همايش ملی پيشرفت های معماری سازمانی

صفحه اصلی / نهمین همایش پیشرفت‌های معماری سازمانی ایران

FedRLEA: Federated Aggregation of Parallel SLMs with RL-Based Prompt Optimization for Smart Enterprise Architecture

نویسندگان :

Keyhan Mohammadi¹ Arman Moradi² Reza Ebrahimi Atani³

1- University of Guilan 2- University of Guilan 3- University of Guilan

کلمات کلیدی :

Federated Learning،Privacy-Preserving AI،Small Language Models (SLMs)،Prompt Optimization،Reinforcement Learning،Hash- graph Consensus،Smart Enterprise Architecture.

چکیده :

The rapid adoption of artificial intelligence (AI) in enterprise architectures has created new opportunities for intelligent automation and decision support while simultane- ously raising concerns regarding data privacy, scalability, and adaptability. Traditional centralized training of large language models (LLMs) is impractical for many organizations due to high computational demands and privacy risks associated with data aggregation. This paper introduces a federated aggregation framework for parallel Small Language Models (SLMs) that applies Hashgraph-based decentralized learning to eliminate reliance on central servers. To further enhance adaptability and task-specific accuracy, the framework integrates a reinforcement learning (RL)-based prompt optimization mechanism, enabling continuous refinement of SLM prompts through interaction driven feedback. The proposed system supports cross organi- zational collaboration under privacy-preserving constraints and is designed for deployment within smart enterprise architectures. Our formal model analysis demonstrates efficient distributed model training, scalability across heterogeneous nodes, and re- duced communication overhead compared to traditional feder- ated learning approaches. The integration of RL-based prompt tuning yields measurable improvements in model relevance and responsiveness to evolving enterprise data. The results highlight the potential of federated SLM aggregation to provide a secure, adaptive, and resource-efficient foundation for next-generation AI-based enterprise architectures.