INFOCOM 2026 Papers

Federated Learning (FL) has emerged as a powerful paradigm for distributed model training that preserves data privacy while aggregating model updates through a central server. However, efficiently implementing federated learning in resource-constrained and heterogeneous edge environments remains challenging due to the trade-off between convergence performance and computation/communication costs. Most of the existing approaches only heuristically tune the compression and computation parameters without directly minimizing the theoretical convergence bound. In this paper, we propose the first FL framework, Convergence-driven Federated learning (ConFed), that explicitly formulates and minimizes the non-convex convergence error upper bound under time-averaged expected resource constraints, providing a principled foundation for joint optimization of computation and compression. ConFed supports diverse compression mechanisms, including Top-k sparsification, random quantization, and their combinations. Extensive experiments on Fashion-MNIST (FMNIST) and CIFAR-10 demonstrate that our method achieves faster convergence and lower resource costs compared to state-of-the-art approaches.

Contemporary network applications require ultra-reliable and low-latency connectivity. However, existing 5G mobility procedures, including the newly introduced 3GPP Release 18 mobility procedure, called L1/L2-Triggered Mobility (LTM), still face challenges in providing fairness, wasting reserved resources, and high signaling overhead, especially in dense networks. In this work, we investigate max-min fair mobility management that jointly optimizes user-to-Base Station (BS) assignment, resource allocation, and cell preparation decisions. We formulate a multi-objective optimization problem with the goal of providing max-min fairness in user data rates while minimizing resource reservation overhead due to the unnecessary cell preparations. We derive an upper bound through a problem transformation and decision variable relaxation. Then, we propose a context-aware Deep Reinforcement Learning (DRL)-based solution, with decisions guided via action masking. Our approach uses a two-phase decision process that separates BS selection and resource allocation from cell preparation decisions for a better performance. Extensive 3GPP-compliant simulations in dynamic macro/micro scenarios demonstrate that our DLR-based algorithm significantly improves fairness and resource efficiency, reducing resource reservation by ≈ 60% compared to the baselines, and achieves an optimality gap of 15%.

The early diagnosis and effective management of Pulmonary Hypertension (PH) are critically dependent on the accurate monitoring of Pulmonary Arterial Pressure (PAP). However, current clinical methods are either invasive or resource-intensive, which limits their application for continuous PAP monitoring. To overcome this challenge, we introduce a novel, noninvasive method for PAP monitoring utilizing in-ear audio. Our approach traces the Heart Sound (HS) through a physics-guided modeling from in-ear audio. Subsequently, we derive PAP indicators from the second heart sound (S2) using a sophisticated signal processing pipeline. This pipeline incorporates mode decomposition, spectral clustering, and Morphological Component Analysis (MCA). Finally, a Dendritic Neuron Regression (DNR) network is employed to determine absolute PAP values from the extracted indicators. Validation of our method on real-world data from 26 participants demonstrates clinically acceptable error margins in the estimation of mean PAP (mPAP), with an estimation error of 1.31 ± 2.83 mmHg. These promising results highlight the potential of our in-ear audio approach for developing wearable, noninvasive, and continuous PAP monitoring devices. This could significantly improve the management of PH by providing a more accessible and patient-friendly monitoring solution.

Remote Direct Memory Access (RDMA) has emerged as a cornerstone of datacenter networks (DCNs). Offloading the protocol stack to RDMA NIC (RNIC) makes data transmission depend on metadata caches, such as Queue Pair Context (QPC) and Work Queue Element (WQE). As DCNs expand, the number of Queue Pairs (QPs) grows, RNIC’s limited cache capacity becomes insufficient, triggering frequent cache misses and leading to serious connection scalability issues. Hardware-based solutions are prohibitively expensive. Existing software-based approaches mainly focus on QPC cache. Some attempt to optimize WQE cache, but only address Send WQEs (SWQEs). In this paper, to deal with unpredictable message arrival for optimizing Receive WQEs (RWQEs), we propose Symphony, a coordinated sender-receiver approach along with immediate data driven feedback that optimizes both SWQE and RWQE cache. We validate Symphony through NS3 simulation and a prototype implementation in RDMA driver, we also analyze its stability via the Laplace domain. The simulation results demonstrate that Symphony outperforms Weir in diverse scenarios, and testbed experiments indicate that Symphony can maintain 96Gbps with low CPU overhead. Compared to Weir, Symphony prominently reduces observable cache misses, ICM by approximately 20%, MTT/MPT by approximately 24% and RWQE by approximately 74%.

Collaborative perception enhances situational awareness among vehicles by enabling information sharing across agents. However, network-induced delays reduce temporal consistency and introduce spatial divergence in shared observations, posing a critical challenge in distributed, delay-sensitive vehicular networks. Although delay alignment mitigates part of the effect, alignment errors and inherent multi-view discrepancies often lead to inconsistencies when synchronizing delayed data to a common timestamp. Crucially, these inconsistencies can accumulate during fusion, propagating cascading errors into downstream trajectory prediction. To address this, we propose STCC, a prediction-oriented collaborative perception framework based on Spatio-Temporal Co-Sensing Calibration. STCC adopts a temporal-spatial cascade design: a history-aware temporal alignment module captures dynamic alignment patterns to compensate for variable delays, while an Interaction-Aware Uncertainty Quantification scheme explicitly models environmental interaction risks to correct cross-view representation divergence. This joint calibration ensures consistent multi-agent fusion and reliable downstream prediction under network-induced delay. We evaluate STCC on large-scale public cooperative perception benchmarks. Experimental results show that STCC consistently outperforms state-of-the-art methods across diverse delay conditions, significantly reducing the Average Displacement Error (ADE) and Final Displacement Error (FDE), demonstrating superior robustness and consistency.

Vision-language models (VLMs) have demonstrated impressive multimodal comprehension capabilities and are being deployed in an increasing number of online video understanding applications. While recent efforts extensively explore advancing VLMs’ reasoning power in these cases, deployment constraints are overlooked, leading to overwhelming system overhead in real-world deployments. To address that, we propose Venus, an on-device memory-and-retrieval system for efficient online video understanding. Venus proposes an edge–cloud disaggregated architecture that sinks memory construction and keyframe retrieval from cloud to edge, operating in two stages. In the ingestion stage, Venus continuously processes streaming edge videos via scene segmentation and clustering, where the selected keyframes are embedded with a multimodal embedding model to build a hierarchical memory for efficient storage and retrieval. In the querying stage, Venus indexes incoming queries from memory, and employs a threshold-based progressive sampling algorithm for keyframe selection that enhances diversity and adaptively balances system cost and reasoning accuracy. Our extensive evaluation shows that Venus achieves a 15×–131× speedup in total response latency compared to state-of-the-art methods, enabling real-time responses within seconds while maintaining comparable or even superior reasoning accuracy.

Quantum key distribution (QKD) networks provide information-theoretically secure keys for distant parties, emerging as a vital alternative to classical cryptography infrastructures threatened by quantum computing. In QKD networks, the immediacy of key supply service is crucial to the security and performance of applications, as their data must be encrypted before transmission. While key buffering can enable instant key supply services, existing schemes rely on heuristic solutions that incur prohibitive key resource consumption, thus significantly hindering practical deployment. To address this issue, we propose QuIKS, an instant key supply scheme based on adaptive buffering, offering the dominant advantage of near-zero key supply latency while consuming ultra-low key resources (i.e., ultra-low buffer size). Specifically, it is built upon a novel analytical model that determines the minimum buffer size required to guarantee near-zero-latency key supply performance. Guided by this model, QuIKS introduces a lightweight two-phase control algorithm that dynamically determines key relaying requests and adjusts the buffer size by probing real-time application patterns and network conditions. Experiments on a real QKD network testbed demonstrate that QuIKS achieves near-zero key supply latency while providing a more than 10-fold reduction in key buffer size compared to state-of-the-art schemes.

High-throughput wireless communication relies on high-resolution channel estimates to accurately compensate for the distortions imposed by the underlying wireless multipath propagation. However, due to the frequency-dependent superposition of delayed propagation paths, certain components may undergo destructive interference at specific frequencies, thus obscuring their contribution. This leads to a partial reconstruction of the channel which ultimately degrades communication performance. As such, we propose BandWeave, a new general-purpose multi-band channel frequency response (CFR) fusion approach that significantly improves channel estimation accuracy by leveraging complementary multipath information across multiple frequency bands. BandWeave enables adaptive, robust and efficient CFR fusion through a three-phase progressive learning algorithm. We experimentally evaluated BandWeave with an IEEE 802.11ac multi-user (MU)-multiple-input multiple-output (MIMO) testbed and a 60 GHz single-user (SU)-MIMO system in three propagation environments with different fading conditions. Experimental results show that BandWeave channel estimation outperforms single-band estimation by achieving more than 16% of throughput improvement and up to 10.2% bit error rate (BER) reduction. BandWeave achieves a gain of more than 7.5× in communication performance compared to state of the art (SOTA) band-merging approaches. Measurements on Jetson Nano, Jetson Orin Nano and Raspberry Pi 5 show that BandWeave reduces inference time and energy consumption by up to 20× and 18×, respectively, compared to SOTA methods. A video demonstration of BandWeave is available1.

Magnetic field sensing is essential for electrical safety and early hazard detection. While existing solutions are generally considered reliable and accurate, they are hindered by battery-dependent challenges, invasive detection, and poor scalability. To address these limitations, in this paper, we propose a novel μT-level magnetic field sensing method based on RFID tags, named RF-Gaussmeter. The design of RF-Gaussmeter involves embedding a Tunnel Magnetoresistance (TMR) sensor into the COTS RFID tags. We consider the magnetic field generated by current-carrying conductors as the sensing target. When the TMR sensor captures the magnetic field generated by a conductor, it can output the sensing voltage signal to the RFID tag, altering the chip’s impedance and thereby influencing the received signal. To amplify weak magnetic fields, we explore the interaction model between the MOSFET in RFID tags and differential voltage signals. Based on this model, we propose a differential signal amplification design based on RFID to enhance sensing sensitivity. To extract the magnetic-field-related sensing features, we develop a self-differential module based on RFID to remove multi-path interference. Experimental results show that RF-Gaussmeter achieves an absolute error of 0.0825A for current sensing and 1.32μT for magnetic field sensing within 80cm range, demonstrating excellent practicality and accuracy.

Today, large DNN models are often trained in parallel and distributed environments, which inevitably incur high costs for fault tolerance. Existing solutions rely on periodic checkpoints to save training state in run-time and recomputing from the latest checkpoint to recover upon a failure. Despite recent efforts to reduce checkpoint cost, data transfer bandwidth can become a bottleneck, limiting checkpoint frequency, and consequently leading to high recomputing cost. This paper focuses on efficient fault tolerance for large model training and proposes Controlled Predicting-assisted Self-Recovery (CPSR). Instead of resource-consuming recomputation, it features a lightweight predictor, fed by routine checkpoints, to predict training state prior to a failure. The prediction error exhibits as a minor perturbation that can be self-corrected by the training process itself. We propose a quantified model of predicting-based recovery cost during rehabilitation and introduce a novel checkpoint interval problem that seeks to minimize the overall fault tolerance cost. We present a solution to compute the optimal checkpoint interval configuration for a given setting, balancing checkpoint and recovery cost. Extensive testbed experiment data demonstrate that CPSR reduces the recovery cost by 41.66% on average compared with state-of-the-art approaches, while introducing a small GPU memory footprint (less than 200MB).

Millimeter-wave (mmWave) technology has enabled emerging applications such as vital-sign-based healthcare monitoring, user authentication, and emotion-aware human-computer interaction. By capturing subtle chest displacements induced by heartbeat and respiration, mmWave systems provide high-resolution, contactless chest-vibration sensing. However, the mmWave signals that power these applications are vulnerable to spoofing attacks, posing serious risks such as identity impersonation and falsified health assessments. While prior studies have demonstrated the feasibility of spoofing mmWave sensing, existing methods often require access to raw mmWave data or rely on expensive, specialized RF equipment, limiting their real-world applicability. In this work, we present a real-time, low-cost spoofing attack using a programmable actuator concealed under clothing to physically mimic a target user’s chest vibrations. Our attack allows adversaries to bypass authentication systems or falsify health data, potentially granting unauthorized access, or concealing critical medical conditions and triggering false emergency responses. To ensure high-fidelity spoofing, we introduce a mitigation strategy that integrates IMU-assisted compensation and quaternion-based alignment to mitigate interference from the attacker’s own chest motion. We further employ deep learning to dynamically adjust actuator behavior in real time. Experiments with eight participants over six months validate the attack’s effectiveness, revealing a critical security vulnerability in emerging mmWave-based sensing systems.

Real-time communication (RTC) applications demand not only low average latency but also tight tail-delay bounds to ensure smooth user experience. However, sudden fluctuations in wireless networks can cause in-flight packets to accumulate in bottleneck queues, delaying or invalidating subsequent frames. Traditional mechanisms focus on rate adaptation but largely overlook managing already enqueued packets that contribute to tail delay. We present Gecko, a lightweight end-to-network coordination mechanism that enables frame-aware queue flushing without requiring any in-network packet modification or protocol negotiation. Gecko-enabled routers monitor queuing delay and implicitly signal the sender, which then makes frame-skipping decisions and conveys flushing intent through minimal in-band RTP markings. This approach preserves end-to-end integrity and is broadly compatible with existing RTC applications. We evaluate Gecko via trace-driven simulations and real-world experiments. Results show that Gecko reduces average frame delay by 21.8% and cuts tail-delay frame ratios by 25% to 91%, demonstrating both its effectiveness and deployability in wireless RTC environments.

Current Cyber Deception as a Service (CDaaS) offerings force a trade-off between low-cost, low-fidelity services and expensive, high-fidelity on-premise deployments. To break this impasse, this study proposes DDCA (Dynamic Deception Cloudification Architecture), a novel cloudified defense framework. DDCA leverages a programmable data plane to construct a high-fidelity cloud deception environment with dynamic topology and controllable characteristics, enabling physically static deception resources to exhibit dynamic behaviors, thus confounding protected services and decoys at the network layer. We first built the DDCA attack-defense theoretical model capable of quantifying attacker's cognitive uncertainty relying on partially observable data. Furthermore, we designed a deception orchestration engine based on Deep Reinforcement Learning (DRL) that transforms defensive constraints into executable strategies of DDCA, jointly optimizing the attack surface hopping, topology reconfiguration, and obfuscated characteristic injection. Experiments conducted on a prototype platform demonstrated that the DDCA can significantly increase expected attack costs at an acceptable expense while effectively countering advanced reconnaissance, providing a more economical systematic solution for CDaaS.

The proliferation of autonomous vehicles necessitates advanced environmental understanding through collaborative perception, yet achieving high accuracy while maintaining communication efficiency and strict latency remains a challenge. Existing region selection strategies, often relying on ego-centric uncertainty, lead to redundant data transmission or the omission of critical information, and their inherent request-response protocols introduce significant latency bottlenecks that hinder real-time deployment. To address these critical limitations and ensure timely, high-value information exchange, we propose BlindMap, a novel collaborative perception framework. BlindMap introduces an interpretable, request-free, and deadline-aware communication paradigm where collaborators proactively infer the ego vehicle’s occluded regions from a third-person perspective, selectively transmitting only critical, unobservable features. This decoupled approach enables parallel processing, significantly reducing system latency. Our framework incorporates a Spatial-Temporal BlindMap Predictor for accurate, proactive blind-area identification and a Blind-Aware Multi-Scale Fusion module that robustly integrates features under noise and misalignment. Extensive experiments on DAIR-V2X-C, OPV2V, and V2XSet demonstrate that BlindMap achieves a state-of-the-art perception accuracy improvement of up to 9.45% under a 1MB bandwidth constraint, while drastically reducing collaboration latency to 21 ms—a 78% decrease compared to existing methods. Even under tight latency budgets, BlindMap consistently maintains high accuracy and stable throughput, proving its potential for robust, real-time, deadline-constrained collaborative perception in dynamic autonomous driving environments. The code is available at: https://github.com/AlexZhu2000/BlindMap

Gait recognition has become an increasingly attractive and prevalent technology in ubiquitous computing. Recent works have explored radio frequency (RF) signals to directly sense walking postures for non-invasive and privacy-preserving gait recognition. However, these approaches face fundamental limitations including environment-dependent nature and line-of-sight (LoS) constraints. To overcome the above limitations, we consider leveraging footstep-induced mechanical signals (e.g., vibration) on the floor for gait recognition. In this paper, we present a multi-user gait recognition system, VibraGait, which leverages a commercial off-the-shelf (COTS) mmWave radar to sense footstep-induced floor vibrations for identifying individuals. VibraGait first separates mmWave-sensed vibrations from different users using a FaSNet-based neural beamforming model in multi-user scenarios, and then constructs individual vibration patterns using a pulse-response model. Furthermore, VibraGait extracts path-invariant vibration patterns by a vibration pattern standardization method to ensure robustness across arbitrary walking paths. Finally, VibraGait extracts distinctive gait features and recognizes users’ identities using a ConvNeXt model. Extensive experiments in real-world environments show that VibraGait effectively identifies users in multi-user scenarios with an average accuracy of 95.5%.

With the rise of decentralized finance, fiat-to-cryptocurrency exchange platforms have become popular en-try points into the cryptocurrency ecosystem. However, these platforms frequently fail to ensure adequate privacy protection, as evidenced by real-world breaches that exposed personally identifiable information (PII) and crypto addresses. Such leaks enable adversaries to link real-world identities to cryptocurrency transactions, undermining the presumed anonymity of cryptocurrency use.We propose FC-GUARD, a privacy-preserving exchange system designed to preserve user anonymity without compromising regulatory compliance in the exchange of fiat currency for cryptocurrencies. Leveraging verifiable credentials and zero-knowledge proof techniques, FC-GUARD enables fiat-to-cryptocurrency exchanges without revealing users’ PII or fiat account details. This breaks the linkage between users’ real-world identities and their cryptocurrency addresses, thereby upholding anonymity, a fundamental expectation in the cryptocurrency ecosystem. In addition, FC-GUARD complies with key regulations over cryptocurrency usage, such as know-your-customer requirements and auditability for tax reporting obligations by integrating a lawful deanonymization mechanism that allows the auditing authority to identify misbehaving users. This ensures regulatory compliance while defaulting to privacy protection. We implement our system on both desktop and mobile platforms, and our evaluation shows its feasibility for practical deployment.

Backdoor attacks pose a significant security threat to deep neural networks (DNNS) by implanting hidden malicious behaviors triggered by specific input patterns. While these attacks typically retain high clean accuracy, their effectiveness can be severely diminished through post-deployment adaptations including fine-tuning, model pruning and quantization techniques commonly applied in edge environments, especially using imbalanced local data. Such scenarios frequently occur in practice due to limited user data, natural distribution skews in downstream tasks, and resource constraints requiring model compression. Existing backdoor injection methods generally overlook these challenges, leaving them vulnerable to suppression during real-world model adaptation. To address this critical limitation, we propose TrojanEdge, a mutual information-enhanced training framework designed for robust and persistent backdoor attacks. TrojanEdge explicitly maximizes mutual information between gradients derived from poisoned data and class-imbalanced data while incorporating dropout-based regularization to enhance robustness against parameter perturbations from pruning and quantization, effectively mitigating gradient drift and structural modifications during post-deployment optimizations. Experiments conducted on MNIST, ,CIFAR-10, and ImageNet-10 demonstrate that TrojanEdge consistently achieves high attack success rates (ASR ≥ 96%) after balanced and imbalanced fine-tuning, pruning, and quantization scenarios, while maintaining competitive clean accuracy.

In cellular networks, inter-cell interference management has been studied for decades, yet its real-world effectiveness remains under-explored. To bridge this gap, we conduct a first measurement study of inter-cell interference for operational 4G/5G networks. Our findings reveal the prevalence of inter-cell interference and a surprising absence of interference coordination among operational base stations (BSs). As a result, user equipments (UEs) experience unnecessary interference, which causes significant signal quality degradation, especially under frequency-selective channel fading. We examine the inter-cell interference issues from four major perspectives: (1) network deployment, (2) channel assignment, (3) time-frequency resource allocation, and (4) network configuration. In none of these dimensions is inter-cell interference effectively managed. Notably, even when spectrum resources are underutilized and simple strategies could effectively mitigate inter-cell interference, BSs consistently prioritize using the same set of time-frequency resources, causing interference across cells. Our measurements reveal substantial opportunities for improving signal quality by inter-cell interference management.