision-making. An attention-based coupling mechanism controls the flow of information between the two levels, allowing for dynamic adaptation and efficient use of resources. Based on this model, a heterogeneous hardware architecture is proposed that maps subconscious processing to NPU/GPU accelerators, and conscious processing to CPU units. The framework is extended to multi-agent systems, in which each agent implements a conscious–subconscious hierarchy, and the emergent coordination is achieved through a collective conscious level. The approach supports distributed intelligence, scalability, and adaptive service composition.