Federated Learning (FL) is a novel paradigm of collaboratively training a model using local datasets of multiple participants. FL maintains data privacy and keeps local datasets confined to the participants. This poster presents a novel noise-resilient federated learning approach that suppresses the negative impact of noisy labels in the local datasets of the participants. The approach starts with the estimation of noise ratio without using prior information about the concentration of noisy labels. Next, the server generates different groups of participants using the estimated noise ratio. The FL-based training starts with the group having the least noise ratio, and subsequent groups are added later. We also introduce a noise robust loss function that incorporates dynamic variables to reduce the impact of noisy labels. The proposed approach reduces the overall training time and achieves adequate accuracy despite noisy labels.

This paper proposes a learning-based loss differentiation method (LLD) for wireless congestion control. LLD uses a neural network to distinguish between wireless packet loss and congestion packet loss in wireless networks. It can work well in combination with classical packet loss-based congestion control algorithms, such as Reno and Cubic. Preliminary results show that our method can effectively differentiate losses and thus improve throughput in wireless scenarios while maintaining the characteristics of the original algorithms.

One of the key visions for 6G is to enable Internet of Intelligence at the network edge. However, many battery-less devices face the dilemmas of energy shortage and spectrum deficiency. To tackle these challenges, we propose a simultaneous information transmission and federated learning (SITFL) scheme for the purpose of overcoming communication bottlenecks and accelerating data processing in wireless power transfer networks. For mean-square-error minimization, a low-complexity solution is developed to optimize the transmit and receive beamforming jointly. Simulation results demonstrate the effectiveness of the proposed solution for wireless powered SITFL networks.

Optimal transport (OT) is a framework that allows for optimal allocation of limited resources in a network consisting of sources and targets. The standard OT paradigm does not extend over a large population of different types. In this paper, we establish a new OT framework with a large and heterogeneous population of target nodes. The heterogeneity of targets is described by a type distribution function. We consider two instances in which the distribution is known and unknown to the sources, i.e., transport designer. For the former case, we propose a fully distributed algorithm to obtain the solution. For the latter case in which the targets' type distribution is not available to the sources, we develop a collaborative learning algorithm to compute the OT scheme efficiently. We evaluate the performance of the proposed learning algorithm using a case study.

Federated learning distributes model training among multiple clients who, driven by privacy concerns, perform training using their local data and only share model weights for iterative aggregation on the server. In this work, we explore the threat of collusion attacks from multiple malicious clients who pose targeted attacks (e.g., label flipping) in a federated learning configuration. By leveraging client weights and the correlation among them, we develop a graph-based algorithm to detect malicious clients. Finally, we validate the effectiveness of our algorithm with different numbers of attackers and normal training clients using a widely adopted Fashion-MNIST dataset.

Unmanned aerial vehicle (UAV) assisted heterogeneous network (HetNet) is a promising solution to outdoor hotspots. This poster proposes a coordination-mini-batch with action mask (CMB-AM) multi-agent deep reinforcement learning (DRL) based resource allocation scheme for uplink spectrum sharing in a UAV-assisted two-tier heterogeneous network (HetNet). We utilize a centralized training and distributed execution mechanism and consider the correlation among actions to make the agents collaborate implicitly. Due to independent state and collective reward design, our resource allocation scheme can be robust and scalable to the varying number of agents. Evaluation shows that the proposed scheme achieves better performance on the aspects of sum capacity, model applicability, and training stability than other baseline schemes.

This paper proposes an inverse reinforcement learning (IRL)-based method to optimize power allocation for multi-user cellular networks. A optimization problem is formulated to maximize the achievable sum information rate of all receivers. In contrast to traditional reinforcement learning (RL)-based methods, the proposed IRL-based one does not require to manually design the reward function manually, which is able to determine the reward function efficiently and automatically from the expert policy. The weighted minimum mean square error (WWMSE) method is used to serve as an expert policy to obtain the reward function, and the action space and sate space are designed. Simulation results show that the proposed IRL-based method achieves about 99 % of the sum information rate achieved by the pure WMMSE method, but the running time of the proposed IRL-based one is about 1/19 of that required of pure WMMSE method.

Smart grid systems provide reliable and efficient power through a smart information and communication technology. Reliability of smart grid systems are of great importance as any critical issue in the system will affect several millions of device connected through communication network. The reliability of smart grid can be compromised by cyber attacks. This entails continuous cyber-security monitoring for smart grid systems. In this work, a machine learning-based cyber attacks detection is proposed. The proposed mechanism is shown to identify false-data injection attaks which is one of the most substantial attacks in smart grid systems. In the proposed scheme, we generated the datasets using an IEEE-34 bus system with cyber attacks implementation. In addition, we have shown that our machine learning models can succesfully identify the attack in smart grid systems.

This paper proposes a deep reinforcement learning (DRL)-based solution framework to maximize the age-energy efficiency (AEE), i.e., the achievable age of information (AoI) gain per consumed energy, in a wireless powered communication network (WPCN), where an edge node (EN) first charges sensors and then the sensors transmit their sensed data to controllers via the EN. To maximize the system AEE, an optimization problem is formulated by jointly optimizing the sensors scheduling and the EN's transmit power, which is modeled by a Markov decision process via wise definitions of state spaces, action spaces, and rewards. Simulation results show that compared with the random-scheduling-based method, our proposed DRL-based framework can improve the AEE by about 4 times with the number of sensors being 30, and the more the number of sensors, the larger the AEE performance can be improved.

In an IoT-based large decentralized smart-system, many devices need to collaborate with each other to achieve the desired goals in a time and energy-efficient manner. Simultaneous communication is one of the key tools to solve the scalability issues in the underlying communication protocols used in these systems. Frequency Division Multiple Access (FDMA) has been one of the elegant strategies to support simultaneous communication. However, massive sharing of the licence free ISM bands among many technologies and many devices results in fast depletion of the availability of the orthogonal frequencies/channels for use in the IoT-systems. To address this resource scarcity, there have been efforts which demonstrate that truly orthogonal channels may not be always necessary to achieve fruitful simultaneous communication. Recent studies have shown that the use of concurrent-transmission can appropriately exploit special radio-features to achieve in-parallel communication even without changing channels. This article summarizes the full spectrum of these works and brings forth an aerial view of the overall problem space. An approximate division of the problem space is also derived and the existing solution approaches are positioned in their respective zones.

Higher refueling time of the Electric-Vehicles (EVs) is one of the major concerns in their wide-spread use for transportation. A well-planned charge scheduling of the EVs, hence, is extremely important for proper utilization of the limited charging infrastructure and also limit the size of the waiting queue in the Charging Stations (CSs). Almost all the existing works on this topic are theoretical and assume the availability of global data of the EVs and the CSs. In this work, we take an endeavor to derive a practically useful solution to this problem through efficient EV-CS coordination. In particular, we perceive the EVs and the CSs to be connected with each other through a Low-Power Wide Area Network (LPWAN) and propose to achieve dynamic EV-CS coordination through the use of Concurrent-Transmission (CT) based mechanism. Through extensive simulation and testbed based studies we demonstrate how the goal can be fruitfully achieved in a quite scalable fashion despite the requirement of a very wide area coverage and active participation of an enormous number of EVs and CSs.

This paper analyzes the NFV-SLB problem, in which NFV performance hinges critically on states. To address the problem, we present a novel state-aware load balancer Statoeuver that judiciously takes state access patterns into consideration and intelligently aligns state sizes, the number of live flows, and CPU loads into consideration for the near-optimal load balancing. We present the paper here to call for insightful and valuable comments for our future work to finalize Statoeuver.

We propose LoRaCoin, a decentralized blockchain-based service to manage the generation and storage of sensor data generated by IoT devices. A novel feature of LoRaCoin is that it rewards both IoT devices that generate data and gateways that offer the Internet connectivity to the sensor nodes. With such double rewards, LoRaCoin aims to incentivize individuals to host sensor nodes and gateways and contribute to the growing need of the society for environmental monitoring applications.

This poster presents the design, development, and test results of an energy consumption analysis module developed over ns3 Millimeter Wave (mmWave) communication, which can analyze the power consumption characteristics of 5G eNodeB/gNodeB Base Stations. This module is essential for research and exploration of the energy consumption behavior of the 5G communication protocols under the New Radio (NR) technology. To the best of our knowledge, the designed module is the first of its kind that provides a comprehensive energy analysis for the 5G mmWave base stations.

In this paper, we focus on the Short Packet Communication (SPC) in the typical massive Machine Type Communication (mMTC) scenario of 5G/6G networks. For the conventional scheme, tremendous Machine Type Communication Devices (MTCDs) send short status update packets directly to the central Base Station (BS) using the Transmission Control Protocol (TCP), which leads to a huge burden on the BS and may cause severe communication congestion. To solve this problem, we propose a frame-level data aggregation SPC scheme based on the Quick UDP Internet Connection (QUIC) protocol. By using the stream multiplexing feature of QUIC, some MTCDs selected as aggregators receive the short status update packets from their neighboring MTCDs, pack the data into new QUIC packets, and forward these new packets to the BS. The QUIC-based SPC scheme is evaluated in the 5G network environment. It is proved that our scheme reduces the communication overhead of the BS by about 10% and the computing burden of the BS by average 40% in CPU usage.

This paper jointly considers computation offloading, service caching and resource allocation in a three-tier mobile cloud-edge computing structure, in which Mobile Users (MUs) have subscribed to Cloud Service Center (CSC) for computation offloading services and paid related fees monthly or yearly, and the CSC provides computation services to subscribed MUs and charges service fees. The problem is formulated as Mixed Integer Non-Linear Programming (MINLP), aiming to meet the
delay requirements of MUs while reducing the cost of the CSC. Then, an Asynchronous Advantage Actor-Critic-based (A3C-based) method is proposed to solve the optimization problem. The simulation results show that the proposed A3C-based method significantly outperforms the other baseline methods in different scenarios.

It is an important problem to study multi-hop communication networks with unreliable links and various nodes forwarding costs, where the freshness of the messages is significant. On unreliable networks, existing time-sensitive utility-based routing protocols provide efficient routing based on a simple utility model that is linear with time. In this work, we introduce an Age of Information (AoI)-sensitive utility model for unreliable networks, in which each periodically generated message has an attached time-sensitive total utility that decays over time following the AoI model. This model provides a good balance between cost and delay. We propose an optimal routing algorithm that guarantees the total expected utility of the messages would be maximized. Our algorithm maximizes expected utilities by forwarding the messages via nodes along the optimal path whenever the beneficial reward will cover the expected total decrease in utility, and dropping them whenever it would not. Finally, we conduct a simulation to evaluate the effectiveness of our algorithm.

Recently, much progress has been achieved virtualizing edge computing and integrating end systems like modern vehicles as both edge servers as well as users. Previously, it was assumed that all participating vehicles share the same vehicle-to-vehicle (V2V) communication technology to exchange data. Uplinks and downlinks to the cloud or a back end data center are provided by gateway nodes that also have a vehicle-to-cloud (V2C) communication interface. We now go beyond this initial architecture and consider quite heterogeneous communication technologies deployed at each vehicle. In particular, we assume that each vehicle is equipped with either V2V or V2C communication, or both so that it can also act as a gateway between the different worlds. We call the resulting system hybrid micro clouds. In this paper, we present means for hybrid micro cloud formation such that every vehicle can exchange data with other vehicles as well as with the back end data center. In our performance evaluation, we looked at the position error of neighboring vehicles in the local knowledge bases compared to the ground truth as a metric.

The use of drones poses a considerable challenge to people's privacy and security. However, vision, interference, and costs limit existing drone detection methods. In this poster, we explore the use of ubiquitous IoT transceivers to collect LoRa signals with and without a drone in flight and design a neural network to train the data. Our detection accuracy on a single link is close to 95%, and we can roughly track the drone's flight path using the network grid.

When considering multi-value objects, the inevitable unbalanced data distribution is overlooked by the existing truth discovery methods. In this work, we propose a confidence interval based approach (CIMTD) to tackle this issue. We estimate source reliability from two aspects, i.e., the ability to claim the correct number of value(s) and specific value(s). To reflect real reliability for both "big" and "small" sources, confidence intervals of enriched estimation are considered. While estimating source reliability, uncertainty degrees are introduced to model object differences. Confidence intervals are also considered to reflect real uncertainty degrees for both "hot" and "cold" objects. CIMTD outperforms baseline methods on real-world datasets.

To guarantee the public safety, important access sections such as corridors and fire passages are required not to be obstructed, thus, it is of great significance to monitor whether there are illegal static obstacles. Previous approaches usually leverage cameras to keep the target area under constant surveillance, but they encounter serious privacy concerns. In this paper, we present SOD, a non-intrusive Static Obstacle Detection system using acoustic technologies. The basic idea is to detect obstacles first, and then identify whether they are illegally static. Particularly, considering narrow access sections, we exploit one speaker to transmit chirp signals, and one microphone array to receive reflected signals. For obstacle detection, we extract the reflection intensity (RI) to depict the spatial structure of target area, and leverage the differential RI (DRI) to detect candidate obstacles. Further, we calculate the average DRI of multiple microphones to filter fake obstacles and achieve robust performance indoors. For obstacle identification, we propose a stable-window-based method to estimate the lasting time of detected obstacles, and alert the obstruction warning when the duration of certain obstacle exceeds the threshold. We implement SOD, and evaluate it in real world. Experiment results show that we can detect static obstacles within 5m with accuracy over 97%.

With the Industrial Internet of Things springs up, it is necessary to provide a remote and automatic meter reading solution for traditional enterprises and factories to avoid huge labor costs during production. However, traditional object detection solutions cannot be deployed on the current lightweight edge devices due to the high computational complexity of deep neural networks. In this paper, we propose a remote meter reading system, named EdgeMeter, which can provide a robust and realtime meter reading solution for lightweight edge devices. To ensure real-time and high-precision performance, we propose a lightweight feature point matching strategy to amortize the high latency of object detection, and we perform perspective correction to minimize the influence of meter orientations. To further improve the robustness of the system, we design a customized auxiliary device to eliminate the interference of complex outdoor environments. Real-world experiment results show that EdgeMeter achieves the meter reading error within 2.2â—¦ with a latency of less than 42 ms.

This poster explores how to construct a LoRa propagation model using a modest amount of RSS to achieve a complement for a failed GPS in some specific environments. Considering the distribution of LoRa IoT devices and LoRa signal characteristics, we use Thin Plate Spline to construct a 3D model with 36 sampling points to achieve a fit of 99%. On this basis, we further gain a localization error of <10m in the space of 150*50*22m, achieving a good balance between overhead and accuracy.

The ever-increasing number of android malware still poses a critical security challenge to the smartphone ecosystem. Literature suggests that machine and deep learning models can detect android malware with high accuracy and low false positivity. However, the result of arms-race in the adversarial setting of these detection models will shape their integration in real-world applications. Therefore, we first constructed four different malware detection models by applying machine and deep learning algorithms. Then, we stepped into the attacker's (malware developer) shoes and created an adversarial setting framework to investigate the robustness of the above detection models. We developed an evasion attack (Gradient Modification Attack) to exploit the vulnerabilities and force massive misclassifications in the above detection models. The attack drastically reduces the average accuracy of the above four detection models from 95.13% to 59.97%. Later, we also developed a potential defense mechanism (Correlated Distillation Retraining) to mitigate such adversarial attacks. Finally, we conclude that investigation of malware detection models in adversarial settings is essential for improving their robustness and real-world deployment.

Security is of vital importance in wireless industrial communication system. When spoofing attacking has occurred, leading to economic losses or even safety accidents. So as address the concern, existing approaches mainly rely on traditional cryptographic algorithms. However, these method cannot meet the needs of short delay and lightweight. In this paper, we propose a CSI-based PHY-layer security authentication scheme to detect spoofing detection. The main idea takes advantage of the uncorrelated nature of wireless channels to identification of spoof-ing nodes in physical layer. We demonstrate a MIMO-OFDM based spoofing detection prototype in industrial environments. Firstly, utilizing Universal Software Radio Peripheral (USRPs)to establish MIMO-OFDM communication systems is presented. Secondly, our proposed a security scheme of CSI-based PHY-layer authentication is demonstrated. Finally, the effectiveness of the proposed approach has been verified via attack experiments.

This paper proposes a dual-RFID-tag based localization method with multiple-aperture to estimate the location and orientation of the object, which preserves high resolution of the large aperture of the antenna array while eliminating position ambiguity by using the small aperture of the two tags attached to the object. The observations obtained by the large and small apertures are synthesized to form linear equations where the object orientation is first solved and then used to estimate the position. To evaluate its performance, the proposed method is implemented using the commercial RFID equipment. Results show that the proposed method can achieve a localization accuracy of about 9.4 cm in the two-dimensional (2-D) area and outperforms the state-of-the-art antenna-array-based localization methods.

In public clouds, compute instances not sold at regular prices become idle resources, which could be sold at highly reduced prices.
Pricing idle resources is crucial to the cloud ecosystem, but its difficulty is underestimated because user strategies are treated as simple.
In this paper, we model users as smart who want to minimize total cost over time and characterize an extensive-form repeated game between a cloud supplier and multiple users, which captures incentives in real world scenarios like Amazon Web Service spot instance.
We model smart user responses to prices by designing a no regret online bidding strategy and prove that the game among users converges to a coarse correlated equilibrium.
Conversely, we design a robust pricing mechanism for the cloud supplier, which guarantees worst case revenue and achieves market equilibrium.

In this article we present our novel SNE2EE mechanism that is under implementation. The mechanism that offers both software and hardware solutions extends encryption technologies and techniques to the end nodes to increase privacy. The SNE2EE mechanism can tackle spyware and stalkerware at both in individual and community level.

This work concerns the control problem of a networked-based unmanned surface vehicle (USV) system subject to communication delays, external disturbance and false data injection (FDI) attacks. The communication channel between the control station and the actuator is vulnerable to cyber attacks. In this work, we propose an adaptive compensation module to track the given trajectory in presence of FDI attacks. Moreover, an event-triggered mechanism is introduced to handle the the communication delay from the sampler to control station actuator via communication network. Finally, numerical simulation demonstrates that the method we proposed is effective.

Middlebox is primarily used in Software-Defined Network (SDN) to enhance operational performance, policy compliance, and security operations. Therefore, the security of the middlebox itself is essential because incorrect use of the middlebox can cause severe cybersecurity problems for SDN. Existing attacks against middleboxes in SDN, for instance, the middlebox-bypass attack, uses methods such as cloned tags from the previous packets to justify that the middlebox has processed the injected packet. Flowcloak as the latest solution to defeat such an attack creates a defence using a tag by computing the hash of certain parts of the packet header. However, the security mechanisms proposed to mitigate these attacks are compromise-able since all parts of the packet header can be imitated, leaving the middleboxes insecure. To demonstrate our claim, we introduce a novel attack against SDN middleboxes by hijacking TCP/IP headers. The attack uses crafted TCP/IP headers to receive the tags and signatures and successfully bypasses the middleboxes.

Extracting semantic information from trajectory traces (timestamped location information) is a challenging task as conventional information retrieval techniques fail to detect the underlying interpretations of movement history. This work proposes a semantic mobility analytics framework to automatically annotate the trajectories with trip-intent and trip-purposes. Also, transfer learning is proposed to extract and transfer the semantic knowledge from source to target trajectory (region) where labelled data is unavailable.

This paper extends the concept of imagechain by a new feature. Now it is possible to store multiple images per block. If necessary, the block is split into parts and each of them is embedded in an individual image. All images from the previous
block take part in hash computation. Described construction uses secret splitting for creating shares and steganographic algorithms for embedding and recovering data.

Kernel32 is one of the most used dynamic link libraries (DLLs) for application programming interface (API) calls on the Microsoft Windows operating system. Each DLL file contains many functions, and each function has its own memory address once loaded in memory. The API memory address is essential for any API call. In the past, the memory address for each API was fixed to a specific hex value. If an attacker was able to obtain these API addresses on one operating system, it could be used on any other Windows operating system as well. In this paper, we examine two existing methods and propose two novel methods to find kernel32 base address. The objective is to optimally combine all the methods to increase the detection rate of unknown malware, and perform experimental evaluation for malware detection in next-generation communication networks.

In the past few years, researchers have shown a growing interest in techniques for automated detection of regrettable disclosures (things people wish they had not shared) on social media. Most of these proposals formulate the task of automatically detecting potentially regrettable disclosures as a supervised classification problem. In such a setting, the underlying classification model is trained and validate on a dataset labeled accordingly. However, despite growing efforts, existing approaches remain limited, partly due to the lack of high-quality corpus of regrettable messages and comments shared on social media. Previous work tend to confuse regrettable disclosure with related concepts such as hate speech, profanity and offensive language, ignoring empirical findings on the reasons, the types of contents, and disclosure contexts that often lead to regrets. Moreover, corpora used in prior work are typically limited in size and w.r.t. their source domains (i.e., social media platforms) and scope (i.e., range of regret-related topical content used as labels). The goal of this paper is to contribute towards lowering the barrier for developing effective systems for automated detection of regret-related posts. We propose a novel methodology for large-scale data collection and semi-automated annotation. We introduce REGRETS, a new large-scale corpus of 4,7 million regrettable text-only posts and comments with high-quality annotations. Further, we propose regret-specific embeddings models pre-trained on our corpus of user-generated social media texts which were extracted from various popular social media ecosystems. Lastly, we report on analyses that demonstrate the feasibility of partly automating the annotation of social media texts, and the richness of the resulting corpus. We release our findings as resources to facilitate further interdisciplinary research: https://bit.ly/3fO36Ex.