Post doc - Native Semantic Communications Between AI Agents".

about the role

Your role as a Post-doctoral researcher will focus on "Native Semantic Communications Between AI Agents".

This post-doctoral position is part of the collaborative SNS 6GARROW project, which aims to develop AI-native 6G architecture, optimize networks via AI agents, and explore semantic communications (https://6garrow.com).

Context: Large Language Models (LLMs) are experiencing rapid and continuous evolution, now incorporating multimodal capabilities, reasoning, and action functions. This evolution enables the emergence of new AI agent services, deployed as personal or professional assistants. These agents must collaborate autonomously to process user requests, gradually transforming these services into machine-to-machine interactions.

Research Problem: To communicate effectively, AI agents must be able to exchange information using a common representation. This representation, specific to AI exchanges, will likely take a form similar to that used for internal model representations: embeddings. However, each agent currently has its own embedding model, which cannot be natively understood by other agents. This is why the Unified Representations research field focuses on techniques that enable compatibility between multiple embedding models [1]. Some of these techniques specifically promote methods without joint or end-to-end learning phases. These are referred to as zero-shot stitching [2], [3]. They are particularly interesting for scaling and considering the native construction of a joint representation space.

Research Objectives:

·         Develop zero-shot stitching techniques in an operational context of agent communications across a network.

·         Design algorithms and protocols to establish native agent communications, considering the operational constraints of a communications network.

[1] Jha, R., Zhang, C., Shmatikov, V., & Morris, J. X. (2025). Harnessing the Universal Geometry of Embeddings. arXiv preprint arXiv:2505.12540. https://arxiv.org/abs/2505.12540

[2] Moschella, L., Maiorca, V., Fumero, M., Norelli, A., Locatello, F., & Rodolà, E. (2023). Relative representations enable zero-shot latent space communication. arXiv preprint arXiv:2209.15430. https://arxiv.org/abs/2209.15430

[3] García-Castellanos, A., Marchetti, G. L., Kragic, D., & Scolamiero, M. (2025). Relative Representations: Topological and Geometric Perspectives. arXiv preprint arXiv:2409.10967. https://arxiv.org/abs/2409.10967