Technical Sessions

Session T1S5

Networking and Management

Conference
9:00 AM — 10:30 AM HKT
Local
Dec 15 Thu, 8:00 PM — 9:30 PM EST

User-Perceived QoE Adaptation for Accelerated Playback in Mobile Video Streaming

Xiongfeng Hu, Yibo Jin, Kefeng Wu, Zhuzhong Qian, Sanglu Lu

1
User-perceived quality of experience (QoE) is critical as mobile video streaming experiences a substantial growth. User¡¯s demands are becoming diversified where accelerated playback is the preference of a considerable part of users. However, the limited and fluctuate mobile bandwidth is often not capable of satisfying user¡¯s demand of watching video at 2x or higher speed because of consequential frequent rebuffering. Previous adaptive bitrate (ABR) algorithms hardly consider the variety of user playback rates. In this work, we fully exploit the relation between user-perceived, i.e., subjective video quality and the characteristic of video con

K-Means Based Grouping of Stations with Dynamic AID Assignment in IEEE 802.11ah Networks

Eduardo C. Oliveira, Stephanie M. Soares, Marcelo M. Carvalho

0
The IEEE 802.11ah amendment extends the success of traditional IEEE 802.11 networks to sub-GHz frequency bands by allowing higher signal coverage and introducing new capabilities to handle a massive number of stations in applications for the Internet of Things. The restricted access window (RAW) mechanism is the key technique to limit contention by dividing stations into RAW groups and RAW time slots according to AID numbers that can be changed dynamically via AID Switch Response frames. To date, however, such feature has not been investigated, especially in cooperation with some grouping scheme. This paper presents the first investigation on the use of AID Switch Response frames to implement dynamic grouping. The goal is to improve throughput fairness among stations that are geographically distributed over some area. The grouping scheme is based on the K-Means algorithm, which takes as input the length and signal strength of data frames received by the AP, and the modulation and coding schemes (MCS) adopted by stations. Simulation results show significant gains in throughput fairness obtained with the proposed grouping scheme compared to random grouping over different network scenarios.

RLRBM: A Reinforcement Learning-based RAN Buffer Management Scheme

Huihui Ma, Du Xu

0
The development of 5G pushes the research community to concentrate on more innovative 5G beyond/6G networks to provide instantaneous connectivity to futuristic applications, such as e-health, autonomous vehicles, and entertainment services broadcasting, etc., which leads to a huge traffic data explosion. How to mitigate the traffic congestion and the bufferbloat problem is a formidable task. The interaction between transport congestion control protocol and the radio access network (RAN) buffer management scheme drastically impacts the congestion and bufferbloat problem. In this paper, we aim to study the RAN buffer management problem from a whole new perspective by leveraging emerging Deep Reinforcement Learning (DRL). We propose a model-free approach, RLRBM, which enables the agent to learn the best buffer size tuning policy as human beings learn skills. We simulate several typical scenarios to evaluate RLRBM. Experimental results show that RLRBM achieves best trade-off between high throughput and low latency.

MilliFit: A Millimeter-Wave Wireless Sensing Based At-Home Exercise Classification

Edward M Sitar IV; Sanjib Sur

0
The proliferation of smart, ubiquitous devices has inspired many researchers to develop at-home personal documentation systems. One application of such systems is at-home exercise monitoring, which is important for remote healthcare and fitness regimens. This work explores a millimeter-wave (mmWave) wireless sensing based at-home exercise monitoring using commodity devices. We leverage the mmWave signals reflected off a person exercising and design a deep-learning network that uses a combination of CNN and LSTM to classify the activities. We evaluate the performance of our classifier extensively, using several input signal representations.

Session Chair

Wenzheng Xu, Sichuan University, China

Session T4S1

Algorithm Based on Big Data

Conference
9:00 AM — 10:30 AM HKT
Local
Dec 15 Thu, 8:00 PM — 9:30 PM EST

Truthful Auction Mechanism for Data Trading with Share-Averse Data Consumers

Zhenni Feng, Qiyuan Wang, Yanmin Zhu

0
In the paper we focus on a promising research problem of data trading, under the scenario that data items can be reproduced easily and inexpensively. Apart from a Bayesian optimal mechanism based data trading approach, we also propose a prior-free data trading approach to organize the data trading process between selfish data owners and share-averse data consumers, with the goal of maximizing revenue of data owners and meanwhile determining the optimal number of data copies. Rigorous theoretical analysis and extensive experiment results are offered to verify the effectiveness of proposed methods in terms of sum of valuations, revenue, individual rationality and incentive compatibility.

Efficient Semantic Segmentation Backbone Evaluation for Unmanned Surface Vehicles based on Likelihood Distribution Estimation

Yulong Zhang, Jingtao Sun, Mingkang Chen, Qiang Wang

0
Obstacle detection using semantic segmentation shows a great promise for unmanned surface vehicles(USVs) in unstable marine environments. Unlike traditional machine learning, semantic segmentation models need to define suitable backbones in advance to extract features of key pixels. However, although the variety and number of backbones are massive, choosing the best one for the developer?€?s environment in the practical application can be a daunting task. Past researches attempt to explore the ranking of backbones in specific scenarios by retraining all mainstream backbone models, which has a certain effect on some single and unchanged land scenes, but cannot be adapted to the unstable marine environment. Therefore, this paper proposes a method to quickly evaluate the suitable backbone, by extracting the representation models of different backbones without retraining and fine-tuning, separating the super-pixels of their feature distribution maps, comparing the features of different models according to likelihood distribution,and finally providing corresponding evaluation scores to give reference for backbone selection. Experimental results show that the proposed approach can provide precise backbone evaluation scores without increasing the computational effort, which can help developers quickly and accurately select the best backbone suitable for their own environment, and further design more accurate semantic segmentation models for unmanned surface vehicles.

Scene Classification through Knowledge Distillation Enabled Parameter-free Attention Model for Remote Sensing Images

Yubing Han, Zongyin Liu, Jiguo Yu, Anming Dong, Huihui Zhang

1
Remote sensing image scene classification is to label remote sensing images as a specific scene category by understanding the semantic information of the images. It is an essential link in remote sensing image analysis and interpretation and has important research value. Convolutional neural networks (CNNs) have been dominant in remote sensing image scene classification due to their powerful feature extraction capabilities. The general trend has been to make deeper and wider CNN architectures to achieve higher classification accuracy. However, these advances to improve accuracy enlarge the network, creating too many parameters and high computational costs. Large models are difficult to deploy on resource-constrained edge devices for practical applications. Furthermore, CNNs can effectively capture local information but are weak in extracting global features. To overcome these drawbacks, we propose a novel knowledge distillation (KD) based method by employing Swin Transformer as a teacher network for guiding MobileNetV2 with Parameter-Free Attention (MobileNetV2-PFA). First, we modify MobileNetV2 by introducing PFA into the inverted bottleneck block; this improvement helps the model learn more latent and robust features without extra parameters. Second, Swin Transformer is an excellent architecture for capturing long-range dependencies via shifted window-based attention. So, we utilize the long-range dependency information from the Swin Transformer to assist MobileNetV2-PFA training through KD. Experimental results on the challenging NWPU-RESISC45 dataset show that the proposed method outperforms the original MobileNetV2 in classification accuracy with low computational consumption.

A Lightweight Deep Learning framework for Human Activity Recognition using Multivariate Time Series

Rui Xi

0
With the rapid development of wearable sensors and ambient sensors, human activity recognition is becoming one of the fundamental research areas in mobile computing. Especially, emerging deep learning has made great progress in feature representations and achieved high recognition accuracy. Still, massive parameters are necessary for the network model, limiting its practical usability in energy-constrained devices. In this work, we present AST, a lightweight deep learning-based human activity recognition framework. The key insight of AST is that each sensor has an association with other sensors at any time step, and the relationships are dynamic and have temporal dependencies. On this basis, we develop a 2D feature extraction model with a handful of parameters that can achieve excellent performance. We conduct some evaluations on public datasets and the results reveal AST?€?s superiority.

Adapitch: Adaption Multi-Speaker Text-to-Speech Conditioned on Pitch Disentangling with Untranscribed Data

Xulong Zhang, Jianzong Wang, , Ning Cheng, Jing Xiao

2
In this paper, we proposed Adapitch, a multispeaker TTS method that makes adaptation of the supervised module with untranscribed data. We design two self supervised modules to train the text encoder and mel decoder separately with untranscribed data to enhance the representation of text and mel. To better handle the prosody information in a synthesized voice, a supervised TTS module is designed conditioned on content disentangling of pitch, text, and speaker. The training phase was separated into two parts, pretrained and fixed the text encoder and mel decoder with unsupervised mode, then the supervised mode on the disentanglement of TTS. Experiment results show that the Adaptich achieved much better quality than baseline methods

Session Chair

Weifeng Sun, Dalian University Of Technology, China

Session T6S1

Smart City

Conference
9:00 AM — 10:30 AM HKT
Local
Dec 15 Thu, 8:00 PM — 9:30 PM EST

Surface Recognition from Wheelchair-induced Noisy Vibration Data: A Tale of Many Cities

Rochishnu Banerjee, Md Fourkanul Islam, Shaswati Saha, Md Osman Gani and Vaskar Raychoudhury

1
Despite the active legislation in many countries supporting the accessibility of public spaces by mobility-impaired users, the reality is far from ideal. Wheelchair users often struggle to navigate the built environment let alone the natural areas. While barriers to wheeled mobility can be caused by broken/uneven surfaces, steep slopes, and unfavorable weather conditions, the effects of many such factors and others are not properly investigated. In this paper, we aim to classify various built and natural surfaces through their characteristic vibration patterns using different deep learning algorithms. The surface vi-bration data is collected from various cities in Europe (including Paris (FR), Mannheim (DE), Dresden (DE), Munich, Nuremberg (DE), and Salzburg (AT)) while a user drives a manual wheelchair attached with three differently oriented smartphones placed at different heights. Extensive experiments show that a Deep Neural Network model classifies surfaces using a denoised dataset with a 98.9% accuracy which is significantly higher than our previous state-of-the-art.

Face Recognition based Beauty Algorithm in Smart City Applications

Ming Tao, Kaiyan Lin

1
Recently, social networking software is widespread used to achieve intelligent life in the construction of smart city, and sharing selfies on social platforms has been a trend of interaction. As a result, beauty software as a popular tool has been widely welcomed by social media users. Along with the development of beauty technology, the popular beauty software in the market include Qingyan, Meitu, VSCO and so on, whose functions continue to be expanded as users demand for image beautification. However, those functions tend to be manually adjusted by users, which would easily result in facial asymmetry. To address this issue, the principles and key points of several classical beauty algorithms are thoroughly investigated in this paper, such as skin beauty, beauty makeup, image artistry, portrait deformation and sharpening filter. In the experiments, the 68 face feature points in Dlib library are used to perform face detection and locate feature points, and Gaussian filtering, bilateral filtering, local translation and bilinear interpolation methods are used to realize the functions of these beauty algorithms. The analysis results have been shown to demonstrate the efficiency of the investigations.

Traffic Event Augmentation via Vehicular Edge Computing: A Vehicle ReID based Solution

Hao Jiang, Penglin Dai, Kai Liu, Feiyu Jin, Hualing Ren, Songtao Guo

1
Traditional traffic event monitoring and detection solutions mainly rely on roadside surveillance cameras. However, existing solutions cannot be applied for traffic event augmentation due to both restricted monitoring angles and limited camera coverage. Therefore, this paper investigates a novel architecture for traffic event augmentation via vehicular edge computing. In particular, multiple vehicles can collaborate with roadside infrastructures for detecting, re-identification and augmenting certain traffic event via vehicle-to-vehicle (V2V) and vehicle-toinfrastructure (V2I) communications. To enable such an application, we formulate the problem of multi-view augmentation task offloading (MATO) by considering the heterogeneous capabilities of vehicles and edge servers, which aims at minimizing average request delay. On this basis, we design the offloading scheduling framework and propose an adaptive real-time offloading algorithm (ARTO), which makes online offloading decision of object detection and re-identification, by balancing real-time workload among heterogeneous devices. Finally, we implement the hardware-in-the-loop testbed for performance evaluation. The comprehensive results demonstrate the superiority of the proposed algorithm in various realistic traffic scenarios.

Real-time Simulation and Testing of a Neural Network-based Autonomous Vehicle Trajectory Prediction Model

Cheng Wei, Fei Hui, Xiangmo Zhao, Shan Fang

0
Autonomous vehicle trajectory prediction is an important component of autonomous driving assistance algorithms (ADAAs), which can help autonomous driving systems (ADSs) better understand the traffic environment, assess critical tasks in advance thus improve traffic safety and traffic efficiency. However, some existing neural network-based trajectory prediction models focus on theoretical numerical analysis and are not tested in real time, leading to doubts about the practical usability of these trajectory prediction models. To address the above limitations, this study first proposes a collaborative simulation environment integrating traffic scenario construction, driving environment perception, and neural network modeling, afterwards used the co-simulation environment for trajectory data and driving environment data collection. In addition, based on the characteristics of the collected data, a trajectory prediction model based on Bi-Encoder-Decoder and deep neural network (DNN) is proposed and pre-trained. Finally, the pre-trained completed model is embedded in the co-simulation environment and tested in real-time with different batches of data. The simulation results show that the proposed trajectory prediction model can predict trajectories well under specific training data batches, and the best performing trajectory prediction model has a prospective time of 4.9 s and a prediction accuracy of 91.55%

Transfer Learning based City Similarity Measurement Methods

Chenxin Qu, Xiaoping Che, Ganghua Zhang

0
In recent years, in order to solve the problem of deep learning in data deficient cities, especially the cold start problem. Researchers put forward a new idea: transfer the model and knowledge from data abundant cities to data scarce cities, also called urban transfer learning. However, in urban transfer learning, the cost for transferring different target cities and source cities cannot be known in advance. In other words, the effectiveness of urban transfer learning need to be improved. In order to solve this problem, we propose a general method for city similarity measurement in urban transfer learning. Through this method, we carry out transfer learning among the cities with higher degree of similarity, which obviously improve the effectiveness of transfer learning at the data level. At the same time, we have also effectively combined this city similarity measurement method with urban transfer learning, and demonstrated the relevant experiment results

Session Chair

Hui Fei, Chang'an University, China

Session T4S2

Prediction, Detection and Classification

Conference
11:00 AM — 12:30 PM HKT
Local
Dec 15 Thu, 10:00 PM — 11:30 PM EST

MSJAD: Multi-Source Joint Anomaly Detection of Web Application Access

Xinxin Chen, Jing Wang, Xingyu Wang, Chengsen Wang, Guosong Lv, Jiankun Li, Dewei Chen, Bo Wu, LianYuan Li, Wei Yu

0
Fixed broadband internet service can provide a stable broadband network of up to 100 megabits or even gigabits and users at home can use fixed broadband service for all kinds of internet surfing, including website and application access, watching videos, playing games, etc. Traditional maintenance for fixed broadband networks primarily uses human manual methods, supplemented by some low-level semi-automation operations. Since the long processes with numerous network elements in the fixed broadband network, it is difficult for traditional operation and maintenance to support effectively with high quality. When abnormalities occur, it is quite manpower cost and time cost to monitor and locate faults. Therefore, to improve the autonomous capability of the fixed broadband network, intelligent operation and maintenance methods are necessary. First of all, a brandnew data pre-process method is proposed to detect anomalies and problems of slow access by selecting web services access commonly visited by users. Secondly, as the fixed broadband network is a multi-level and complex structure with only a small amount of anomaly sample data, we propose a multi-source joint anomaly detection model called MSJAD model on multidimensional features data. The model validation results on real datasets from the real fixed broadband network are state-of-theart. The accuracy rate reaches 98% and the recall is over 99%. We have already begun to deploy the model on the real fixed broadband network and have achieved good feedback.

ResNect: An Accurate and Efficient Backbone Network for Text Detection Model

Bowei Zhang, Weifeng Sun, Minghui Ji and Kelong Meng

0
As an instance segmentation model, Mask R-CNN can be well applied to text detection tasks, but the accuracy and efficiency of its backbone network, such as ResNet or ResNeXt, are relatively low. To improve the accuracy and computational efficiency, we propose a novel backbone network for Mask R-CNN, called ResNect (Residual Network with channel mixing). ResNect increases model accuracy (reflected by the F1 score on MTWI dataset) by mixing multi-scale features, and improves the efficiency (reflected by the runtime tested on CIFAR-100 dataset) of the backbone network by reducing the module expansion. Through these two methods, the computational requirements are reduced while increasing the accuracy. The experimental results show that, compared with the backbone networks ResNet, ResNeXt and Res2Net, the runtime of ResNect tested on CIFAR-100 is reduced by 15.8%, 34.5% and 29.3%, and the Mask R-CNN with ResNect as the backbone network also has the highest F1 score on MTWI.

Multi-timescale History Modeling for Temporal Knowledge Graph Completion

Chenchen Peng, Xiaochuan Shi, Rongwei Yu, Chao Ma, Libing Wu and Dian Zhang

0
Temporal knowledge graph (TKG) has received great attention in recent years. However, the TKG is not always complete due to the missing of important facts, which has seriously hindered its wide application. Inferring missing facts in TKG is a critical and challenging task due to its highly dynamic nature. Most of the existing methods mainly focus on modeling the structural features and temporal dependencies of TKG to solve the temporal knowledge graph completion problem (TKGC). However, those methods only operate at a single timescale without considering the latent time variability of TKG and thus limit the performance of TKGC solutions. Therefore, we propose a novel method named MtGCN (Multi-timescale history modeling framework based on Graph Convolutional Networks) for completing TKG by self-adaptively modeling the multitimescale history of the incomplete TKG. Firstly, MtGCN uses a structural encoder with a graph convolutional network to mine the latent semantic information and structural features of the TKG. Secondly, MtGCN uses GRU-based temporal encoder to learn the historical information at various timescales of the TKG. Finally, it generates effective entity and relation representations to infer the missing facts for the originally incomplete TKG. By conducting comprehensive experiments on 5 public datasets, the experimental results show that our proposed method MtGCN significantly outperforms the baselines by achieving the highest MRR and HITS@1,3,10.

Improving Speech Representation Learning via Speech-level and Phoneme-level Masking Approach

Xulong Zhang, Jianzong Wang, Ning Cheng, Kexin Zhu and Jing Xiao

1
Recovering the masked speech frames is widely applied in speech representation learning. However, most of these models use random masking in the pre-training. In this work, we proposed two kinds of masking approaches: (1) speech-level masking, making the model to mask more speech segments than silence segments, (2) phoneme-level masking, forcing the model to mask the whole frames of the phoneme, instead of phoneme pieces. We pre-trained the model via these two approaches, and evaluated on two downstream tasks, phoneme classification and speaker recognition. The experiments demonstrated that the proposed masking approaches are beneficial to improve the performance of speech representation.

Attention Based End-to-End Network for Short Video Classification

Hui Zhu, Chao Zou, Zhenyu Wang, Kai Xu and Zihao Huang

0
It has been proved that three-dimensional (3D) convolutional kernel can effectively capture local features in the spatiotemporal range of videos, leading to impressive results of various models in video-related tasks. With the introduction of Transformer and the rise of self-attention mechanism, more selfattention models have been used on video representation learning recently. However, there exist limitations of local perception and self-attention operations respectively in both two types of models. Inspired by the global context network (GCNet), we take advantages of both 3D convolution and self-attention mechanism to design a novel operator called the GC-Conv block. The block performs local feature extraction and global context modeling with channel-level concatenation similarly to the dense connectivity pattern in DenseNet, which maintains the lightweight property at the same time. Furthermore, we apply it for multiple layers of our proposed end-to-end network in short video classification task while the temporal dependency is captured via dilated convolutions and bidirectional GRU for better representation. Finally, our model outperforms both stateof- the-art convolutional models and self-attention models on three human action recognition datasets with considerably fewer parameters, which demonstrates the effectiveness.

Session Chair

Hejun Wu, Sun Yat-sen University, China

Session T5S1

Systems

Conference
11:00 AM — 12:30 PM HKT
Local
Dec 15 Thu, 10:00 PM — 11:30 PM EST

Speed Up IPv4 Connections via IPv6 Infrastructure

Ruiyu Fang, Guoliang Han, Xin Wang, Congxiao Bao, Xing Li, Yang Chen

0
Although IPv6 has been proposed to solve the IP address exhaustion problem for decades, the transition process from IPv4 to IPv6 is rather slow due to the possible loss of users and increased costs for ISPs compared with the potential profits. In order to accelerate this process and make full use of the IPv6 network, in this paper, we propose a user-transparent solution named NetBoost by transferring IPv4 traffic through the IPv6 core network. We also implement a simulator called NetBoostSim to further verify the usefulness and prospective performance gain of NetBoost in different network environments. By deploying our system upon both real and simulated network environments, we showcase that better performance for IPv4 endto-end connections can be acquired by utilizing the light-loaded IPv6 network to transfer traffic from heavy-loaded IPv4 core network, using stateless IPv4/IPv6 translation techniques. In this way, our system can serve as an incentive for ISPs to upgrade to pure IPv6 networks gradually without concerns for the user churn.

Lattice-Based Fine-grained Data Access Control and Sharing Scheme in Fog and Cloud Computing Environments for the 6G Systems

Bei Pei, Xianbin Zhou, Rui Jiang

0
The 6G (the sixth generation mobile communication) network is a heterogeneous network, which includes the cloud computing and the fog computing environment. And, the data securities in the 6G Systems may face great challenges. On the one hand, fine-grained data access control in fog and cloud computing environments is essential to the 6G network. On the other hand, quantum computing attacks may cause great security threats for the 6G network. In order to solve the above problems, we propose a lattice-based fine-grained data access control and sharing scheme in fog and cloud computing environments (LB-DACSS) for the 6G network. Firstly, our LBDACSS scheme can achieve secure data sharing and fine-grained access control in fog and cloud computing environments. Secondly, we apply a lattice encryption algorithm to ensure our LB-DACSS scheme can withstand quantum computing attacks. Thirdly, our LB-DACSS scheme can achieve secure attribute revocation. Fourthly, we bind a unique secret value related to the user?€?s attribute to construct the user's secret key. Finally, both formal security analysis and performance analysis show that our LB-DACSS scheme is efficient and secure

Towards Adaptive Quality-aware Complex Event Processing in the Internet of Things

Majid Lotfian Delouee, Boris Koldehofe, Viktoriya Degeler

0
This paper investigates how to complement Complex Event Processing (CEP) with dynamic quality monitoring mechanisms and support the dynamic integration of suitable sensory data sources. In the proposed approach, queries to detect complex events are annotated with consumer-definable quality policies that are evaluated and used to autonomously assign (or even configure) suitable data sources of the sensing infrastructure. We present and study different forms of expressing quality policies and explore how they affect the process of quality monitoring including different modes of assessing and applying quality-related adaptations. A performance study in an IoT scenario shows that the proposed mechanisms in supporting quality policy monitoring and adaptively selecting suitable data sources succeed in enhancing the acquired quality of results while fulfilling consumers?€? quality requirements. We show that the quality-based selection of sensor sources also extends the network?€?s lifetime by optimizing the data sources?€? energy consumption

DDF-GAN: A Generative Adversarial Network with Dual-Discriminator for Multi-Focus Image Fusion

Shiyu Chen, Xin Jin, Qian Jiang, Jie Yang, Ting Cao, Xiuliang Xi, Yunyun Dong

0
Multi-focus image fusion can overcome the issues that optical lens imaging cannot focus multiple targets simultaneously due to the depth of field limitation. In this paper, we propose a generative adversarial network (DDF-GAN) which consists of a generator and two discriminators to directly generate fused images without decision maps and post-processing. In training process, the source all-in-focus image and the fused image generated by generator are used as input to one of the dual discriminators. Meanwhile, the gradient map of the fused image and the source gradient map of the source all-in-focus image are used as input to another discriminator. An adversarial relationship is established to enhance texture details of fused image. In addition, we create a data set and use it as the main training set for the proposed model. Abundant experiments were carried out to verify the availability of our method. Experimental results prove that our method has advantages in subjective visual perception of human and quantitative measurement.

Personalized news headline generation system with fine-grained user modeling

Jiaohong Yao

0
Personalized news headline generation aims to summarize a news article as a news headline, according to the preference of a specific user. It can help users filter their interested news quickly and increase the news click rates for news providers. However, in this field, when learning user interests from their historically clicked news, existing research only learned user interests on word and news level, ignoring sentence level informativeness. This paper proposes a user model, adding sentence-level informativeness to learn user interests, and further guide the news headline generation. To be more detailed, based on attention layers, sentence and news are represented as the weighted sum of word and sentence representations, respectively. To further explore the correlation between different news contents (news title, body, and topic information), the query vectors in the attention layers are replaced by news content. Experiments on the dataset PENS show that the performance of these two models is better than the baseline model on the evaluation metrics ROUGE. Finally, some future directions are discussed, including interactions across informativeness levels and contents.

Session Chair

Xiaodong Wang, Victoria University, Australia

Session T6S2

Smart Home and Healthcare

Conference
11:00 AM — 12:30 PM HKT
Local
Dec 15 Thu, 10:00 PM — 11:30 PM EST

Alz-Sense+: An Auto Time-synchronized Multi-class Algorithm for Dementia Detection

S. M. Shovan and Sajal K. Das

1
Dementia, a cognitive disease that affects more than 50 million people, causes some degree of disability in remembering simple things and following basic instructions with unusual delays. Researchers proposed different pre-clinical methods with mediocre performance leaving the door open for further improvement. One of the most successful pre-clinical tests, SLUMS (Saint Louis University Mental Status), incorporates verbal responses in the form of standardized questionnaires. It involves expert judgment to label patients such as dementia, MCI (Mild Cognitive Impairment), or healthy based on an overall score. However, a nonverbal stress response is also taken into account in the Alz-Sense algorithm, which has a few underlying false assumptions, i) uniformity of answering duration, ii) equity of questions stress level, and iii) unfair stress penalty while discarding healthy patient detection. Moreover, the stress data of the corresponding question is manually synchronized using the examiner?€?s hand-shaken data of the wearable device. As a goal to improve the original Alz-Sense algorithm, Alz-Sense+ is proposed to handle these three assumptions by incorporating the windowing process, statistical and visual approach. Besides, it also automated the synchronization between questions and corresponding sensor data by estimating time slots while proposing an optimal ordering of questions that mitigates the unintended consequences. Alz-Sense+ achieved 81.39%, 80.76%, and 82.35% accuracy, sensitivity, and specificity, respectively, which is 7.39%, 0.01%, and 15.75% improvement over the original Alz-Sense algorithm. In a nutshell, the new Alz-Sense+ algorithm outperformed the existing algorithm by addressing a few underlying assumptions while eliminating a few limitations of the original algorithm.

Towards Socially Acceptable Food Type Recognition

Junjie Wang, Jiexiong Guan, Y.Alicia Hong, Hong Xue and Shuangquan Wang

0
Automatic food type recognition is an essential task of dietary monitoring. It helps medical professionals recognize a user?€?s food contents, estimate the amount of energy intake, and design a personalized intervention model to prevent many chronic diseases, such as obesity and heart disease. Various wearable and mobile devices are utilized as platforms for food type recognition. However, none of them has been widely used in our daily lives and, at the same time, socially acceptable enough for continuous wear. In this paper, we propose a food type recognition method that takes advantage of Airpods Pro, a pair of widely used wireless in-ear headphones designed by Apple, to recognize 20 different types of food. As far as we know, we are the first to use this socially acceptable commercial product to recognize food types. Audio and motion sensor data are collected from Airpods Pro. Then 135 representative features are extracted and selected to construct the recognition model using the lightGBM algorithm. A real-world data collection is conducted to comprehensively evaluate the performance of the proposed method for seven human subjects. The results show that the average f1-score reaches 94.4% for the ten-fold crossvalidation test and 96.0% for the self-evaluation test.

Accurately Identify and Localize Commodity Devices from Encrypted Smart Home Traffic

Xing Guo, Jie Quan, Jiahui Hou, Hao Zhou, Xin He, and Tao He

0
Nowadays, Internet of Things (IoT) based smart home system is equipped with a large number of smart devices, such as smart speakers and cameras, which can greatly facilitate users to control and automate their home environment. However, recent studies have shown that smart home system is at great risk of privacy leakage. Especially, external attackers can infer user privacy information by passively sniffing encrypted smart home network traffic. Traditional methods mainly focus on sniffing WiFi devices but pay less attention to other commodity devices such as Zigbee and Bluetooth (BLE). In this paper, we focus on inferring fine-grained sensitive details about users using diverse commodity devices. We apply deep learning techniques to infer users?€? behaviors through identifying and localizing smart home devices being used due to the excellent performance of deep learning in many fields. Specifically, we first pre-process encrypted device traffic, select valid features, and use Convolutional Neural Networks (CNN) for device identification. In addition, we extract the Received Signal Strength Indicator (RSSI) from the frame information of traffic packets and employ Sparse Autoencoder (SAE) to extract stable and distinguishable high-dimensional features for RSSI measurement. Features are fed into a Multilayer Perceptron (MLP) to predict the device?€?s localization. In this way, we can infer human activity by identifying and localizing the devices being used. Extensive experiment results show that our work can achieve a mean position estimation error of 1.34m even in an unseen environment, outperforming other commonused localization algorithms based on RSSI fingerprints.

Multiuser Collaborative Localization based on Inter-user Distance Estimation using Wi-Fi RSS Fingerprints

Tinghao Qi, Chanxin Zhou, Guang Ouyang and Bang Wang

0
Indoor localization based on Wi-Fi received signal strength (RSS) fingerprints has been widely studied in recent years, mainly focusing on how to improve localization accuracy in an independent way. Some studies propose to use additional hardware devices to measure the distance in between users to improve localization accuracy, but these methods suffer from high cost and low practicality. In order to solve this problem, an interuser distance estimation algorithm iDE is proposed in this paper. We first construct user features based on Wi-Fi fingerprints, then train the random forest and nearest neighbor regressors to obtain inter-user distance estimates, and design a multi-layer perceptron to fuse them. We propose a multiuser collaborative localization MCLoc based on inter-user distance estimation. It takes the distance estimation from iDE as a soft constraint to optimize the user?€?s location using gradient descent search. Experiments in real scenarios show that in terms of inter-user distance estimation, the iDE algorithm can reduce the error by 24.2% compared with the single-model algorithm; in terms of positioning performance, the MCLoc algorithm can reduce the localization error by 11.4% compared with the non-collaborative method.

Analytic Correlation Penalty with Variable Window in Multi-task Learning Disease Progression Model

Xiangchao Chang, Menghui Zhou, Fengtao Nan, Yun Yang and Po Yang

0
Alzheimer?€?s Disease (AD) is the most common reason of dementia that causes serious problems in patients?€? congnitive functions. Multi-task learning (MTL) has performed well in studies of longitudinal processes in Alzheimer?€?s disease for revealing the progression of AD. Combined with prior knowledges in disease progression or medical science, regularization MTL framework could introduce empirical constraints more flexibly. Meanwhile, it brings higher cost during optimization. While it shown that most of formulations could not define the disease progression precisely. Existing regression methods with temporal smoothness method eliminated abnormal fluctuation of cognitive scores, and neglected the sophisticated progression in disease. In this article, we proposed an analytic method to define the progression of AD, and a flexible bandwidth method to encourage the points of disease time sequence temporal smoothness in an appropriate way. To solve three non-smooth penalties in our method, we proposed an optimization method combined accelerated gradient descent (AGD) and alternating direction method of multipliers (ADMM).

Session Chair

Ming Tao, Dongguan University of Technology, China

Session T4S3

Federated Learning

Conference
2:00 PM — 3:30 PM HKT
Local
Dec 16 Fri, 1:00 AM — 2:30 AM EST

Anomaly Detection through Unsupervised Federated Learning

Mirko Nardi, Lorenzo Valerio and Andrea Passarella

0
Federated learning (FL) is proving to be one of the most promising paradigms for leveraging distributed resources, enabling a set of clients to collaboratively train a machine learning model while keeping the data decentralized. The explosive growth of interest in the topic has led to rapid advancements in several core aspects like communication efficiency, handling non-IID data, privacy, and security capabilities. However, the majority of FL works only deal with supervised tasks, assuming that clients?€? training sets are labeled. To leverage the enormous unlabeled data on distributed edge devices, in this paper, we aim to extend the FL paradigm to unsupervised tasks by addressing the problem of anomaly detection (AD) in decentralized settings. In particular, we propose a novel method in which, through a preprocessing phase, clients are grouped into communities, each having similar majority (i.e., inlier) patterns. Subsequently, each community of clients trains the same anomaly detection model (i.e., autoencoders) in a federated fashion. The resulting model is then shared and used to detect anomalies within the clients of the same community that joined the corresponding federated process. Experiments show that our method is robust, and it can detect communities consistent with the ideal partitioning in which groups of clients having the same inlier patterns are known. Furthermore, the performance is significantly better than those in which clients train models exclusively on local data and comparable with federated models of ideal communities?€? partition.

PPFM: An Adaptive and Hierarchical Peer-to-Peer Federated Meta-Learning Framework

Zhengxin Yu, Yang Lu, Plamen Angelov, and Neeraj Suri

0
With the advancement in Machine Learning (ML) techniques, a wide range of applications that leverage ML have emerged across research, industry, and society to improve application performance. However, existing ML schemes used within such applications struggle to attain high model accuracy due to the heterogeneous and distributed nature of their generated data, resulting in reduced model performance. In this paper we address this challenge by proposing PPFM: an adaptive and hierarchical Peer-to-Peer Federated Meta-learning framework. Instead of leveraging a conventional static ML scheme, PPFM uses multiple learning loops to dynamically self-adapt its own architecture to improve its training effectiveness for different generated data characteristics. Such an approach also allows for PPFM to remove reliance on a fixed centralized server in a distributed environment by utilizing peer-to-peer Federated Learning (FL) framework. Our results demonstrate PPFM provides significant improvement to model accuracy across multiple datasets when compared to contemporary ML approaches.

DPFed: Toward Fair Personalized Federated Learning with Fast Convergence

Jiang Wu, Xuezheng Liu, Jiahao Liu, Miao Hu and Di Wu

0
Instead of training a single global model to fit the needs of all clients, personalized federated learning aims to train multiple client-specific models to better account for data disparities across participating clients. However, existing solutions suffer from serious unfairness among clients in terms of model accuracy and slow convergence under non-IID data. In this paper, we propose a novel personalized federated learning framework, called DPFed, which employs deep reinforcement learning (DRL) to identify relationship between clients and enable closer collaboration among similar clients. By exploiting such relationships, DPFed can personalize model aggregation for each client and achieve fast convergence. Moreover, by regularizing the reward function of DRL, we can reduce the variance of model accuracy across clients and achieve a higher level of fairness. Finally, we conduct extensive experiments to evaluate the effectiveness of our proposed framework under a variety of datasets and degrees of non-IID data distribution. The results demonstrate that DPFed outperforms other alternatives in terms of convergence speed, model accuracy, and fairness.

Dynamic Unknown Worker Recruitment for Heterogeneous Contextual Labeling Tasks Using Adversarial Multi-Armed Bandit

Wucheng Xiao, Mingjun Xiao, Yin Xu

0
Nowadays, crowdsourcing has become an increasingly popular paradigm for large-scale data annotation. It is crucial to ensure label quality by selecting the most suitable workers for labeling tasks. Many previous works have studied the reliability of unknown workers for crowdsourcing tasks with a stochastic assumption. However, each worker?€?s reliability varies when performing tasks with different categories. Meanwhile, the reliability of each worker is usually unknown and doesn?€?t follow any stochastic distribution. In this paper, we propose an Adversarial multi-armed Bandit-base algorithm to handle the Unknown Worker Recruitment (ABUWR) problem without any prior stochastic assumption. In ABUWR, we determine suitable workers for each task to maximize the accumulated average accuracy of the labeling tasks under a limited budget. Specifically, we model this unknown worker recruitment problem as an adversarial multi-armed bandit game and use the least confidence scheme to ensure the total accumulate accuracy. Meanwhile, we theoretically prove that ABUWR has a sub-linear regret upper bound. Furthermore, we demonstrate its significant performance through extensive simulations on real-world data traces.

Session Chair

Zhenni Feng , Donghua University, China

Session T5S2

Testbed and Simulation

Conference
2:00 PM — 3:30 PM HKT
Local
Dec 16 Fri, 1:00 AM — 2:30 AM EST

DiNS: Nature Disaster in Network Simulations

Nisal Hemadasa, Wanli Yu, Yanqiu Huang, Leonardo Sarmiento, Amila Wickramasinghe and Alberto Garcia-Ortiz

1
Wireless sensor networks (WSNs) is a promising solution for disaster management because of its scalability and low cost operations. However, testing the effectiveness of WSNs in real-world disasters is time consuming, costly and in some cases even infeasible. In this paper, we propose a Disaster in Network Simulations (DiNS) framework based on OMNeT++ to replicate a WSN deployed in a disaster in a simulated environment. DiNS allows researchers to observe how the disaster influences the sensor network and how the network can respond in return. A generic coupling interface is developed to support different disaster types. Moreover, we develop a verification tool for functional debugging and verification during new functions developing of sensor nodes, and an optimization tool to support the mathematical optimization of the network operations in response to the disaster. The functionality of DiNS is demonstrated with a case study using wildfire disaster. It provides an easy way for validating and optimizing the disaster management with WSNs.

UAV Swarm Trajectory and Cooperative Beamforming Design in Double-IRS Assisted Wireless Communications

Yangzhe Liao, Shuang Xia, Ke Zhang and Xiaojun Zhai

0
Non-terrestrial communications have emerged as a technological enabler for seamless connectivity and ubiquitous computation services in the upcoming beyond fifth generation (B5G) and sixth generation (6G) networks. However, there exist numerous practical technical limitations, such as high deployment cost, massive energy consumption, high probability of information transmission blockage and dynamic propagation environments and so forth. Thanks to the rapid developments of meta-materials, the cost-effective and energy-efficiency intelligent reconfigurable surface (IRS) has been globally recognized as a revolutionized technology to construct smart radio environments. In this paper, a novel double-IRS assisted unmanned aerial vehicles (UAV)-swarm-enabled communication network architecture is proposed, where two UAV swarms are integrated with the main IRS reflector and subreflector, respectively. The energy minimization problem of UAV swarm carried main IRS is formulated, subject to a list of quality of service (QoS) constraints. To tackle the formulated challenging problem, we first decouple the original problem into two subproblems. Then, a heuristic algorithm is proposed, where the enhanced differential evolution (DE) algorithm is proposed to optimize the UAV swarm trajectory and the alternate optimization algorithm is utilized to optimize the cooperative reflect beamforming vector. Numerical results validate that the proposed algorithm outperforms several selected advanced algorithms regarding UAV swarm energy consumption. Moreover, the network performance under the different number of IRS elements is investigated.

Towards Energy-efficient Container Data Center: An Online Migratability-aware Orchestrator

Shengjie Wei, Jiayi Li, Tuo Cao, Sheng Zhang, and Zhuzhong Qian

0
The growing demands for cloud computing have led to high power consumption in data centers. To save power, existing works attempt to reduce the number of active servers via scheduling or migrating VMs. Meanwhile, owing to the lightweight, highly-portable and scalable properties, containers have been widely used in data centers. However, comparing with VMs, containers show the migratability property, i.e., some containers can not be consolidated by migration. Thus, one has to take this property into consideration when making scheduling and migrating decisions. To this end, this paper studies the energy-efficient container orchestration problem in data centers. We frame container scheduling and migration as a single problem, and take the migratability of containers into account. We propose an online container orchestration algorithm, based on Lyapunov optimization and Markov approximation. It works without requiring future information and achieves a provable performance guarantee. Simulation results show that our algorithm can effectively reduce the power consumption of data centers.

InstaVarjoLive: An Edge-Assisted 360 Degree Video Live Streaming for Virtual Reality Testbed

Pengyu Li, Feifei Chen, Rui Wang, Thuong Hoang and Lei Pan

0
Virtual Reality (VR) challenges us with the requirements of ultra-low latency and ultra-high bandwidth. Existing methods that rely on cloud computing systems to improve the latency and bandwidth problems cannot satisfy the high computation and fast communication requirements in VR. Edge computing has emerged as a promising solution that can be applied in VR to optimize the latency and bandwidth problems. However, another challenge is applying edge computing technology to improve the seamless for the VR users. Based on this, this paper proposes an edge-end collaboration testbed called InstaVarjoLive and conducts the experiments on the realtime 360?? video live streaming seamless using VR headsets. We compared our experiments with the 360?? videos watched by users from the cloud through VR headset and obtained the results showing that the edge-assisted 360?? videos live streaming method has three major advantages: better real-time delivery, lower response time, and higher bandwidth guaranteed. Furthermore, we tested our experiments and discussed the other possible optimization methods in the future.

Mobile6TiSCH: a Simulator for 6TiSCH-based Industrial IoT Networks with Mobile Nodes

Marco Pettorali, Francesca Righetti, and Carlo Vallati

0
The Internet Engineering Task Force (IETF) has defined the 6TiSCH architecture to enable the Industrial Internet of Things (IIoT). While many industrial applications involve mobile devices (e.g., mobile robots or wearable devices carried by workers), 6TiSCH does not provide any mechanism to manage node mobility. However, recently, a Synchronized Single-hop Multiple Gateway (SHMG) architecture has been proposed to allow an efficient management of mobile nodes in 6TiSCH networks. The SHMG architecture is quite flexible and can be customized to meet the requirements of different IIoT applications. However, find the appropriate configuration to guarantee the application requirements, may not be trivial. In this paper, we present Mobile6TiSCH, a simulation tool based on OMNeT++, that implements the SHMG architecture. The proposed tool is general and allows to evaluate the Quality of Service (QoS) achieved by mobile nodes, in different scenarios. As such, it can be used to evaluate different solutions in a simulated environment before implementing them in practice. We describe the organization and implementation of Mobile6TiSCH and show the simulation results to display its effectiveness.

Session Chair

Lei Pan, Deakin University, Australia

Session T6S3

RFID and Optimization

Conference
2:00 PM — 3:30 PM HKT
Local
Dec 16 Fri, 1:00 AM — 2:30 AM EST

IMRG: Impedance Matching Oriented Receiver Grouping for MIMO WPT System

Lulu Tang, Hao Zhou, Weiming Guo, Wangqiu Zhou, Xing Guo and Xiaoyan Wang

0
In recent years, multiple-input multiple-output (MIMO) technology has been imported into magnetic resonance coupled (MRC) enabled wireless power transfer (WPT) systems for concurrent charging of multiple devices. Besides the traditional performance optimization methods (e.g., TX current scheduling, system frequency adjustment, etc.), receiver (RX) grouping will also severely influence the achieved powerdelivered-to-load (PDL). In this paper, we investigate the optimal RX grouping issue to maximize the proportional fairness of RX achieved PDL, which is a joint optimization problem involving RX grouping and time-slice allocation among groups. By decoupling the problem, we solve the group generation subproblem with a impedance-matching based greedy algorithm to generate potential RX group candidates, and we further solve the time slice allocation sub-problem with a genetic algorithm to distribute resources among group candidates. We prototype the proposed system, denoted as IMRG, and conduct extensive experiments to evaluate the performance. The experimental results validate the effectiveness of the proposed algorithm, e.g., IMRG achieves average 59.6% PDL improvement through RX grouping compared to the simultaneous charging scheme.

Compact Unknown Tag Identification for Large-Scale RFID Systems

Kai Lin, Honglong Chen, Na Yan, Zhichen Ni, Zhe Li

0
Nowadays, Radio Frequency IDentification (RFID) technology is profoundly affecting all walks of life. Unknown tag identification, as an important service for RFID-enabled applications, aims to exactly collect all EPCs (Electronic Product Code) of unknown tags that are not recorded by the back-end server in the RFID systems. Efficient unknown tag identification is significant to accurately discover the unregistered or newly entering tags in many scenarios, such as warehouse management and retail industry. However, the replies of known tags and the unpredictable behaviors of unknown tags bring serious challenges for accurate and efficient identification of unknown tags. To handle these tough issues, we propose a Compact Unknown Tag identification protocol (CUT) to collect unknown tag EPCs in large-scale RFID systems. Firstly, we introduce a compact indicator vector to simultaneously label unknown tags and deactivate known tags. Then the unknown tags are instructed to reply their EPCs via another compact reply based indicator vector. In each indicator vector, the amount of expected empty and singleton slots is increased to greatly improve the labeling, deactivation and collection efficiency. After that, we validate the effectiveness of proposed CUT protocol by extensive theoretical analyses and simulations. The simulation results demonstrate that CUT protocol outperforms the state-of-the-art one.

Analytic Hierarchy Process Based Compatibility Measurement for RFID Protocols

Weiping Zhu, Changyu Huang, and Chao Ma

0
In recent years, radio frequency identification (RFID) based information query is widely used in many applications. In order to meet various application requirements, different kinds of RFID protocols are proposed, such as ID collection, category estimation, and missing tag identification. We find that the structure and function used in these RFID protocols are quite similar. For example, empty slot skipping and collision slot reconciling are used in many protocols. An improvement in one protocol may also be applied in another protocol, or a combination of two compatible protocols can fulfill a new application requirement. However, currently there is no approach to measure the compatibility of two RFID protocols. In this study, we theoretically proposed the concept of RFID protocol compatibility and designed an approach to measure it. Analytic hierarchy process approach is revised for this purpose. The important features of an RFID protocol are identified, and then determine their weights according to their importance. The similarity of two protocols are computed by the weighted similarity of lowest level features. We validate this approach by using eight typical RFID protocols, and show useful information for the protocol design. For example, the results show the compatibility between CLS and SFMTI reaches 89.5%, while the compatibility between CLS and TKQ is only 20.71%, this conforms the characteristics of these protocols.

Trajectory Optimization Model of Connected and Autonomous Vehicle at Unsignalized Intersections

Yu Kong, Chen Mu, Xinyu Chen

0
Intersections are typical bottle necks in urban traffic system. Unsignalized intersection technology, together with the Connected and Autonomous vehicles (CAVs), has a great potential to alleviate traffic congestion and has gained rapidly increasing interest. This work presents a Mixed Integer Nonlinear Programming (MINLP) model to generate the CAV trajectories that maximize average speed, so as to guide CAVs to pass an unsignalized intersection in a safe and efficient manner. Specifically, potential conflict points between CAVs from different lanes are analyzed. Upon that, the realistic constraints, e.g. vehicle kinematic limits and collision avoidance under different conflict scenarios, are established. To carry out collision avoidance mechanism and control the time at which CAVs pass the conflict point, we reformulate the arc-shaped CAVs?€? trajectory at an unsignalized intersection into a straight line. Numerical examples are conducted on several representative applications and demonstrate the effectiveness of proposed mathematical model.

Node Selection Strategy Design Based on Reputation Mechanism for Hierarchical Federated Learning

SHEN Xin, LI Zhuo, CHEN Xin

0
With the rapid development of Internet of Things (IoT) and 5G wireless communication technology, a large amount of data is generated at the edge of the network. The combination of mobile edge computing (MEC) and federated learning has become a key technology to improve performance and protect users?€? privacy data in mobile networks. The selection of nodes for Hierarchical Federated Learning (HFL) affects the quality of model training. In this paper, we investigate the optimization problem of node selection accuracy in HFL. In order to improve the quality of model training, we design an algorithm of node selection based on reputation (NSRA). In NSRA, the edge server selects the node with high reputation prediction value to participate in the model training, and the node selects the neighbor node with high transmission capacity to cooperate. D2D communication is adopted for node cooperation. Through extensive simulations, it is verified the performance of NSRA. The mutual trust between nodes is enhanced, so the ideal prediction effect is achieved. We also observe that compared with RSA, the accuracy is improved by 11.48% and 19.38% in MNIST and CIFAR-10, respectively.

AttachSFC: Optimizing SFC Initialization Process through Request Properties

Kaiwen Ning, Hao Wang, Zhiheng Zhang, Zhou Xu, Xiaowei Shu

0
Network Function Virtualization (NFV) technology enables the decoupling of Network Functions (NFs) from hardware by initializing them into virtual machines or containers, which can provide more flexible and customized services to users. However, in current NFV networks, the NFV Orchestrator (NFVO) needs to initialize an Service Function Chain (SFC) containing one or more Virtual Network Functions (VNFs) for each request. This is not appropriate in most cases. On the one hand, the initialization of VNFs also needs to be delayed, which can affect the quality of service to some extent. On the other hand, most of the requests in the network are low-resource and short-occupancy, and such SFCs will be born and died frequently in the network, which will affect the stability of the network. Unfortunately, optimising VNF initialization is a necessary but easily neglected issue. In this paper, we design AD-NFVO for Internet Service Providers (ISPs) to simplify the initialization process of SFCs and increase the overall reward of the network. Firstly, we classify user requests in terms of time and size, respectively, and propose AttachSFC, an adaptive SFC attachment strategy to simplify the initialization process of SFCs. Secondly, we apply AttachSFC to NFVO, and since AttachSFC and SFCs deployment is interdependent, we use the Advantage Actor Critic (A2C) strategy to optimize them to increase the overall reward of ISPs. Finally, we experimentally demonstrate the effectiveness and potential of the AD-NFVO.

Session Chair

Zhuo Li, Beijing Information Science and Technology University, China

Session T4S4

Reinforcement Learning

Conference
4:00 PM — 5:30 PM HKT
Local
Dec 16 Fri, 3:00 AM — 4:30 AM EST

An Opponent-Aware Reinforcement Learning Method for Team-to-Team Multi-Vehicle Pursuit via Maximizing Mutual Information Indicator

Qinwen Wang, Xinhang Li, Zheng Yuan, Yiying Yang, Chen Xu, and Lin Zhang

0
The pursuit-evasion game in Smart City brings a profound impact on the Multi-vehicle Pursuit (MVP) problem, when police cars cooperatively pursue suspected vehicles. Existing studies on the MVP problems tend to set evading vehicles to move randomly or in a fixed prescribed route. The opponent modeling method has proven considerable promise in tackling the non-stationary caused by the adversary agent. However, most of them focus on two-player competitive games and easy scenarios without the interference of environments. This paper considers a Team-to-Team Multi-vehicle Pursuit (T2TMVP) problem in the complicated urban traffic scene where the evading vehicles adopt the pre-trained dynamic strategies to execute decisions intelligently. To solve this problem, we propose an opponentaware reinforcement learning via maximizing mutual information indicator (OARLM2I2) method to improve pursuit efficiency in the complicated environment. First, a sequential encodingbased opponents joint strategy modeling (SEOJSM) mechanism is proposed to generate evading vehicles?€? joint strategy model, which assists the multi-agent decision-making process based on deep Q-network (DQN). Then, we design a mutual informationunited loss, simultaneously considering the reward fed back from the environment and the effectiveness of opponents?€? joint strategy model, to update pursuing vehicles?€? decision-making process. Extensive experiments based on SUMO demonstrate our method outperforms other baselines by 21.48% on average in reducing pursuit time. The code is available at https://github. com/ANT-ITS/OARLM2I2.

Graded-Q Reinforcement Learning with Information-Enhanced State Encoder for Hierarchical Collaborative Multi-Vehicle Pursuit

Yiying Yang, Xinhang Li, Zheng Yuan, Qinwen Wang, Chen Xu, and Lin Zhang

0
The multi-vehicle pursuit (MVP), as a problem abstracted from various real-world scenarios, is becoming a hot research topic in the Intelligent Transportation System (ITS). The combination of Artificial Intelligence (AI) and connected vehicles has greatly promoted the research development of MVP. However, existing works on MVP pay little attention to the importance of information exchange and cooperation among pursuing vehicles under the complex urban traffic environment. This paper proposed a graded-Q reinforcement learning with informationenhanced state encoder (GQRL-IESE) framework to address this hierarchical collaborative multi-vehicle pursuit (HCMVP) problem. In the GQRL-IESE, a cooperative graded Q scheme is proposed to facilitate the decision-making of pursuing vehicles to improve pursuing efficiency. Each pursuing vehicle further uses a deep Q network (DQN) to make decisions based on its encoded state. A coordinated Q optimizing network adjusts the individual decisions based on the current environment traffic information to obtain the global optimal action set. In addition, an informationenhanced state encoder is designed to extract critical information from multiple perspectives and uses the attention mechanism to assist each pursuing vehicle in effectively determining the target. Extensive experimental results based on SUMO indicate that the total timestep of the proposed GQRL-IESE is less than other methods on average by 47.64%, which demonstrates the excellent pursuing efficiency of the GQRL-IESE. Codes are outsourced in https://github.com/ANT-ITS/GQRL-IESE.

Learning-based Dwell Time Prediction for Vehicular Micro Clouds

Max Schettler, Gurjashan Singh Pannu, Seyhan Ucar, Takamasa Higuchi, Onur Altintas, Falko Dressler

0
Vehicular Micro Clouds (VMCs) are an emerging development in the domain of vehicular networks posed to provide local services to users without the need for external infrastructure. This can significantly improve the user experience, in particular due to the low latencies that such systems can achieve. Due to the distributed nature of such a VMC, effective local coordination is important while using minimal communication resources. To this end, it is important to know, how long vehicles will be participating in, and contributing to a VMC. In this work, we investigate, how previous, heuristic-based approaches can be improved by incorporating local, learning-based techniques. Our analysis indicates a potential improvement of the accuracy of the prediction, and resulted in an improved simulation environment within which the learning-based approach can be deployed.

A Motion Propagation Prediction based Sim2Real Strategy Migration for Clutter Removal

Jiaxin Zhang, Ping Zhang

0
When objects are densely placed, training in the simulation with artificial samples and removing clutter are helpful to reduce the cost and risk. However, the performance of control strategy decreases in sim2real is still a challenge. This paper introduces a clutter removal method of sim2real using object motion propagation prediction. In this method, based on deep reinforcement learning, push and grasp actions are used to remove clutter. The reward of push action is calculated based on the object divergence of quadtree. The action strategy is trained in the simulation environment. Due to the position error caused by the robot pushing the object in the simulation and real environment, the object motion propagation prediction network based on graph neural network is used to predict the pushing results in the real environment and replace the real push action to training pushing strategy to improve the reward value. The pushing strategy learned in the simulation is subject to finetuning based on differential evolution. Compared with applying the action strategy directly to the real environment, the method in this paper has higher action efficiency and completion rate.

Session Chair

Anming Dong, Shandong Polytechnic University, China

Session T6S4

Multimedia Application

Conference
4:00 PM — 5:30 PM HKT
Local
Dec 16 Fri, 3:00 AM — 4:30 AM EST

Throughput Prediction-Enhanced RL for Low-Delay Video Application

Yong Liu, Chaokun Zhang, , Jingshun Du, Tie Qiu

0
Maximizing user quality of experience (QoE) is the ultimate goal of video players, and adaptive bitrate (ABR) is recognized as one of the most effective solutions. Approaches employing reinforcement learning (RL) have performed well as hybrid ABR algorithms, due to the ability to learn autonomously. However, throughput, which plays a crucial role in low-delay video streaming, is difficult to predict simply in mobile and wireless networks, and the inaccurately predicted throughput can lead to the wrong selection of bitrates. Worse, the general RL approaches are prone to frequent bitrate switching due to bandwidth fluctuation. These obstacles make the RL-based ABR approach unable to truly reflect the user QoE. We propose TPRL, an application that makes ongoing decisions to maximize user QoE. To realize this, TP-RL adopts three ideas: (i) It takes the RL neural network as the main body of decision-making, which will inherit the advantages of RL and improve on this basis; (ii) Explore Mogrifier LSTM for throughput prediction, and replace the throughput part in the state space of the original RL neural network with a prediction module; (iii) The decided bitrate is further processed to achieve better smoothness when the bandwidth fluctuates. The performance of TP-RL is evaluated in different experimental environments, and experiments show that it can improve QoE by about 14% to 20.7% compared with the best baseline.

Towards Reliable AI Applications via Algorithm-Based Fault Tolerance on NVDLA

Mustafa Tarik Sanic, Cong Guo, Jingwen Leng, Minyi Guo, Weiyin Ma

2
With the development of deep neural networks (DNNs), more complex accelerators have been designed for more sophisticated networks. Naturally, the complexity of accelerators makes them vulnerable to transient errors. Also, some DNN accelerators are widely used the safety-critical systems, such as autonomous vehicles. Therefore, the susceptibility to transient errors makes research on mitigation techniques more significant, and errors of accelerators should be limited to none. Some researchers proposed the modular redundancy method, which offers a highly reliable way but also considerably increases overhead. In this regard, algorithm-based solutions offer cheaper solutions. However, their implementation is primarily observed in software-based error injections. In this study, we propose a novel approach that focuses on implementing algorithm-based error detection (ABED) for RTL-level (hardware-based) error injections. Previous studies generally focused on the impact of soft errors in memory structures of embedded system-based accelerators. However, the main goal of this research is to study the impact of soft errors in processing elements and how to mitigate them. We implement an algorithm-based error detection that utilizes checksums for verifying convolution operations with low overhead. We first explain how to overcome the challenges of implementing ABED on FPGA-based accelerators, then how to implement it. We implement and evaluate our solution on an industry-level DNN accelerator called NVIDIA deep learning accelerator (NVDLA). In this study, our error injection method is constructed to test the most common soft error scenarios in processing units. The results of the research show that algorithm-based fault tolerance can detect all silent data corruptions (SDC) while maintaining a very low overhead (6-23%) on runtime.

Adaptive Progressive Image Enhancement for Edge-Assisted Mobile Vision

Daipeng Feng, Liekang Zeng, Lingjun Pu, Xu Chen

0
Recent advances in deep learning models have pushed Super-Resolution (SR) techniques to an unprecedented altitude, enabling high-quality image rendering with variable scaling size and natural fidelity. To deploy them on resourceconstrained mobile devices, however, confronts significant challenges of excessively long latency and poor user experience. To this end, we propose Apie, an edge-assisted adaptive image rendering system that allows low-latency, progressive image enhancement for a smooth user experience. Apie adopts a data parallel strategy across the end device and the edge server, along with a residual learning mechanism to judiciously retrieve information for SR models. Besides, a novel progressive image reconstruction is developed by exploiting content-aware image blocking and incremental image rendering, towards improved quality of user experience. Furthermore, Apie can dynamically adjust the choice of employed SR models with respect to the networking conditions, striking a good balance upon the latency-quality trade-off. Extensive evaluations show that Apie performs 7.33?? faster than on-device GPU execution and 1.42?? faster compared to the partial offloading method, while achieves 2.84dB higher PSNR compared to the interpolation method using conventional JPEG image compression and 0.74dB higher PSNR compared to the partial offloading method.

PeTrack: Smartphone-based Pedestrian Tracking in Underground Parking Lot

Xiaotong Ren, Shuli Zhu, Chuize Meng, Shan Jiang, Xuan Xiao, Dan Tao, and Ruipeng Gao

0
Although location awareness is prevalent outdoors due to GNSS systems and devices, pedestrians are back into darkness in indoor buildings such as underground parking lots. Frequently we forget where we park the car and get confused by such maze-like structure. In order to track pedestrians without any additional equipment and map support, we propose PeTrack which is a smartphone-only approach that collects the inertial measurement unit (IMU) data for long-term tracking. Our intuition is to train the tracking model with crowdsourced outdoor trajectories, and infer customized user?€?s trace with only inertial readings at indoors. Specially, we propose an inertial sequence learning framework with outdoor geo-tags. We also exploit opportunistic landmark detection and structure cues to refine the trajectory. We have developed a prototype and conducted experiments in an underground parking lot, and results have shown our effectiveness.

Session Chair

Cong Guo, Shanghai Jiao Tong University, China

Session T6S5

Other Areas

Conference
4:00 PM — 5:30 PM HKT
Local
Dec 16 Fri, 3:00 AM — 4:30 AM EST

Joint Convolutional and Self-Attention Network for Occluded Person Re-Identification

Chuxia Yang, Wanshu Fan, Dongsheng Zhou, Qiang Zhang

1
Occluded person Re-Identification (Re-ID) is built on cross views, which aims to retrieve a target person in occlusion scenes. Under the condition that occlusion leads to the interference of other objects and the loss of personal information, the efficient extraction of personal feature representation is crucial to the recognition accuracy of the system. Most of the existing methods solve this problem by designing various deep networks, which are called convolutional neural networks (CNN)-based methods. Although these methods have the powerful ability to mine local features, they may fail to capture features containing global information due to the limitation of the gaussian distribution property of convolution operation. Recently, methods based on Vision Transformer (ViT) have been successfully employed to person Re-ID task and achieved good performance. However, since ViT-based methods lack the capability of extracting local information from person images, the generated results may severely lose local details. To address these deficiencies, we design a convolution and self-attention aggregation network (CSNet) by combining the advantages of both CNN and ViT. The proposed CSNet consists of three parts. First, to better capture personal information, we adopt DualBranch Encoder (DBE) to encode person images. Then, we also embed a Local Information Aggregation Module (LIAM) in the feature map, which effectively leverages the useful information in the local feature map. Finally, a Multi-Head Global-to-Local Attention (MHGLA) module is designed to transmit global information to local features. Experimental results demonstrate the superiority of the proposed method compared with the stateof-the-art (SOTA) methods on both the occluded person Re-ID datasets and the holistic person Re-ID datasets.

A DMA-based Swap Mechanism of Hybrid Memory System

Weijie Zhang, Lidang Xu, Dingding Li, Haoyu Luo

0
Typical applications of smart cities, such as smart public services, require a large memory footprint to store user data and facilitate the responsive results of user queries, thus inevitably activating the memory swap mechanism between memory and storage to expand the capacity of main memory. Frequent page swapping can cause performance interference for hard real-time operating systems such as SylixOS. In a hybrid memory architecture, namely the novel persistent memory (PM) alongside the conventional DRAM, the swap mechanism often uses the PM to act as the swap partition and executes memory copying to transfer the data between DRAM and PM, resulting in frequent I/O operations and high CPU consumption. Eventually, the memory performance is sub-optimal. By leveraging a general DMA technology of memory-to-memory (M2M), namely Intel I/OAT, we propose PM-Swap, a swap mechanism without heavy CPU consumption. PM-Swap further contains three techniques: (1) a new memory reclamation algorithm based on instruction sampling and page awareness, which reduces the unnecessary swap operations; (2) according to the data size, a switching strategy selects the suitable swapping path between the original CPU and the DMA, to maintain reasonable memory performance; (3) bulk transferring is employed for improving the overall throughput of the page swapping. We implement PM-Swap in a stable Linux kernel (5.17.9). The experimental results show that PM-Swap can decrease CPU overhead by more than 39% and increase page swapping bandwidth by up to 1.76??.

CUE: compound uniform encoding for writer retrieval

Jiakai Luo, Hongwei Lu, Xin Nie, Shenghao Liu, Xianjun Deng, Chenlu Zhu

0
Writer retrieval is crucial in document forensics and historical document analysis. However, due to the difference in syntactic structure between Chinese and other languages, the existing methods may not be directly applied to Chinese writer retrieval. Previous work on Chinese writer retrieval does not overcome the performance degradation problem when the number of samples grows. In this paper, we propose a novel compound uniform encoding algorithm (CUE) for Chinese writer retrieval, which mainly consists of a combined feature extraction module (CFE) and a prototype substitution module (PS). The CFE module combines two complementary features from image filter response and character contour. It counts local symmetries and edge co-occurrence pairs. PS module substitutes the outliers with the class prototypes to alleviate the influence of the outliers. Finally, the weighted Chi-square distance is applied to measure the similarity between writer and text. To verify the superiority of our proposed method, experiments are conducted on four public datasets and our built dataset. The results validate that CUE outperforms the state-of-the-art algorithms on mAP metric

A Temperature Prediction-Assisted Approach for Evaluating Propagation Delay and Channel Loss of Underwater Acoustic Networks

Rui Gao, Jun liu, Shanshan Song, En Wang, Yu Gou, Tong Zhang and Jun-hong Cui

0
Propagation delay and channel loss are two vital factors affecting reliability of Underwater Acoustic Networks (UANs). Different from land networks, UANs have long propagation delay and poor channel quality, which lead to serious data collision and high bit error rate, respectively. However, complex underwater environments impose great challenges to evaluate propagation delay and channel loss. As temperature is the most critical factor affecting them, in this paper, we propose to employ temperature to evaluate them. However, existing temperature prediction research are insufficient for accuracy or efficiency. This paper proposes a temperature prediction-assisted approach for evaluating propagation delay and channel loss, aiming to improve reliability and performance of underwater acoustic networks. We build a nonlinear autoregressive dynamic neural network-based temperature prediction model to improve prediction accuracy and reduce time complexity. Then, we evaluate propagation delay and channel loss considering different marine environments, including shallow and deep sea. Extensive simulation results show that our approach performs better than five advanced baselines.

A Hybrid Link Connectivity Model for Opportunistic Routing in IoV Networks under Viaduct Scenarios

Xing Tang, Yongbiao Tao, Wei Liu, Bing Shi, Jing Wang

0
Viaduct structures have been built to ease traffic congestion in large-sized cities, especially in many large-medium cities in China. In Internet of Vehicles (IoV) networks, interlevel links in urban viaduct scenarios are established between two vehicles on different levels with various heights. The bridge surface of the viaduct acts as an obstacle in the propagation paths of inter-level links and may cause a degradation of the link quality. However, inter-level links may bring more communication opportunities when the traffic is sparse or no intra-level link can be found. Compared with traditional three-dimensional wireless networks, inter-level links for viaduct shadowing scenarios reveal unique characteristics which are not addressed by previous studies. This paper aims to estimate the unique link connectivity characteristics and figure out whether inter-level links can enhance opportunistic routing performance. To this end, we first build an accurate hybrid link availability and reliability model by considering the unique structure of viaducts. Second, we design a new metric over opportunistic routing combined with our link model. Simulation results show that taking advantage of inter-level links can improve the performance of opportunistic routing.

Session Chair

Dingding Li, South China Normal University, China

Made with in Toronto · Privacy Policy · © 2022 Duetone Corp.