We propose an analytical framework based on a simple metric and capable of analyzing mobile network data so as to identify changes in consumption patterns across antennas due to the occurrence of massive public protests. We collect data from an operational network in France and analyze how it was impacted by the 2023 French pension reform strikes. We are able to identify a number of antennas that were clearly affected by the strike, and to follow the corresponding events in the mobile traffic demand as it propagates in space and time along the designated route followed by the marchers. The proposed framework is a stepping stone for more robust classification models on the impacts of massive protests on mobile networks, paving the road to network-based solutions for a pervasive and cost-effective monitoring of such events.

The backpressure routing method offers a dynamic resource allocation approach that achieves optimal throughput. However, the method's sole dependence on the first-order backlog difference between nodes leads to poor convergence and may result in unnecessarily long paths, potentially causing loops when unrestricted. In this work, we address both transmission cost and computational complexity concerns. We introduce a second-order backpressure algorithm (SBP) featuring adaptive shortest routing, a second-order backlog metric, and a two-level queue mapping structure. We validate this novel backlog metric through Lyapunov optimization techniques. Simulation results demonstrate that while ensuring throughput, our approach effectively reduces end-to-end delay and eliminates routing loops without imposing specific constraints.

In a network topology G, we say a set of st-paths are disaster-disjoint if no disaster strikes more than one path. In this poster, we explore the basic capabilities and limitations of greedy and more advanced algorithms for computing maximal collections of such paths in planar networks. An algorithm is greedy if it generates consecutive paths P1, P2,... according to a simple rule. In the simplest setting, the only rule is that P_i+1 is the closest clockwise disaster-disjoint from P_i. We find that the simplest greedy may fail even when 1) G is planar, 2) each disaster region is connected, and 3) each node failure (apart from s and t ) is considered possible. Adding a simple rule explained in [1] yields a correct polynomial-time algorithm for the above problem. Finally, we digest a recent related near-linear runtime algorithm of [2] solving a more general problem and discuss the underlying relations among the foundations of these algorithms.

Federated Learning (FL) has emerged as a promising decentralized machine learning (ML) paradigm where distributed clients collaboratively train models without sharing their private data. However, due to their limited resources and heterogeneous properties, only a small subset of clients can participate at a given time. Furthermore, the high dimensions of ML models incur a massive communication overhead which considerably slows down the convergence of FL. To address the aforementioned challenges, we propose FedHSC, a framework that considers both system and statistical heterogeneity. Specifically, at each communication round, the clients are sampled based on their data properties combined with the importance of their local learning update. After completing their local training, the selected clients share compressed updates with the server for aggregation. The compression rate is adjusted for each client to meet the communication delay requirement. Experimental results on CIFAR-10 show the efficiency of our approach and its robustness to Non-IID data.

Cooperative perception for Connected and Autonomous Vehicles (CAVs) improve road user safety. However, the increase in complexity of the resource intensive vision based algorithms involved can result in large amount of computing and communication resource consumption on energy-limited CAVs. We believe CAVs can leverage the trajectories of nearby road users and plan to navigate around them instead of sharing large amount of sensory data. We propose a novel federated learning based energy efficient approach for CAVs to predict nearby road users Global Positioning System (GPS) coordinates in real-time and share with other vehicles in proximity. We employ sequential and transformer-based models for predicting CAV trajectories with the proposed approach and show its potential to reduce resource consumption with high prediction accuracy.

The rapid progress in wearable sensors, smartphones equipped with sensors, and seamless cloud integration has ignited significant research into the creation of IoT-driven intelligent systems designed for human activity recognition (HAR). However, the reliance on external data curation for training machine learning (ML)-based recognition models renders the system susceptible to adversarial attacks. The limited existing research in this area calls for extensive efforts to address its gaps, notably the absence of vital contextual information crucial for HAR systems. In this poster, we present our ongoing research effort on investigating context-aware spatiotemporal data poisoning attacks on the intelligence of HAR systems. These attacks involve attackers exploiting specific spatiotemporal patterns and conditions to manipulate the labels of activity data. This manipulation is strategically designed to mislead the recognition system, thereby highlighting the pressing need for improved security measures in this rapidly evolving domain.

Acoustic sensing holds great promise in the realm of human-computer interaction owing to the ubiquitous availability of speakers and microphones in smart homes. In this work, we propose a promising solution aided by contactless acoustic sensing, where people with motor neuron disease (MND) who have lost their motion and speech capabilities can input text with only a signal finger. However, it is still challenging for finger sensing due to the tiny signal variation and severe interference from uncontrolled body motion or other objects. To address these issues, we propose a search-based strategy to extract tiny motions and an adaptive origin transformation method to eliminate the interference. Extensive experiments demonstrate that it can achieve a recognition accuracy of 96.2% for single-finger text inputting.

Federated learning (FL) has emerged as a promising paradigm for privacy-preserving collaborative learning, enabling multiple devices to jointly train a global model without sharing their raw data. However, the bias in FL training significantly reduces its performance. This poster presents a novel FL algorithm to counteract bias for performance improvement. First, we provide a global perspective for analyzing the causes of bias, data heterogeneity and transmission probability. Then, we propose a method that introduces a regularized local training method and a reweighted aggregation strategy to jointly mitigate bias. Through extensive experiments on real-world datasets, we demonstrate that our method significantly outperforms various baseline FL methods in terms of convergence speed and accuracy.

Multi-robot systems are adept at handling complex tasks in large-scale, dynamic, and cold-start scenarios such as wildfire control. This paper introduces FireHunter to tackle the challenge of coordinating fire monitoring and suppression tasks simultaneously in unpredictable environments. FireHunter utilizes a confidence-aware assessment method to identify optimal locations and a priority graph-based algorithm to coordinate robots efficiently. It effectively manages the dynamic planning inclinations for sensing and operational tasks, ensuring real-time information collection and timely environmental intervention. Experimental results from simulation show that FireHunter reduces fire expansion ratio by 59% compared to state-of-the-art solutions.

As a leading approach to improving the scalability of Ethereum, Rollup projects (e.g., Arbitrum) have proliferated in recent years, leading to an increased demand for cross-rollup solutions. Unfortunately, existing cross-rollup solutions are either slow and expensive or unsuitable for complex cross-rollup interaction. We present ROLLUPNET, the first state channel that synchronously executes cross-rollup contracts to support complex cross-rollup applications while providing both security and efficiency guarantees. To prevent parties from misbehaving, we design an Ethereum-based dispute resolution to ensure the consistency of state between different rollups. The results show that the ROLLUPNET is practical for Ethereum rollups.

In vehicular networks (VNets), ensuring both road safety and dependable network connectivity is of utmost importance. Achieving this necessitates the creation of resilient and efficient decision-making policies that prioritize multiple objectives. In this paper, we develop a Variational Quantum Circuit (VQC)-based multi-objective reinforcement learning (MORL) framework to characterize efficient network selection and autonomous driving policies in a vehicular network (VNet). Numerical results showcase notable enhancements in both convergence rates and rewards when compared to conventional deep-Q networks (DQNs), validating the efficacy of the VQC-MORL solution.

We propose and demonstrate a novel converged system cascading a fiber, sub-terahertz (sub-THz), fiber, and millimeter-wave (mmWave) link for radio signal transmission from a central station to indoor users. Direct conversion between optical and sub-THz signal and from sub-THz to mmWave signal was performed using photonic technology for transparent radio signal transmission. We successfully transmitted 5G New Radio standard-compliant signals in the 38-GHz band over a converged system in the 172-GHz band and achieved satisfactory performance.