arXiv:2506.05577v3 Announce Type: replace 
Abstract: Agentic AI aims to create systems that set their own goals, adapt proactively to change, and refine behavior through continuous experience. Recent advances suggest that, when facing multiple and unforeseen tasks, agents could benefit from sharing machine-learned knowledge and reusing policies that have already been fully or partially learned by other agents. However, how to query, select, and retrieve policies from a pool of agents, and how to integrate such policies remains a largely unexplored area. This study explores how an agent decides what knowledge to select, from whom, and when and how to integrate it in its own policy in order to accelerate its own learning. The proposed algorithm, \emph{Modular Sharing and Composition in Collective Learning} (MOSAIC), improves learning in agentic collectives by combining (1) knowledge selection using performance signals and cosine similarity on Wasserstein task embeddings, (2) modular and transferable neural representations via masks, and (3) policy integration, composition and fine-tuning. MOSAIC outperforms isolated learners and global sharing approaches in both learning speed and overall performance, and in some cases solves tasks that isolated agents cannot. The results also demonstrate that selective, goal-driven reuse leads to less susceptibility to task interference. We also observe the emergence of self-organization, where agents solving simpler tasks accelerate the learning of harder ones through shared knowledge.

تسعى الذكاء الاصطناعي الوكيلة إلى إنشاء أنظمة تحدد أهدافها الخاصة وتتكيف مع التغيرات. تستكشف هذه الدراسة كيف يمكن للوكلاء اختيار ودمج المعرفة من وكلاء آخرين بشكل فعال لتحسين عمليات التعلم الخاصة بهم. الخوارزمية المقترحة، MOSAIC، تستخدم إشارات الأداء وتضمينات المهام لتحسين استرجاع السياسات وتكوينها، مما يسرع التعلم في الأنظمة الوكيلة التعاونية.

La IA agentica se centra en crear sistemas que establecen sus propios objetivos y se adaptan a los cambios. Este estudio investiga cómo los agentes pueden seleccionar e integrar efectivamente el conocimiento de otros agentes para mejorar sus procesos de aprendizaje. El algoritmo propuesto, MOSAIC, utiliza señales de rendimiento y embeddings de tareas para mejorar la recuperación y composición de políticas, acelerando así el aprendizaje en sistemas colaborativos agenticos.

L'IA agentique vise à créer des systèmes qui fixent leurs propres objectifs et s'adaptent aux changements. Cette étude examine comment les agents peuvent sélectionner et intégrer efficacement les connaissances d'autres agents pour améliorer leurs processus d'apprentissage. L'algorithme proposé, MOSAIC, utilise des signaux de performance et des embeddings de tâches pour améliorer la récupération et la composition des politiques, accélérant ainsi l'apprentissage dans les systèmes collaboratifs agentiques.

Agentic AI focuses on creating systems that autonomously set goals and adapt to changes. This study investigates how agents can effectively select and integrate knowledge from other agents to enhance their learning processes. The proposed algorithm, MOSAIC, utilizes performance signals and task embeddings to improve policy retrieval and composition, thus accelerating the learning in collaborative agentic systems.

Policy Search, Retrieval, and Composition via Task Similarity in Collaborative Agentic Systems

Was this article worth reading? Share it

LucidQuery AI

Augmeta

GoAgentic