3rd International Workshop on Artificial Intelligence Applications in Internet of Things

Session AI2OT


10:30 AM — 12:00 PM HKT
Dec 13 Tue, 9:30 PM — 11:00 PM EST

Image Classification of Alzheimer's Disease based on Residual Bilinear and Attentive Models

Xue Lin, Yushui Geng, Jing Zhao, Wenfeng Jiang, Zhen Yan

Due to the characteristics of high noise and low resolution in medical images, it is difficult to extract local features, which affects the accuracy of image diagnosis and classification. To exploit the discriminative features of local image regions, we propose a network model method that combines improved residual bilinear and attention mechanism. First, in the ResNeXt model, it performs segmentation and convolution on the original residual unit structure to extract multi-scale features of the image. And it replaces the VGGNet model in bilinear. Then, it uses channel nonlinear attention to obtain expressive features when extracting features, and employs spatial attention for weight region selection to achieve BAP (Bilinear Attention Pooling) fusion. Finally, it implements classification in the SVM classifier and tests our model on the Alzheimer?€?s Disease Neuroimaging Initiative (ADNI) dataset. The results show that the model has better accuracy and robustness than other models in AD diagnosis classification.

Anslysing and Evaluating Complementarity of Multi-Modaility Data Fusion in AD diagnosis

Zhaodong Chen, Fengtao Nan, Yun Yang, Jiayu Wang, Po Yang

The clinical progression of Alzheimer?€?s disease( AD ) can?€?t be accurately evaluated by single modality data alone. Multi-modal data have a good effect on the diagnosis of AD. Clarifying the complementarity between modalities is crucial for the assessment of each stage of AD. Few studies have specifically explored the complementarity between different modalities due to the lack of completely aligned and paired multi-modal data and the limitation of sample size. However, collecting the full set of aligned and paired data is expensive or even impractical. In addition, the limited number of samples poses a great challenge to the robustness of the model. In this paper, different machine learning( ML ) methods were used to explore data complementarity between T1-weighted magnetic resonance imaging ( MRI ), cerebrospinal fluid ( CSF ), and fluorodeoxyglucose-positron emission tomography ( FDG-PET ) modalities. The different modal data of Alzheimer?€?s Neuroimaging Initiative ( ADNI ) and the self-extracted neuroimaging data were experimentally explored. Experiments show that there is obvious complementarity between MRI and CSF. By fusing MRI and CSF data, three binary classification tasks using multi-modal fusion data have achieved varying degrees of improvement. At the same time, we also explored the important features of multi-modal fusion data through SHapley Additive exPlanations ( SHAP ), and found that most important features are supported by relevant literature.

MetaSpeech: Speech Effects Switch Along with Environment for Metaverse

Xulong Zhang, Jianzong Wang, Ning Cheng, Jing Xiao

Metaverse expands the physical world to a new dimension, and the physical environment and Metaverse environment can be directly connected and entered. Voice is an indispensable communication medium in the real world and Metaverse. Fusion of the voice with environment effects is important for user immersion in Metaverse. In this paper, we proposed using the voice conversion based method for the conversion of target environment effect speech. The proposed method was named MetaSpeech, which introduces an environment effect module containing an effect extractor to extract the environment information and an effect encoder to encode the environment effect condition, in which gradient reversal layer was used for adversarial training to keep the speech content and speaker information while disentangling the environmental effects. From the experiment results on the public dataset of LJSpeech with four environment effects, the proposed model could complete the specific environment effect conversion and outperforms the baseline methods from the voice conversion task.

Potential Game Based Connectivity Preservation for UAV-Assisted Public Safety Rescue

Jingjing Wang, Yanjing Sun, Bowen Wang, Toshimitsu Ushio

In public safety networks (PSNs), it is an important issue how reliable data transmission recovers when some base stations (BSs) are damaged by natural disasters. An unmanned aerial vehicle (UAV) is used as a temporal relay station to transmit data of ground users (GUs) to an undamaged BS. In this paper, we consider a swarm of UAVs and introduce the following three roles for its management: (1) Relay UAVs (RUs) sacrifice their coverage capabilities to preserve the network connectivity; (2) Air BS UAVs (BUs) perform a covering task; (3) Standby UAVs (SUs) remain inactive. Then, we formulate an optimal coverage problem where we assign a role to each UAV to maximize the number of GUs that can transmit their data to the undamaged BS. First, we transform the problem into an exact potential game (EPG) whose utility function is designed based on the number of GUs served by each UAV. Next, we propose a learning algorithm to obtain an optimal role assignment and utilize the Fiedler eigenvalue, which represents the algebraic connectivity of the network topology of the swarm, to update the strategy selection probabilities. Finally, by simulation, it is shown that the proposed algorithm can strike a better balance between coverage and connectivity preservation than other benchmark algorithms.

Three-dimensional Key Distribution Scheme in Wireless Sensor Networks

Wanqing Wu, Ziyang Zhang, Yahua Dong, Caixia Ma

One of the major security challenges faced by wireless sensor networks(WSNs) is establishing a secure link for communication between neighboring sensor nodes. Finding a balance between connection, overhead, and resilience against node capture attacks is difficult due to the resource limits of sensor nodes. We propose a new three-dimensional key distribution scheme for wireless sensor networks based on polynomial and random key distribution schemes. The key pool is divided into two sections in the proposed scheme: key pool 1 is generated by the polynomial pool, and key pool 2 is generated by key pool 1. A three-dimensional key distribution model is constructed using the key pool and the coefficients of the polynomials. It can enhance network resilience while maintaining good connectivity by dynamically adjusting the degree of polynomials and the size of the polynomial pool. This paper analyzes the performance of the proposed scheme and compares it with other schemes. The results show that the proposed scheme has better local connectivity and resilience against node capture attacks when compared with the previous schemes.

Application identification under Multi-Service Integration Platform

Ziyang Wu, Yi Xie

Multi-service integration platform (MIP) is becoming a new way for mobile applications to provide services, such as the ChatBot of Facebook and the applet of WeChat. However, currently there are no special means and filtering strategies to supervise the services running on various MIPs. Existing solutions for program detection and traffic analysis are not suitable for MIP scenarios, which creates favorable conditions for the dissemination of illegal content through MIP . To address this issue, in this work we propose a new approach to identify mobile applications running on MIP platforms. The proposed approach uses IP flow to reconstruct data units of both transport and application layers respectively. By this way, we can capture the data transmission behavior of multi-protocol layers and obtain richer semantic features for application identification. Then, multikernel convolutional neural networks (CNNs) and long short term memory (LSTM) neural networks are employed to extract and aggregate the multi-scale features from the perspective of both protocol layer and time series. Finally, the fused features generated by the models are used to identify the category of the pending applications by a classifier composed of a fully connected neural network. We validate the proposed approach by three real datasets. The experimental results show that the proposed approach outperforms most existing benchmark methods in performance.

UAV Visual Navigation System based on Digital Twin

Jingsi Miao, Ping Zhang

In recent years, many scholars have carried out researchs on UA V digital twin from various aspects. However, the research is still in the preliminary stage, and there are still some problems, such as incomplete data and model fusion, poor migration of algorithm policy, poor relation between virtual and physical space, and lack of extensibility of application scenarios. In order to explore the application potential of digital twin technology in UA V fields, this paper introduces digital twin into UA V monocular visual navigation. Therefore, this paper proposes a digital twin(DT)-based framework integrating with deep neural network, which consists of physical space, virtual space, twin data layer and application layer. Next, the multi-modal decision model with decoupling methods in application layer consisting of perception model and control model is built to explore the global optimal solution and control the behaviors of UA V . Finally, the digital twin system and decision model are verified in virtual space and physical space respectively. The results shows that the UA V visual navigation system based on digital twin reduces the cost of application, algorithm development and deployment, and improves the migration ability of navigation policy. Compared with the baselines, the proposed decision model has the best navigation performance in both virtual space and physical space. Compared with the navigation policy without the decoupling method, the performance index is improved by about 8.6% in virtual space and 2.7 times in physical space.

Applications of Reinforcement Learning in Virtual Network Function Placement: A Survey

Cong Zhou, Baokang Zhao, Jing Tao, Baosheng Wang

In recent years, network function virtualization has attracted massive attention in academia and industry,and the virtual network functions placement problem is one of them. Reinforcement learning has been widely applied in network control and decision, which can learn the optimal policy according to the environment feedback automatically. This paper presents a new summary of the virtual network functions placement problem based on reinforcement learning. We will give a detailed description of how to use reinforcement learning to solve virtual network function placement in different scenarios, then the prospect of further research is forecasted preliminarily.

Session Chair

Xuan Liu, Hunan University, China

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