Publications
How Far Do Simple Transformations Translate Across Text Embedding Models?
Sid Ali Hamideche, Louis-Adrien Dufrène, Quentin Lampin, Guillaume Larue
August 06, 2026 — arXiv preprint
We investigate whether simple transformations can translate representations across heterogeneous text embedding models. Understanding how independently trained models organize semantic information is an enabler for AI-to-AI latent communication without decoding into human-readable text. Focusing on lightweight translators such as linear mappings, we test the literature hypothesis of latent universality in a realistic text setting beyond simplified benchmarks. Across nine embedding models differing in architecture, pooling strategy, and training objective, we evaluate compatibility using CKA, downstream transfer, fidelity, and retrieval. Simple translators recover meaningful shared structure and support transfer for some compatible pairs, but fail sharply for others. Compatibility depends jointly on architecture, training objective, pooling, and data distribution. Overall, the results show that heterogeneous embedding spaces are not universally related by simple mappings as often suggested in some literature.
Learning Compression Rules for Network Traffic
Quentin Lampin, Éloi Sainte-Beuve, Louis-Adrien Dufrène, Guillaume Larue, Massih-Reza Amini
August 05, 2026 — arXiv preprint
We study the problem of learning compact rule-based compressors for structured network traffic. Each packet is a record of header fields that are highly redundant within a flow, and a compressor is a small set of rules matching such records and replacing predictable fields with short codes. We cast rule learning as a two-stage problem: (i) an unsupervised structure-discovery stage that recursively partitions training packets using a normalized entropy-ratio criterion robust to small samples, and (ii) a constrained selection stage that uses dynamic programming to pick the rule subset maximizing expected compression gain under a hard budget on the number of installable rules. We instantiate the framework on Static Context Header Compression (SCHC), the IETF standard for rule-based header compression in constrained networks, and evaluate it on four real-world Internet-of-Things and 5G core-network datasets. Our method, Robust Entropy Clustering for Adaptive comPression (RECAP), surpasses expert-engineered rule sets with a small number of learned rules and removes the need for manual rule design.
Learning High-Dimensional Parity Functions with Product Networks using Gradient Descent
Guillaume Larue, Louis-Adrien Dufrène, Quentin Lampin, Hadi Ghauch, Ghaya Rekaya
May 30, 2026 — ICML 2026 (accepted, to appear)
Parity functions are fundamental Boolean operations with critical applications across machine learning, cryptography, and error correction. Yet, learning high-dimensional parity functions poses significant challenges: in a general setting, standard neural network architectures typically require exponential sample complexity, making gradient-based optimization intractable for large number of inputs $N$. We demonstrate that compact product-based neural architectures combined with stochastic data sparsity (Bernoulli inputs with $p_e \leq 1/N$) and appropriate hyperparameter choice enable efficient parity learning, with theoretical guarantees of convergence. Experiments validate our theory across dimensions up to $N = 100{,}000$, with empirical evidence showing optimal hyperparameter choices for $p_e$ and learning rate $\alpha$, as well as polynomial complexity scaling laws. This work establishes fundamental connections between architectural inductive bias and data sparsity, opening new possibilities for neural arithmetic, structured reasoning, binary neural networks, and machine learning applied to automated protocol discovery.
Bridging the Data Gap: Digital Twin as a New Paradigm for AI-based Radio Sensing
Éloi Sainte-Beuve, Guillaume Larue, Louis-Adrien Dufrène, Quentin Lampin, Ali Al Khansa
April 01, 2026 — 34th European Conference on Signal Processing, EUSIPCO 2026 (accepted, to appear)
We present a methodology that places a 3D digital twin (DT) of the environment as the main enabler behind the development of radio sensing at scale. The DT acts as a world model, providing geometry, materials, and transmitter/receiver placements to a ray-tracing engine that generates time-indexed channel impulse responses (CIRs) for large numbers of plausible scenes (moving people and objects, layout variants, seasonal/weather conditions, etc). From these synthetic sequences, we train a sequential neural network that maps CIR time series to spatial occupancy estimates, enabling device-free localization (DFL) without instrumented targets. We posit that sensing is best approached as an environment-conditioned learning problem: rather than seeking a single global model, we advocate training or fine-tuning local models specialized to a site-specific DT. As a first experiment, we introduce a novel State Space Model architecture, trained and evaluated across multiple room geometries. The localization performances obtained demonstrate the potential of the approach.
Compacting Semantic Matryoshka Representations with Product Quantization
Leonardo Roque, Quentin Lampin, Louis-Adrien Dufrène, Guillaume Larue, Grégoire Lefebvre, Mohamad Assaad
December 08, 2025 — Workshop AI-Native Communications Systems — IEEE GLOBECOM 2025
This paper investigates efficient compression techniques for embedding representations in semantic communication systems. As communication networks evolve to support AI agent-based interactions, there is a growing need for methods that preserve semantic content while reducing transmission overhead. We explore Product Quantization (PQ) as a specialized compression approach for embeddings constructed using Matryoshka Representation Learning (MRL), which organizes semantic features hierarchically by relevance. This structure allows a reduction in model complexity by using truncated embeddings instead of full-size representations. Within a framework based on network-application cooperation and following to the Separation of Concerns principle, we analyze how PQ serves as an effective source coding strategy for semantically structured representations. Our comparative analysis demonstrates that PQ delivers better task accuracy performances and efficiency compared to simple truncation methods, especially when compressing MRL-based embeddings.
Target Handover in Distributed Integrated Sensing and Communication
Yu Ge et al.
August 12, 2025 — IEEE International Conference on Communications 2025, Montreal, Canada
The concept of 6G distributed integrated sensing and communications (DISAC) builds upon the functionality of integrated sensing and communications (ISAC) by integrating distributed architectures, significantly enhancing both sensing and communication coverage and performance. In 6G DISAC systems, tracking target trajectories requires base stations (BSs) to hand over their tracked targets to neighboring BSs. Determining what information to share, where, how, and when is critical to effective handover. This paper addresses the target handover challenge in DISAC systems and introduces a method enabling BSs to share essential target trajectory information at appropriate time steps, facilitating seamless handovers to other BSs. The target tracking problem is tackled using the standard trajectory Poisson multi-Bernoulli mixture (TPMBM) filter, enhanced with the proposed handover algorithm. Simulation results confirm the effectiveness of the implemented tracking solution.
AI-Native 6G Networks: the 6GARROW Integrated Device-Network Approach
Emilio Calvanese Strinati et al.
June 30, 2025 — 2025 Joint European Conference on Networks and Communications & 6G Summit (EuCNC/6G Summit), Poznan, Poland
This paper introduces the concept of seamlessly embedding AI-native Radio Access Networks (RANs) and integrated device-network approaches, aiming to extend AI capabilities across diverse devices e.g., smartphones, Internet of Things (IoT) sensors, autonomous vehicles and network infrastructures. By embedding AI deeply within RANs and fostering tight integration between devices and network architectures, we unlock unprecedented levels of intelligence, efficiency, flexibility, and performance in wireless communication systems. This integration enables devices and networks to collaboratively process, adapt, and optimize data in real-time, creating a synergistic AI-enabled ecosystem that forms the backbone of AI-native 6G. Unlike traditional wireless communication systems, which rely on static configurations and manual interventions, AI-native networks dynamically adapt to changing conditions, autonomously optimize resource allocation, and respond to diverse user demands and environmental factors. The 6GARROW project explores these transformative approaches, demonstrating how AI-native RANs and integrated device-network solutions deliver higher throughput, lower latency, and improved reliability. Beyond enhanced performance, this paradigm shift supports a diverse range of innovative services, including spatio-temporal communications, critical and compute-AI applications, omnipresent IoT, immersive user experiences, and global broadband connectivity. This paper outlines the vision, methodologies, and potential impact of the project, highlighting its role in advancing toward a more connected, intelligent, and adaptive wireless ecosystem for the future.
Learning Linear Block Codes with Gradient Quantization
Guillaume Larue, Louis-Adrien Dufrène, Quentin Lampin
March 20, 2025 — IEEE Transactions on Communications
This study investigates the problem of learning linear block codes optimized for Belief-Propagation decoders significantly improving performance compared to the state-of-the-art. Our previous research is extended with an enhanced system design that facilitates a more effective learning process for the parity check matrix. We simplify the input dataset, restrict the number of parameters to learn and improve the gradient back-propagation within the model. We also introduce novel optimizers specifically designed for discrete-valued weights. Based on conventional gradient computation, these optimizers provide discrete weights updates, enabling finer control and improving explainability of the learning process. Through these changes, we consistently achieve improved code performance, provided appropriately chosen hyper-parameters. To rigorously evaluate the performance of learned codes in the context of short to medium block lengths, we propose a comprehensive code performance assessment framework. This framework enables a fair comparison between our learning methodology and random search approaches, ensuring statistical significance in our results. The proposed model pave the way for a new approach to the efficient learning of linear block codes tailored to specific decoder structures.
Dynamic Power Allocation in OFDM ISAC for Time of Arrival Estimation
Ali Al Khansa, Giyyarpuram Madhusudan, Guillaume Larue, Louis-Adrien Dufrène
November 27, 2024 — 2024 IEEE Conference on Standards for Communications and Networking (CSCN)
Resource allocation in Integrated Sensing and Communication (ISAC) systems is critical for balancing communication and sensing performance. This paper introduces a novel dynamic power allocation strategy for Orthogonal Frequency Division Multiplexing (OFDM) ISAC systems, optimizing communication capacity while adhering to Peak Side-lobe Level (PSL) and sensing accuracy constraints, particularly for Time of Arrival (ToA) estimation. Unlike conventional methods that address either PSL or accuracy in isolation, our approach dynamically allocates power to satisfy both constraints. Additionally, it prioritizes communication when sensing performance is insufficient, avoiding any loss in communication capacity. Numerical results validate the importance of considering both sensing constraints and demonstrate the effectiveness of the proposed dynamic power allocation strategy.
Semantic Communications Services within Generalist Operated Networks
Quentin Lampin, Louis-Adrien Dufrène, Guillaume Larue
October 07, 2024 — 2024 IEEE 25th International Workshop on Signal Processing Advances in Wireless Communications (SPAWC)
This paper addresses the challenge of integrating semantic communication principles into operated networks, traditionally optimized based on network-centric metrics rather than application-specific needs. Operated networks strongly adhere to the principle of "separation of concerns", which emphasizes a clear distinction between network operation and application. Despite the initial perceived incompatibility between semantic communication and the principles of operated networks, this paper provides solutions to reconcile them. The foundations of these solutions include the adoption of non-arbitrary semantic representations as a standard encoding for communications, the establishment of a standard interface between the application and network, and a dedicated network control plane. These enable the application to describe the data typology and the nature of the task, and to agree upon a transmission scheme tailored to the supported task. Through three scenarios involving an application transmitting text representations, we illustrate the implementation of the proposal and demonstrate the potential of the approach.
AI Models for Digital Signal Processing in Future 6G-IoT Networks
Guillaume Larue
January 10, 2023 — Ph.D. dissertation, Institut Polytechnique de Paris
Wireless technologies are of paramount importance to today's societies and future 6th generation communication networks are expected to address many societal and technological challenges. While communications infrastructures have a growing environmental impact that needs to be reduced, digital technologies also have a role to play in reducing the impact of all sectors of the economy. To this end, the future networks will not only have to enable more efficient information transfer, but also meet the growing need for data exchange capacity. This is particularly the role of the Internet of Things use cases, where a massive number of sensors allow to monitor complex systems. These use cases are associated with many constraints such as limited energy resources and complexity. Therefore, an efficient and low-complexity physical layer - responsible for the transmission of information between the network nodes - is absolutely crucial. In this regard, the use of artificial intelligence techniques is relevant. On the one hand, the mathematical framework of neural networks allows for efficient and low-cost generic hardware implementations. On the other hand, the application of learning procedures can improve the performance of certain algorithms. In this work, we are interested in the use of neural networks and machine learning for digital signal processing in the context of 6G-IoT networks. First, we are interested in the transcription of certain equalisation, demodulation and decoding algorithms from the digital communications literature into neural networks. Secondly, we are interested in the application of learning mechanisms on these neural network structures in order to improve their performance. A linear block decoder is proposed which allows the blind discovery of a decoding scheme whose performance is at least equivalent to that of the reference decoder. Finally, an end-to-end structure is presented, allowing joint learning of an encoding/decoding scheme with performance and complexity comparable to state-of-the-art solutions.
Neural Belief Propagation Auto-Encoder for Linear Block Code Design
Guillaume Larue, Louis-Adrien Dufrène, Quentin Lampin, Hadi Ghauch, Ghaya Rekaya
September 21, 2022 — IEEE Transactions on Communications
The growing number of Internet of Thing (IoT) and Ultra-Reliable Low Latency Communications (URLCC) use cases in next generation communication networks calls for the development of efficient Forward Error Correction (FEC) mechanisms. These use cases usually imply using short to mid-sized information blocks and requires low-complexity and/or fast decoding procedures. This paper investigates the joint learning of short to mid block-length coding schemes and associated Belief-Propagation (BP) like decoders using Machine Learning (ML) techniques. An interpretable auto-encoder (AE) architecture is proposed, ensuring scalability to block sizes currently challenging for ML-based linear block code design approaches. By optimizing a coding scheme w.r.t. the targeted decoder, the proposed system offers a good complexity/performance trade-off compared to various codes from literature with length up to 128 bits.
Pervasive Artificial Intelligence in Next Generation Wireless: The Hexa-X Project Perspective
Miltiadis C. Filippou et al.
July 21, 2022 — AI6G’22: First International Workshop on Artificial Intelligence in beyond 5G and 6G Wireless Networks
The European 6G flagship project Hexa-X has the objective to conduct exploratory research on the next generation of mobile networks with the intention to connect human, physical and digital worlds with a fabric of technology enablers. Within this scope, one of the main research challenges is the ambition for beyond 5G (B5G)/6G systems to support, enhance and enable real-time trustworthy control by transforming Artificial Intelligence (AI) / Machine Learning (ML) technologies into a vital and trusted tool for large-scale deployment of interconnected intelligence available to the wider society. Hence, the study and development of concepts and solutions enabling AI-driven communication and computation co-design for a B5G /6G communication system is required. This paper focuses on describing the possibilities that emerge with the application of AI & ML mechanisms (with emphasis on ML) to 6G networks, identifying the resulting challenges and proposing some potential solution approaches.
Blind Neural Belief Propagation Decoder for Linear Block Codes
Guillaume Larue, Louis-Adrien Dufrène, Quentin Lampin, Paul Chollet, Hadi Ghauch, Ghaya Rekaya
June 11, 2021 — 2021 Joint European Conference on Networks and Communications & 6G Summit (EuCNC/6G Summit)
Neural belief propagation decoders were recently introduced by Nachmani et al. as a way to improve the decoding performance of belief propagation iterative algorithm for short to medium length linear block codes. The main idea behind these decoders is to represent belief propagation as a neural network, enabling adaptive weighting of the decoding process. In the present paper an efficient recurrent neural network architecture, based on gating and weights sharing mechanisms, is proposed to perform blind neural belief propagation decoding without prior knowledge of the coding scheme used by the encoder. The proposed architecture is able to learn to decode BCH (15,11) and BCH (15,7) codes and significantly improves the decoding performance over a standard belief propagation algorithm. A particular emphasis is given to the interpretability and complexity of the proposed model to ensure scalability to larger codes.
Low-Complexity Neural Networks for Baseband Signal Processing
Guillaume Larue et al.
December 11, 2020 — 2020 IEEE Globecom Workshops, Taipei, Taiwan
This study investigates the use of neural networks for the physical layer in the context of Internet of Things. In such systems, devices face challenging energy, computational and cost constraints that advocate for a low-complexity baseband signal processing. In this work, low-complexity neural networks are proposed as promising candidates. They present adaptability to operating conditions, high performance to complexity ratio and also offer a good explainability, crucial in most communications systems that cannot rely on "black-box" solutions. Moreover, recent advances in dedicated hardware for neural networks processing bring new perspectives in terms of efficiency and flexibility that motivate their use at the physical layer. To illustrate how classical baseband signal processing algorithms can be translated to minimal neural networks, two models are proposed in this paper to realize single-path equalization and demodulation of M-QAM signals. These models are assessed using both simulation and experimentation and achieve near optimal performances.