Deniz Gündüz

Yang Wang, Zhen Gao, Dezhi Zheng et al.

It is anticipated that 6G wireless networks will accelerate the convergence of the physical and cyber worlds and enable a paradigm-shift in the way we deploy and exploit communication networks. Machine learning, in particular deep learning (DL), is expected to be one of the key technological enablers of 6G by offering a new paradigm for the design and optimization of networks with a high level of intelligence. In this article, we introduce an emerging DL architecture, known as the transformer, and discuss its potential impact on 6G network design. We first discuss the differences between the transformer and classical DL architectures, and emphasize the transformer's self-attention mechanism and strong representation capabilities, which make it particularly appealing for tackling various challenges in wireless network design. Specifically, we propose transformer-based solutions for various massive multiple-input multiple-output (MIMO) and semantic communication problems, and show their superiority compared to other architectures. Finally, we discuss key challenges and open issues in transformer-based solutions, and identify future research directions for their deployment in intelligent 6G networks.

Michal Yemini, Rajarshi Saha, Emre Ozfatura et al.

We present a semi-decentralized federated learning algorithm wherein clients collaborate by relaying their neighbors' local updates to a central parameter server (PS). At every communication round to the PS, each client computes a local consensus of the updates from its neighboring clients and eventually transmits a weighted average of its own update and those of its neighbors to the PS. We appropriately optimize these averaging weights to ensure that the global update at the PS is unbiased and to reduce the variance of the global update at the PS, consequently improving the rate of convergence. Numerical simulations substantiate our theoretical claims and demonstrate settings with intermittent connectivity between the clients and the PS, where our proposed algorithm shows an improved convergence rate and accuracy in comparison with the federated averaging algorithm.

Francesco Pase, Deniz Gündüz, Michele Zorzi

We consider a rate-constrained contextual multi-armed bandit (RC-CMAB) problem, in which a group of agents are solving the same contextual multi-armed bandit (CMAB) problem. However, the contexts are observed by a remotely connected entity, i.e., the decision-maker, that updates the policy to maximize the returned rewards, and communicates the arms to be sampled by the agents to a controller over a rate-limited communications channel. This framework can be applied to personalized ad placement, whenever the content owner observes the website visitors, and hence has the context, but needs to transmit the ads to be shown to a controller that is in charge of placing the marketing content. Consequently, the rate-constrained CMAB (RC-CMAB) problem requires the study of lossy compression schemes for the policy to be employed whenever the constraint on the channel rate does not allow the uncompressed transmission of the decision-maker's intentions. We characterize the fundamental information theoretic limits of this problem by letting the number of agents go to infinity, and study the regret that can be achieved, identifying the two distinct rate regions leading to linear and sub-linear regrets respectively. We then analyze the optimal compression scheme achievable in the limit with infinite agents, when using the forward and reverse KL divergence as distortion metric. Based on this, we also propose a practical coding scheme, and provide numerical results.

Ozan Aygün, Mohammad Kazemi, Deniz Gündüz et al.

We examine federated learning (FL) with over-the-air (OTA) aggregation, where mobile users (MUs) aim to reach a consensus on a global model with the help of a parameter server (PS) that aggregates the local gradients. In OTA FL, MUs train their models using local data at every training round and transmit their gradients simultaneously using the same frequency band in an uncoded fashion. Based on the received signal of the superposed gradients, the PS performs a global model update. While the OTA FL has a significantly decreased communication cost, it is susceptible to adverse channel effects and noise. Employing multiple antennas at the receiver side can reduce these effects, yet the path-loss is still a limiting factor for users located far away from the PS. To ameliorate this issue, in this paper, we propose a wireless-based hierarchical FL scheme that uses intermediate servers (ISs) to form clusters at the areas where the MUs are more densely located. Our scheme utilizes OTA cluster aggregations for the communication of the MUs with their corresponding IS, and OTA global aggregations from the ISs to the PS. We present a convergence analysis for the proposed algorithm, and show through numerical evaluations of the derived analytical expressions and experimental results that utilizing ISs results in a faster convergence and a better performance than the OTA FL alone while using less transmit power. We also validate the results on the performance using different number of cluster iterations with different datasets and data distributions. We conclude that the best choice of cluster aggregations depends on the data distribution among the MUs and the clusters.

Bowen Xie, Yuxuan Sun, Sheng Zhou et al.

Federated learning (FL) is a promising approach to enable the future Internet of vehicles consisting of intelligent connected vehicles (ICVs) with powerful sensing, computing and communication capabilities. We consider a base station (BS) coordinating nearby ICVs to train a neural network in a collaborative yet distributed manner, in order to limit data traffic and privacy leakage. However, due to the mobility of vehicles, the connections between the BS and ICVs are short-lived, which affects the resource utilization of ICVs, and thus, the convergence speed of the training process. In this paper, we propose an accelerated FL-ICV framework, by optimizing the duration of each training round and the number of local iterations, for better convergence performance of FL. We propose a mobility-aware optimization algorithm called MOB-FL, which aims at maximizing the resource utilization of ICVs under short-lived wireless connections, so as to increase the convergence speed. Simulation results based on the beam selection and the trajectory prediction tasks verify the effectiveness of the proposed solution.

Omar J. Faqir, Eric C. Kerrigan, Deniz Gündüz

Communication energy in a wireless network of mobile autonomous agents should be considered as the sum of transmission energy and propulsion energy used to facilitate the transfer of information. Accordingly, communication-theoretic and Newtonian dynamic models are developed to model the communication and locomotion expenditures of each node. These are subsequently used to formulate a novel nonlinear optimal control problem (OCP) over a network of autonomous nodes. It is then shown that, under certain conditions, the OCP can be transformed into an equivalent convex form. Numerical results for a single link between a node and access point allow for comparison with known solutions before the framework is applied to a multiple-node UAV network, for which previous results are not readily extended. Simulations show that transmission energy can be of the same order of magnitude as propulsion energy allowing for possible savings, whilst also exemplifying how speed adaptations together with power control may increase the network throughput.

Jiakang Chen, Di You, Deniz Gündüz et al.

Joint source-channel coding schemes based on deep neural networks (DeepJSCC) have recently achieved remarkable performance for wireless image transmission. However, these methods usually focus only on the distortion of the reconstructed signal at the receiver side with respect to the source at the transmitter side, rather than the perceptual quality of the reconstruction which carries more semantic information. As a result, severe perceptual distortion can be introduced under extreme conditions such as low bandwidth and low signal-to-noise ratio. In this work, we propose CommIN, which views the recovery of high-quality source images from degraded reconstructions as an inverse problem. To address this, CommIN combines Invertible Neural Networks (INN) with diffusion models, aiming for superior perceptual quality. Through experiments, we show that our CommIN significantly improves the perceptual quality compared to DeepJSCC under extreme conditions and outperforms other inverse problem approaches used in DeepJSCC.

Deniz Gündüz, Michèle A. Wigger, Tze-Yang Tung et al.

Semantic- and task-oriented communication has emerged as a promising approach to reducing the latency and bandwidth requirements of next-generation mobile networks by transmitting only the most relevant information needed to complete a specific task at the receiver. This is particularly advantageous for machine-oriented communication of high data rate content, such as images and videos, where the goal is rapid and accurate inference, rather than perfect signal reconstruction. While semantic- and task-oriented compression can be implemented in conventional communication systems, joint source-channel coding (JSCC) offers an alternative end-to-end approach by optimizing compression and channel coding together, or even directly mapping the source signal to the modulated waveform. Although all digital communication systems today rely on separation, thanks to its modularity, JSCC is known to achieve higher performance in finite blocklength scenarios, and to avoid cliff and the levelling-off effects in time-varying channel scenarios. This article provides an overview of the information theoretic foundations of JSCC, surveys practical JSCC designs over the decades, and discusses the reasons for their limited adoption in practical systems. We then examine the recent resurgence of JSCC, driven by the integration of deep learning techniques, particularly through DeepJSCC, highlighting its many surprising advantages in various scenarios. Finally, we discuss why it may be time to reconsider today's strictly separate architectures, and reintroduce JSCC to enable high-fidelity, low-latency communications in critical applications such as autonomous driving, drone surveillance, or wearable systems.

Francesc Wilhelmi, Jernej Hribar, Selim F. Yilmaz et al.

As wireless standards evolve, more complex functionalities are introduced to address the increasing requirements in terms of throughput, latency, security, and efficiency. To unleash the potential of such new features, artificial intelligence (AI) and machine learning (ML) are currently being exploited for deriving models and protocols from data, rather than by hand-programming. In this paper, we explore the feasibility of applying ML in next-generation wireless local area networks (WLANs). More specifically, we focus on the IEEE 802.11ax spatial reuse (SR) problem and predict its performance through federated learning (FL) models. The set of FL solutions overviewed in this work is part of the 2021 International Telecommunication Union (ITU) AI for 5G Challenge.

Milad Ganjalizadeh, Hossein Shokri Ghadikolaei, Deniz Gündüz et al.

Recent advances in distributed intelligence have driven impressive progress across a diverse range of applications, from industrial automation to autonomous transportation. Nevertheless, deploying distributed learning services over wireless networks poses numerous challenges. These arise from inherent uncertainties in wireless environments (e.g., random channel fluctuations), limited resources (e.g., bandwidth and transmit power), and the presence of coexisting services on the network. In this paper, we investigate a mixed service scenario wherein high-priority ultra-reliable low latency communication (URLLC) and low-priority distributed learning services run concurrently over a network. Utilizing device selection, we aim to minimize the convergence time of distributed learning while simultaneously fulfilling the requirements of the URLLC service. We formulate this problem as a Markov decision process and address it via BSAC-CoEx, a framework based on the branching soft actor-critic (BSAC) algorithm that determines each device's participation decision through distinct branches in the actor's neural network. We evaluate our solution with a realistic simulator that is compliant with 3GPP standards for factory automation use cases. Our simulation results confirm that our solution can significantly decrease the training delays of the distributed learning service while keeping the URLLC availability above its required threshold and close to the scenario where URLLC solely consumes all wireless resources.

Seyyed Amirhossein Ameli Kalkhoran, Mehdi Letafati, Ecenaz Erdemir et al.

In this paper, a generalization of deep learning-aided joint source channel coding (Deep-JSCC) approach to secure communications is studied. We propose an end-to-end (E2E) learning-based approach for secure communication against multiple eavesdroppers over complex-valued fading channels. Both scenarios of colluding and non-colluding eavesdroppers are studied. For the colluding strategy, eavesdroppers share their logits to collaboratively infer private attributes based on ensemble learning method, while for the non-colluding setup they act alone. The goal is to prevent eavesdroppers from inferring private (sensitive) information about the transmitted images, while delivering the images to a legitimate receiver with minimum distortion. By generalizing the ideas of privacy funnel and wiretap channel coding, the trade-off between the image recovery at the legitimate node and the information leakage to the eavesdroppers is characterized. To solve this secrecy funnel framework, we implement deep neural networks (DNNs) to realize a data-driven secure communication scheme, without relying on a specific data distribution. Simulations over CIFAR-10 dataset verifies the secrecy-utility trade-off. Adversarial accuracy of eavesdroppers are also studied over Rayleigh fading, Nakagami-m, and AWGN channels to verify the generalization of the proposed scheme. Our experiments show that employing the proposed secure neural encoding can decrease the adversarial accuracy by 28%.

Mehdi Sattari, Deniz Gündüz, Tommy Svensson

Efficient channel state information (CSI) compression is essential in frequency division duplexing (FDD) massive multiple-input multiple-output (MIMO) systems due to the substantial feedback overhead. Recently, deep learning-based compression techniques have demonstrated superior performance for CSI feedback. However, their performance often degrades under distribution shifts across wireless environments, largely due to limited generalization capability. To address this challenge, we consider a full-model fine-tuning scheme, in which both the encoder and decoder are jointly updated using a small number of recent CSI samples from the target environment. A key challenge in this setting is the transmission of updated decoder parameters to the receiver, which introduces additional communication overhead. To mitigate this bottleneck, we explicitly incorporate the bit rate of model updates into the fine-tuning objective and entropy-code the model updates jointly with the compressed CSI. Furthermore, we employ a structured prior that promotes sparse and selective parameter updates, thereby significantly reducing the model-update communication cost. Simulation results across multiple CSI datasets demonstrate that full-model fine-tuning substantially improves the rate-distortion performance of neural CSI compression, despite the additional cost of model updates. We further analyze the impact of the evaluation horizon, the quantization resolution of model updates, and the size of the target-domain dataset on the overall feedback efficiency.

Ruiqing Mao, Haotian Wu, Yukuan Jia et al.

Collaborative perception (CP) is emerging as a promising solution to the inherent limitations of stand-alone intelligence. However, current wireless communication systems are unable to support feature-level and raw-level collaborative algorithms due to their enormous bandwidth demands. In this paper, we propose DiffCP, a novel CP paradigm that utilizes a specialized diffusion model to efficiently compress the sensing information of collaborators. By incorporating both geometric and semantic conditions into the generative model, DiffCP enables feature-level collaboration with an ultra-low communication cost, advancing the practical implementation of CP systems. This paradigm can be seamlessly integrated into existing CP algorithms to enhance a wide range of downstream tasks. Through extensive experimentation, we investigate the trade-offs between communication, computation, and performance. Numerical results demonstrate that DiffCP can significantly reduce communication costs by 14.5-fold while maintaining the same performance as the state-of-the-art algorithm.

Yanxiao Liu, Yijun Fan, Deniz Gündüz

Change of measure inequalities translate divergences between probability measures into explicit bounds on event probabilities, and play an important role in deriving probabilistic guarantees in learning theory, information theory, and statistics. We propose novel change of measure inequalities via a unified framework based on the data processing inequality, which is surprisingly elementary yet powerful enough to yield novel, tighter inequalities. We provide change of measure inequalities in terms of a broad family of information measures, including $f$-divergences (with Kullback-Leibler divergence and $χ^2$-divergence as special cases), Rényi divergence, and $α$-mutual information (with maximal leakage as a special case). We apply these results to generalization error analysis, PAC-Bayesian theory, differential privacy, and data memorization, obtaining stronger guarantees while recovering best-known results through simplified analyses.

Lorenzo Mario Amorosa, Marco Skocaj, Roberto Verdone et al.

The ever-increasing demand for high-quality and heterogeneous wireless communication services has driven extensive research on dynamic optimization strategies in wireless networks. Among several possible approaches, multi-agent deep reinforcement learning (MADRL) has emerged as a promising method to address a wide range of complex optimization problems like power control. However, the seamless application of MADRL to a variety of network optimization problems faces several challenges related to convergence. In this paper, we present the use of graphs as communication-inducing structures among distributed agents as an effective means to mitigate these challenges. Specifically, we harness graph neural networks (GNNs) as neural architectures for policy parameterization to introduce a relational inductive bias in the collective decision-making process. Most importantly, we focus on modeling the dynamic interactions among sets of neighboring agents through the introduction of innovative methods for defining a graph-induced framework for integrated communication and learning. Finally, the superior generalization capabilities of the proposed methodology to larger networks and to networks with different user categories is verified through simulations.

Bohan Li, Yiming Liu, Xueyan Niu et al.

Diffusion models have achieved remarkable success in generating high quality image and video data. More recently, they have also been used for image compression with high perceptual quality. In this paper, we present a novel approach to extreme video compression leveraging the predictive power of diffusion-based generative models at the decoder. The conditional diffusion model takes several neural compressed frames and generates subsequent frames. When the reconstruction quality drops below the desired level, new frames are encoded to restart prediction. The entire video is sequentially encoded to achieve a visually pleasing reconstruction, considering perceptual quality metrics such as the learned perceptual image patch similarity (LPIPS) and the Frechet video distance (FVD), at bit rates as low as 0.02 bits per pixel (bpp). Experimental results demonstrate the effectiveness of the proposed scheme compared to standard codecs such as H.264 and H.265 in the low bpp regime. The results showcase the potential of exploiting the temporal relations in video data using generative models. Code is available at: https://github.com/ElesionKyrie/Extreme-Video-Compression-With-Prediction-Using-Pre-trainded-Diffusion-Models-

Taowen Liu, Marta Andronic, Deniz Gündüz et al.

LLM training is resource-intensive. Quantized training improves computational and memory efficiency but introduces quantization noise, which can hinder convergence and degrade model accuracy. Stochastic Rounding (SR) has emerged as a theoretically attractive alternative to deterministic rounding, offering unbiased gradient estimates. However, its interaction with other training factors -- especially batch size -- remains under explored. In this paper, we present a theoretical and empirical study of mini-batch stochastic gradient descent (SGD) with SR, showing that increased batch sizes can compensate for reduced precision during back-propagation. Furthermore, we show that quantizing weights and activations impacts gradient variance in distinct ways. Our experiments validate these theoretical insights.

Yanxiao Liu, Chun Hei Michael Shiu, Lele Wang et al.

We study the generalization error of stochastic learning algorithms from an information-theoretic perspective, with a particular emphasis on deriving sharper bounds for differentially private algorithms. It is well known that the generalization error of stochastic learning algorithms can be bounded in terms of mutual information and maximal leakage, yielding in-expectation and high-probability guarantees, respectively. In this work, we further upper bound mutual information and maximal leakage by explicit, easily computable formulas, using typicality-based arguments and exploiting the stability properties of private algorithms. In the first part of the paper, we strictly improve the mutual-information bounds by Rodríguez-Gálvez et al. (IEEE Trans. Inf. Theory, 2021). In the second part, we derive new upper bounds on the maximal leakage of learning algorithms. In both cases, the resulting bounds on information measures translate directly into generalization error guarantees.

Jingyuan Xia, Zhixiong Yang, Shengxi Li et al.

Learning-based approaches have witnessed great successes in blind single image super-resolution (SISR) tasks, however, handcrafted kernel priors and learning based kernel priors are typically required. In this paper, we propose a Meta-learning and Markov Chain Monte Carlo (MCMC) based SISR approach to learn kernel priors from organized randomness. In concrete, a lightweight network is adopted as kernel generator, and is optimized via learning from the MCMC simulation on random Gaussian distributions. This procedure provides an approximation for the rational blur kernel, and introduces a network-level Langevin dynamics into SISR optimization processes, which contributes to preventing bad local optimal solutions for kernel estimation. Meanwhile, a meta-learning-based alternating optimization procedure is proposed to optimize the kernel generator and image restorer, respectively. In contrast to the conventional alternating minimization strategy, a meta-learning-based framework is applied to learn an adaptive optimization strategy, which is less-greedy and results in better convergence performance. These two procedures are iteratively processed in a plug-and-play fashion, for the first time, realizing a learning-based but plug-and-play blind SISR solution in unsupervised inference. Extensive simulations demonstrate the superior performance and generalization ability of the proposed approach when comparing with state-of-the-arts on synthesis and real-world datasets. The code is available at https://github.com/XYLGroup/MLMC.

Mohammad Malekzadeh, Burak Hasircioglu, Nitish Mital et al.

While rich medical datasets are hosted in hospitals distributed across the world, concerns on patients' privacy is a barrier against using such data to train deep neural networks (DNNs) for medical diagnostics. We propose Dopamine, a system to train DNNs on distributed datasets, which employs federated learning (FL) with differentially-private stochastic gradient descent (DPSGD), and, in combination with secure aggregation, can establish a better trade-off between differential privacy (DP) guarantee and DNN's accuracy than other approaches. Results on a diabetic retinopathy~(DR) task show that Dopamine provides a DP guarantee close to the centralized training counterpart, while achieving a better classification accuracy than FL with parallel DP where DPSGD is applied without coordination. Code is available at https://github.com/ipc-lab/private-ml-for-health.

Gongpu Chen, Deniz Gündüz

We study a decentralized linear quadratic Gaussian control problem, in which a leader and a follower must steer a linear system to a target state. The target state is known only to the leader, and no explicit communication channel exists between the agents. To address the challenge posed by this asymmetric information structure, we propose an integrated communication and control (ICoCo) framework in which the control plant itself serves as a communication channel: the leader encodes the target state into its control input through an additive communication term, and the follower decodes it from the resulting state trajectory. We design an implicit coordination scheme based on joint source-channel coding ideas, and prove that the follower's estimation error decreases monotonically to zero, enabling the two agents to coordinate increasingly well and ultimately steer the system to the target state. We then formulate the design of the communication power as an optimal control problem to minimize the overall control cost. In the fully actuated leader case, we derive necessary optimality conditions and in the under-actuated case, we solve the problem numerically. Numerical results show that the proposed scheme effectively coordinates the two agents and achieves a control cost close to that of the explicit-communication lower bound.

Szymon Kobus, Deniz Gündüz

We address the challenge of remote control where one or more actors, lacking direct reward access, are steered by a controller over a communication-constrained channel. The controller learns an optimal policy from observed rewards and communicates action guidance to the actors, which becomes demanding for large or continuous action spaces. To achieve rate-efficient communication throughout this interactive learning and control process, we introduce a novel framework leveraging remote generation. Instead of transmitting full action specifications, the controller sends minimal information, enabling the actors to locally generate actions by sampling from the controller's evolving target policy. This guided sampling is facilitated by an importance sampling approach. Concurrently, the actors use the received guidance as supervised learning data to learn the controller's policy. This actor-side learning improves their local sampling capabilities, progressively reducing future communication needs. Our solution, Guided Remote Action Sampling Policy (GRASP), demonstrates significant communication reduction, achieving an average 12-fold data reduction across all experiments (50-fold for continuous action spaces) compared to direct action transmission, and a 41-fold reduction compared to reward transmission.

Antonio Tarizzo, Mohammad Kazemi, Deniz Gündüz

Federated edge learning (FEEL) enables wireless devices to collaboratively train a centralised model without sharing raw data, but repeated uplink transmission of model updates makes communication the dominant bottleneck. Over-the-air (OTA) aggregation alleviates this by exploiting the superposition property of the wireless channel, enabling simultaneous transmission and merging communication with computation. Digital OTA schemes extend this principle by incorporating the robustness of conventional digital communication, but current designs remain limited in low signal-to-noise ratio (SNR) regimes. This work proposes a learned digital OTA framework that improves recovery accuracy, convergence behaviour, and robustness to challenging SNR conditions while maintaining the same uplink overhead as state-of-the-art methods. The design integrates an unsourced random access (URA) codebook with vector quantisation and AMP-DA-Net, an unrolled approximate message passing (AMP)-style decoder trained end-to-end with the digital codebook and parameter server local training statistics. The proposed design extends OTA aggregation beyond averaging to a broad class of symmetric functions, including trimmed means and majority-based rules. Experiments on highly heterogeneous device datasets and varying numbers of active devices show that the proposed design extends reliable digital OTA operation by more than 10 dB into low SNR regimes while matching or improving performance across the full SNR range. The learned decoder remains effective under message corruption and nonlinear aggregation, highlighting the broader potential of end-to-end learned design for digital OTA communication in FEEL.

Waleed El-Geresy, Deniz Gündüz

We propose the use of adversarial wireless jamming to regularise the latent space of an autoencoder to match a diagonal Gaussian distribution. We consider the minimisation of a mean squared error distortion, where a jammer attempts to disrupt the recovery of a Gaussian source encoded and transmitted over the adversarial channel. A straightforward consequence of existing theoretical results is the fact that the saddle point of a minimax game - involving such an encoder, its corresponding decoder, and an adversarial jammer - consists of diagonal Gaussian noise output by the jammer. We use this result as inspiration for a novel approach to distribution matching in the latent space, utilising jamming as an auxiliary objective to encourage the aggregated latent posterior to match a diagonal Gaussian distribution. Using this new technique, we achieve distribution matching comparable to standard variational autoencoders and to Wasserstein autoencoders. This approach can also be generalised to other latent distributions.

Jiakang Chen, Selim F. Yilmaz, Di You et al.

Joint source-channel coding systems based on deep neural networks (DeepJSCC) have recently demonstrated remarkable performance in wireless image transmission. Existing methods primarily focus on minimizing distortion between the transmitted image and the reconstructed version at the receiver, often overlooking perceptual quality. This can lead to severe perceptual degradation when transmitting images under extreme conditions, such as low bandwidth compression ratios (BCRs) and low signal-to-noise ratios (SNRs). In this work, we propose SING, a novel two-stage JSCC framework that formulates the recovery of high-quality source images from corrupted reconstructions as an inverse problem. Depending on the availability of information about the DeepJSCC encoder/decoder and the channel at the receiver, SING can either approximate the stochastic degradation as a linear transformation, or leverage invertible neural networks (INNs) for precise modeling. Both approaches enable the seamless integration of diffusion models into the reconstruction process, enhancing perceptual quality. Experimental results demonstrate that SING outperforms DeepJSCC and other approaches, delivering superior perceptual quality even under extremely challenging conditions, including scenarios with significant distribution mismatches between the training and test data.

Maximilian Egger, Rawad Bitar, Antonia Wachter-Zeh et al.

We address the prominent communication bottleneck in federated learning (FL). We specifically consider stochastic FL, in which models or compressed model updates are specified by distributions rather than deterministic parameters. Stochastic FL offers a principled approach to compression, and has been shown to reduce the communication load under perfect downlink transmission from the federator to the clients. However, in practice, both the uplink and downlink communications are constrained. We show that bi-directional compression for stochastic FL has inherent challenges, which we address by introducing BICompFL. Our BICompFL is experimentally shown to reduce the communication cost by an order of magnitude compared to multiple benchmarks, while maintaining state-of-the-art accuracies. Theoretically, we study the communication cost of BICompFL through a new analysis of an importance-sampling based technique, which exposes the interplay between uplink and downlink communication costs.

Arpan Mukherjee, Marcello Bullo, Debabrota Basu et al.

While test-time scaling with verification has shown promise in improving the performance of large language models (LLMs), the role of the verifier and its imperfections remain underexplored. The effect of verification manifests through interactions of three quantities: (i) the generator's coverage, (ii) the verifier's region of convergence (ROC), and (iii) the sampling algorithm's sub-optimality. Though recent studies capture subsets of these factors, a unified framework quantifying the geometry of their interplay is missing. We frame verifiable test-time scaling as a transport problem. This characterizes the interaction of coverage, ROC, and sub-optimality, and uncovers that the sub-optimality--coverage curve exhibits three regimes. A transport regime -- where sub-optimality increases with coverage, a policy improvement regime -- where sub-optimality may decrease with coverage, depending on the verifier's ROC, and a saturation regime -- where sub-optimality plateaus, unaffected by coverage. We further propose and analyze two classes of sampling algorithms -- sequential and batched, and examine how their computational complexities shape these trade-offs. Empirical results with Qwen, Llama, and Gemma models corroborate our theoretical findings.

Didrik Bergström, Deniz Gündüz, Onur Günlü

We consider image transmission via deep joint source-channel coding (DeepJSCC) over multi-hop additive white Gaussian noise (AWGN) channels by training a DeepJSCC encoder-decoder pair with a pre-trained deep hash distillation (DHD) module to semantically cluster images, facilitating security-oriented applications through enhanced semantic consistency and improving the perceptual reconstruction quality. We train the DeepJSCC module to both reduce mean square error (MSE) and minimize cosine distance between DHD hashes of source and reconstructed images. Significantly improved perceptual quality as a result of semantic alignment is illustrated for different multi-hop settings, for which classical DeepJSCC may suffer from noise accumulation, measured by the learned perceptual image patch similarity (LPIPS) metric.

Antonio Tarizzo, Mohammad Kazemi, Deniz Gündüz

Federated edge learning (FEEL) enables distributed model training across wireless devices without centralising raw data, but deployment is constrained by the wireless uplink. A promising direction is over-the-air (OTA) aggregation, which merges communication with computation. Existing digital OTA methods can achieve either strong convergence or robustness to noise, but struggle to achieve both simultaneously, limiting performance in low signal-to-noise ratios (SNRs) where many IoT devices operate. This work proposes a learnt digital OTA framework that extends reliable operation into low-SNR conditions while maintaining the same uplink overhead as state-of-the-art. The proposed method combines an unrolled decoder with a jointly learnt unsourced random access codebook. Results show an extension of reliable operation by more than 7 dB, with improved global model convergence across all SNR levels, highlighting the potential of learning-based design for FEEL.

Arpan Mukherjee, Marcello Bullo, Deniz Gündüz

Uniform-reward reinforcement learning from human feedback (RLHF), which trains a single reward model to represent the preferences of all annotators, fails to capture the diversity of opinions across sub-populations, inadvertently favoring dominant groups. The state-of-the-art, MaxMin-RLHF, addresses this by learning group-specific reward models, and by optimizing for the group receiving the minimum reward, thereby promoting fairness. However, we identify that a key limitation of MaxMin-RLHF is its poor performance when the minimum-reward group is a minority. To mitigate this drawback, we introduce a novel framework, termed {\em SharedRep-RLHF}. At its core, SharedRep-RLHF learns and leverages {\em shared traits} in annotations among various groups, in contrast to learning separate reward models across groups. We first show that MaxMin-RLHF is provably suboptimal in learning shared traits, and then quantify the sample complexity of SharedRep-RLHF. Experiments across diverse natural language tasks showcase the effectiveness of SharedRep-RLHF compared to MaxMin-RLHF with a gain of up to 20% in win rate.

Szymon Kobus, Deniz Gündüz

Speculative decoding accelerates large language model inference using a smaller draft model. In this paper, we establish a surprising connection between speculative decoding and channel simulation, which aims at simulating a noisy channel using as few bits as possible. This connection allows us to provide an information-theoretic analysis of the speed up that can be achieved by speculative decoding. Leveraging this link, we derive an explicit relation between generation speed-up and the number of tokens $k$ generated by the draft model for large $k$, which serves as an upper bound for all $k$. We also propose a novel speculative decoding method via exponential race ERSD that matches state-of-the-art performance.

Waleed El-Geresy, Christos Papavassiliou, Deniz Gündüz

In this paper, we examine the problem of information storage on memristors affected by resistive drift noise under energy constraints. We introduce a novel, fundamental trade-off between the information lifetime of memristive states and the energy that must be expended to bring the device into a particular state. We then treat the storage problem as one of communication over a noisy, energy-constrained channel, and propose a joint source-channel coding (JSCC) approach to storing images in an analogue fashion. To design an encoding scheme for natural images and to model the memristive channel, we make use of data-driven techniques from the field of deep learning for communications, namely deep joint source-channel coding (DeepJSCC), employing a generative model of resistive drift as a computationally tractable differentiable channel model for end-to-end optimisation. We introduce a modified version of generalised divisive normalisation (GDN), a biologically inspired form of normalisation, that we call conditional GDN (cGDN), allowing for conditioning on continuous channel characteristics, including the initial resistive state and the delay between storage and reading. Our results show that the delay-conditioned network is able to learn an energy-aware coding scheme that achieves a higher and more balanced reconstruction quality across a range of storage delays.

Marcello Bullo, Seifallah Jardak, Pietro Carnelli et al.

The growing demand for intelligent applications beyond the network edge, coupled with the need for sustainable operation, are driving the seamless integration of deep learning (DL) algorithms into energy-limited, and even energy-harvesting end-devices. However, the stochastic nature of ambient energy sources often results in insufficient harvesting rates, failing to meet the energy requirements for inference and causing significant performance degradation in energy-agnostic systems. To address this problem, we consider an on-device adaptive inference system equipped with an energy-harvester and finite-capacity energy storage. We then allow the device to reduce the run-time execution cost on-demand, by either switching between differently-sized neural networks, referred to as multi-model selection (MMS), or by enabling earlier predictions at intermediate layers, called early exiting (EE). The model to be employed, or the exit point is then dynamically chosen based on the energy storage and harvesting process states. We also study the efficacy of integrating the prediction confidence into the decision-making process. We derive a principled policy with theoretical guarantees for confidence-aware and -agnostic controllers. Moreover, in multi-exit networks, we study the advantages of taking decisions incrementally, exit-by-exit, by designing a lightweight reinforcement learning-based controller. Experimental results show that, as the rate of the ambient energy increases, energy- and confidence-aware control schemes show approximately 5% improvement in accuracy compared to their energy-aware confidence-agnostic counterparts. Incremental approaches achieve even higher accuracy, particularly when the energy storage capacity is limited relative to the energy consumption of the inference model.

Tan Chen, Jintao Yan, Yuxuan Sun et al.

Hierarchical federated learning (HFL) enables distributed training of models across multiple devices with the help of several edge servers and a cloud edge server in a privacy-preserving manner. In this paper, we consider HFL with highly mobile devices, mainly targeting at vehicular networks. Through convergence analysis, we show that mobility influences the convergence speed by both fusing the edge data and shuffling the edge models. While mobility is usually considered as a challenge from the perspective of communication, we prove that it increases the convergence speed of HFL with edge-level heterogeneous data, since more diverse data can be incorporated. Furthermore, we demonstrate that a higher speed leads to faster convergence, since it accelerates the fusion of data. Simulation results show that mobility increases the model accuracy of HFL by up to 15.1% when training a convolutional neural network on the CIFAR-10 dataset.

Marcello Bullo, Seifallah Jardak, Pietro Carnelli et al.

Deep learning (DL) models have emerged as a promising solution for the Internet of Things (IoT). However, due to their computational complexity, DL models consume significant amounts of energy, which can rapidly drain the battery and compromise the performance of IoT devices. For sustainable operation, we consider an edge device with a rechargeable battery and energy harvesting (EH) capabilities. In addition to the stochastic nature of the ambient energy source, the harvesting rate is often insufficient to meet the inference energy requirements, leading to drastic performance degradation in energy-agnostic devices. To mitigate this problem, we propose energy-adaptive dynamic early exiting (EE) to enable efficient and accurate inference in an EH edge intelligence system. Our approach derives an energy-aware EE policy that determines the optimal amount of computational processing on a per-sample basis. The proposed policy balances the energy consumption to match the limited incoming energy and achieves continuous availability. Numerical results show that accuracy and service rate are improved up to 25% and 35%, respectively, in comparison with an energy-agnostic policy.

Michal Yemini, Rajarshi Saha, Emre Ozfatura et al.

Intermittent connectivity of clients to the parameter server (PS) is a major bottleneck in federated edge learning frameworks. The lack of constant connectivity induces a large generalization gap, especially when the local data distribution amongst clients exhibits heterogeneity. To overcome intermittent communication outages between clients and the central PS, we introduce the concept of collaborative relaying wherein the participating clients relay their neighbors' local updates to the PS in order to boost the participation of clients with poor connectivity to the PS. We propose a semi-decentralized federated learning framework in which at every communication round, each client initially computes a local consensus of a subset of its neighboring clients' updates, and eventually transmits to the PS a weighted average of its own update and those of its neighbors'. We appropriately optimize these local consensus weights to ensure that the global update at the PS is unbiased with minimal variance - consequently improving the convergence rate. Numerical evaluations on the CIFAR-10 dataset demonstrate that our collaborative relaying approach outperforms federated averaging-based benchmarks for learning over intermittently-connected networks such as when the clients communicate over millimeter wave channels with intermittent blockages.

Yuxuan Sun, Sheng Zhou, Zhisheng Niu et al.

Federated edge learning (FEEL) is a promising distributed machine learning (ML) framework to drive edge intelligence applications. However, due to the dynamic wireless environments and the resource limitations of edge devices, communication becomes a major bottleneck. In this work, we propose time-correlated sparsification with hybrid aggregation (TCS-H) for communication-efficient FEEL, which exploits jointly the power of model compression and over-the-air computation. By exploiting the temporal correlations among model parameters, we construct a global sparsification mask, which is identical across devices, and thus enables efficient model aggregation over-the-air. Each device further constructs a local sparse vector to explore its own important parameters, which are aggregated via digital communication with orthogonal multiple access. We further design device scheduling and power allocation algorithms for TCS-H. Experiment results show that, under limited communication resources, TCS-H can achieve significantly higher accuracy compared to the conventional top-K sparsification with orthogonal model aggregation, with both i.i.d. and non-i.i.d. data distributions.

Maximilian Egger, Rawad Bitar, Antonia Wachter-Zeh et al.

We consider the distributed SGD problem, where a main node distributes gradient calculations among $n$ workers. By assigning tasks to all the workers and waiting only for the $k$ fastest ones, the main node can trade-off the algorithm's error with its runtime by gradually increasing $k$ as the algorithm evolves. However, this strategy, referred to as adaptive $k$-sync, neglects the cost of unused computations and of communicating models to workers that reveal a straggling behavior. We propose a cost-efficient scheme that assigns tasks only to $k$ workers, and gradually increases $k$. We introduce the use of a combinatorial multi-armed bandit model to learn which workers are the fastest while assigning gradient calculations. Assuming workers with exponentially distributed response times parameterized by different means, we give empirical and theoretical guarantees on the regret of our strategy, i.e., the extra time spent to learn the mean response times of the workers. Furthermore, we propose and analyze a strategy applicable to a large class of response time distributions. Compared to adaptive $k$-sync, our scheme achieves significantly lower errors with the same computational efforts and less downlink communication while being inferior in terms of speed.

Ozan Aygün, Mohammad Kazemi, Deniz Gündüz et al.

Federated learning (FL) over wireless communication channels, specifically, over-the-air (OTA) model aggregation framework is considered. In OTA wireless setups, the adverse channel effects can be alleviated by increasing the number of receive antennas at the parameter server (PS), which performs model aggregation. However, the performance of OTA FL is limited by the presence of mobile users (MUs) located far away from the PS. In this paper, to mitigate this limitation, we propose hierarchical over-the-air federated learning (HOTAFL), which utilizes intermediary servers (IS) to form clusters near MUs. We provide a convergence analysis for the proposed setup, and demonstrate through theoretical and experimental results that local aggregation in each cluster before global aggregation leads to a better performance and faster convergence than OTA FL.

Tze-Yang Tung, David Burth Kurka, Mikolaj Jankowski et al.

Recent works have shown that the task of wireless transmission of images can be learned with the use of machine learning techniques. Very promising results in end-to-end image quality, superior to popular digital schemes that utilize source and channel coding separation, have been demonstrated through the training of an autoencoder, with a non-trainable channel layer in the middle. However, these methods assume that any complex value can be transmitted over the channel, which can prevent the application of the algorithm in scenarios where the hardware or protocol can only admit certain sets of channel inputs, such as the use of a digital constellation. Herein, we propose DeepJSCC-Q, an end-to-end optimized joint source-channel coding scheme for wireless image transmission, which is able to operate with a fixed channel input alphabet. We show that DeepJSCC-Q can achieve similar performance to models that use continuous-valued channel input. Importantly, it preserves the graceful degradation of image quality observed in prior work when channel conditions worsen, making DeepJSCC-Q much more attractive for deployment in practical systems.

Tze-Yang Tung, Deniz Gündüz

We present DeepWiVe, the first-ever end-to-end joint source-channel coding (JSCC) video transmission scheme that leverages the power of deep neural networks (DNNs) to directly map video signals to channel symbols, combining video compression, channel coding, and modulation steps into a single neural transform. Our DNN decoder predicts residuals without distortion feedback, which improves video quality by accounting for occlusion/disocclusion and camera movements. We simultaneously train different bandwidth allocation networks for the frames to allow variable bandwidth transmission. Then, we train a bandwidth allocation network using reinforcement learning (RL) that optimizes the allocation of limited available channel bandwidth among video frames to maximize overall visual quality. Our results show that DeepWiVe can overcome the cliff-effect, which is prevalent in conventional separation-based digital communication schemes, and achieve graceful degradation with the mismatch between the estimated and actual channel qualities. DeepWiVe outperforms H.264 video compression followed by low-density parity check (LDPC) codes in all channel conditions by up to 0.0462 on average in terms of the multi-scale structural similarity index measure (MS-SSIM), while beating H.265 + LDPC by up to 0.0058 on average. We also illustrate the importance of optimizing bandwidth allocation in JSCC video transmission by showing that our optimal bandwidth allocation policy is superior to the naïve uniform allocation. We believe this is an important step towards fulfilling the potential of an end-to-end optimized JSCC wireless video transmission system that is superior to the current separation-based designs.

Yuxuan Sun, Fan Zhang, Junlin Zhao et al.

Distributed computing enables large-scale computation tasks to be processed over multiple workers in parallel. However, the randomness of communication and computation delays across workers causes the straggler effect, which may degrade the performance. Coded computation helps to mitigate the straggler effect, but the amount of redundant load and their assignment to the workers should be carefully optimized. In this work, we consider a multi-master heterogeneous-worker distributed computing scenario, where multiple matrix multiplication tasks are encoded and allocated to workers for parallel computation. The goal is to minimize the communication plus computation delay of the slowest task. We propose worker assignment, resource allocation and load allocation algorithms under both dedicated and fractional worker assignment policies, where each worker can process the encoded tasks of either a single master or multiple masters, respectively. Then, the non-convex delay minimization problem is solved by employing the Markov's inequality-based approximation, Karush-Kuhn-Tucker conditions, and successive convex approximation methods. Through extensive simulations, we show that the proposed algorithms can reduce the task completion delay compared to the benchmarks, and observe that dedicated and fractional worker assignment policies have different scopes of applications.

Mehmet Emre Ozfatura, Junlin Zhao, Deniz Gündüz

We consider federated edge learning (FEEL) over wireless fading channels taking into account the downlink and uplink channel latencies, and the random computation delays at the clients. We speed up the training process by overlapping the communication with computation. With fountain coded transmission of the global model update, clients receive the global model asynchronously, and start performing local computations right away. Then, we propose a dynamic client scheduling policy, called MRTP, for uploading local model updates to the parameter server (PS), which, at any time, schedules the client with the minimum remaining upload time. However, MRTP can lead to biased participation of clients in the update process, resulting in performance degradation in non-iid data scenarios. To overcome this, we propose two alternative schemes with fairness considerations, termed as age-aware MRTP (A-MRTP), and opportunistically fair MRTP (OF-MRTP). In A-MRTP, the remaining clients are scheduled according to the ratio between their remaining transmission time and the update age, while in OF-MRTP, the selection mechanism utilizes the long term average channel rate of the clients to further reduce the latency while ensuring fair participation of the clients. It is shown through numerical simulations that OF-MRTP provides significant reduction in latency without sacrificing test accuracy.

Amir Ahooye Atashin, Behrooz Razeghi, Deniz Gündüz et al.

We study the role of information complexity in privacy leakage about an attribute of an adversary's interest, which is not known a priori to the system designer. Considering the supervised representation learning setup and using neural networks to parameterize the variational bounds of information quantities, we study the impact of the following factors on the amount of information leakage: information complexity regularizer weight, latent space dimension, the cardinalities of the known utility and unknown sensitive attribute sets, the correlation between utility and sensitive attributes, and a potential bias in a sensitive attribute of adversary's interest. We conduct extensive experiments on Colored-MNIST and CelebA datasets to evaluate the effect of information complexity on the amount of intrinsic leakage.

Yuxuan Sun, Sheng Zhou, Zhisheng Niu et al.

Machine learning and wireless communication technologies are jointly facilitating an intelligent edge, where federated edge learning (FEEL) is a promising training framework. As wireless devices involved in FEEL are resource limited in terms of communication bandwidth, computing power and battery capacity, it is important to carefully schedule them to optimize the training performance. In this work, we consider an over-the-air FEEL system with analog gradient aggregation, and propose an energy-aware dynamic device scheduling algorithm to optimize the training performance under energy constraints of devices, where both communication energy for gradient aggregation and computation energy for local training are included. The consideration of computation energy makes dynamic scheduling challenging, as devices are scheduled before local training, but the communication energy for over-the-air aggregation depends on the l2-norm of local gradient, which is known after local training. We thus incorporate estimation methods into scheduling to predict the gradient norm. Taking the estimation error into account, we characterize the performance gap between the proposed algorithm and its offline counterpart. Experimental results show that, under a highly unbalanced local data distribution, the proposed algorithm can increase the accuracy by 4.9% on CIFAR-10 dataset compared with the myopic benchmark, while satisfying the energy constraints.

Fan Mo, Anastasia Borovykh, Mohammad Malekzadeh et al.

In collaborative learning, clients keep their data private and communicate only the computed gradients of the deep neural network being trained on their local data. Several recent attacks show that one can still extract private information from the shared network's gradients compromising clients' privacy. In this paper, to quantify the private information leakage from gradients we adopt usable information theory. We focus on two types of private information: original information in data reconstruction attacks and latent information in attribute inference attacks. Furthermore, a sensitivity analysis over the gradients is performed to explore the underlying cause of information leakage and validate the results of the proposed framework. Finally, we conduct numerical evaluations on six benchmark datasets and four well-known deep models. We measure the impact of training hyperparameters, e.g., batches and epochs, as well as potential defense mechanisms, e.g., dropout and differential privacy. Our proposed framework enables clients to localize and quantify the private information leakage in a layer-wise manner, and enables a better understanding of the sources of information leakage in collaborative learning, which can be used by future studies to benchmark new attacks and defense mechanisms.

Mohammad Malekzadeh, Anastasia Borovykh, Deniz Gündüz

It is known that deep neural networks, trained for the classification of non-sensitive target attributes, can reveal sensitive attributes of their input data through internal representations extracted by the classifier. We take a step forward and show that deep classifiers can be trained to secretly encode a sensitive attribute of their input data into the classifier's outputs for the target attribute, at inference time. Our proposed attack works even if users have a full white-box view of the classifier, can keep all internal representations hidden, and only release the classifier's estimations for the target attribute. We introduce an information-theoretical formulation for such attacks and present efficient empirical implementations for training honest-but-curious (HBC) classifiers: classifiers that can be accurate in predicting their target attribute, but can also exploit their outputs to secretly encode a sensitive attribute. Our work highlights a vulnerability that can be exploited by malicious machine learning service providers to attack their user's privacy in several seemingly safe scenarios; such as encrypted inferences, computations at the edge, or private knowledge distillation. Experimental results on several attributes in two face-image datasets show that a semi-trusted server can train classifiers that are not only perfectly honest but also accurately curious. We conclude by showing the difficulties in distinguishing between standard and HBC classifiers, discussing challenges in defending against this vulnerability of deep classifiers, and enumerating related open directions for future studies.

Mingzhe Chen, Deniz Gündüz, Kaibin Huang et al.

The next-generation of wireless networks will enable many machine learning (ML) tools and applications to efficiently analyze various types of data collected by edge devices for inference, autonomy, and decision making purposes. However, due to resource constraints, delay limitations, and privacy challenges, edge devices cannot offload their entire collected datasets to a cloud server for centrally training their ML models or inference purposes. To overcome these challenges, distributed learning and inference techniques have been proposed as a means to enable edge devices to collaboratively train ML models without raw data exchanges, thus reducing the communication overhead and latency as well as improving data privacy. However, deploying distributed learning over wireless networks faces several challenges including the uncertain wireless environment, limited wireless resources (e.g., transmit power and radio spectrum), and hardware resources. This paper provides a comprehensive study of how distributed learning can be efficiently and effectively deployed over wireless edge networks. We present a detailed overview of several emerging distributed learning paradigms, including federated learning, federated distillation, distributed inference, and multi-agent reinforcement learning. For each learning framework, we first introduce the motivation for deploying it over wireless networks. Then, we present a detailed literature review on the use of communication techniques for its efficient deployment. We then introduce an illustrative example to show how to optimize wireless networks to improve its performance. Finally, we introduce future research opportunities. In a nutshell, this paper provides a holistic set of guidelines on how to deploy a broad range of distributed learning frameworks over real-world wireless communication networks.

David Burth Kurka, Deniz Gündüz

We propose deep learning based communication methods for adaptive-bandwidth transmission of images over wireless channels. We consider the scenario in which images are transmitted progressively in layers over time or frequency, and such layers can be aggregated by receivers in order to increase the quality of their reconstructions. We investigate two scenarios, one in which the layers are sent sequentially, and incrementally contribute to the refinement of a reconstruction, and another in which the layers are independent and can be retrieved in any order. Those scenarios correspond to the well known problems of \textit{successive refinement} and \textit{multiple descriptions}, respectively, in the context of joint source-channel coding (JSCC). We propose DeepJSCC-$l$, an innovative solution that uses convolutional autoencoders, and present three architectures with different complexity trade-offs. To the best of our knowledge, this is the first practical multiple-description JSCC scheme developed and tested for practical information sources and channels. Numerical results show that DeepJSCC-$l$ can learn to transmit the source progressively with negligible losses in the end-to-end performance compared with a single transmission. Moreover, DeepJSCC-$l$ has comparable performance with state of the art digital progressive transmission schemes in the challenging low signal-to-noise ratio (SNR) and small bandwidth regimes, with the additional advantage of graceful degradation with channel SNR.

Dagnachew Azene Temesgene, Marco Miozzo, Deniz Gündüz et al.

To meet the growing quest for enhanced network capacity, mobile network operators (MNOs) are deploying dense infrastructures of small cells. This, in turn, increases the power consumption of mobile networks, thus impacting the environment. As a result, we have seen a recent trend of powering mobile networks with harvested ambient energy to achieve both environmental and cost benefits. In this paper, we consider a network of virtualized small cells (vSCs) powered by energy harvesters and equipped with rechargeable batteries, which can opportunistically offload baseband (BB) functions to a grid-connected edge server depending on their energy availability. We formulate the corresponding grid energy and traffic drop rate minimization problem, and propose a distributed deep reinforcement learning (DDRL) solution. Coordination among vSCs is enabled via the exchange of battery state information. The evaluation of the network performance in terms of grid energy consumption and traffic drop rate confirms that enabling coordination among the vSCs via knowledge exchange achieves a performance close to the optimal. Numerical results also confirm that the proposed DDRL solution provides higher network performance, better adaptation to the changing environment, and higher cost savings with respect to a tabular multi-agent reinforcement learning (MRL) solution used as a benchmark.