Effective management of private keys is crucial to ensure the security and ownership of users' data and digital assets in the Web3 environment. However, existing solutions often fail to adequately address private key management from the user's perspective. Private key leakage and loss incidents occur frequently, resulting in significant losses of digital assets. Moreover, the conventional approach of revoking both the private and public keys after a leakage or loss accident is inconvenient in Web3, where the public key serves as the user's wallet address or digital identity. To tackle the issue of user-side private key management in Web3, this paper presents KeyShield which is a leakage-and-loss-resilient private key protection scheme. KeyShield divides the user's private key into three shares, securely stored across a primary device and a secondary device owned by the user, and a third storage module owned by the user or a semi-trusted service provider. For daily use of the private key, the user only needs to connect the primary and secondary devices. In the event of a leakage or loss, such as device theft or attack, an update process will be triggered to update the three shares, immediately invalidating the leaked or lost share while causing no changes to the public key. As a demonstration of KeyShield, we developed KeyShieldECC accessible on both Android and iOS platforms for managing Elliptic Curve Cryptography (ECC) private keys. The testing results show that for a 256-bit ECC private key, the daily use only needs 0.05 seconds and update needs 0.25 to 0.3 seconds on an ordinary smart phone.

Non-Fungible Token (NFT) creators use digital signatures to ensure the ownership, authenticity, integrity, and nonrepudiation of their digital works. However, if the private key is compromised, an attacker can generate unauthorized NFTs by using the creator’s private key to issue valid signatures. These valid but unauthorized signatures will be accepted in the NFT market and cannot be revoked. Even if the NFT creators update their private-public key pairs, they cannot deny the NFTs generated by the attacker. To mitigate these risks, we propose revocable signature by introducing commitment mechanism and an Auxiliary Embedded Key (AEK) into the signature, while the regular verification process does not involve this AEK. If a valid but unauthorized signature is detected and needs to be revoked, AEK will be disclosed to perform the revocation operation. To illustrate the application of revocable signatures in NFT, we design and implement a revocable Elliptic Curve Digital Signature Algorithm (ECDSA) scheme with provable security. Experimental evaluations on the FIPS-recommended elliptic curves show that the performance of revocable ECDSA is comparable to the basic ECDSA, with additional 0.0303 s (P-256 curve) and 0.15 USD gas fee in Remix VM for revoking a signature.

Understanding the interactions between human-object (HO) pairs is the key to the human-object interaction (HOI) detection task. Visual understanding research has been significantly impacted by recent advances in linguistic-visual contrastive learning. For HOI detection studies, the alignment of linguistic and visual features is usually required to be performed when linguistic knowledge is used for enhancement. This usually results in the demands of extra training data or extended training time. In this study, an effective approach for utilizing multimodal knowledge to enhance HOI learning from global and instance scales is proposed. Model performance on Rare HOI categories can be prominently improved by using projection guided by linguistic knowledge at a global scale and merging multimodal features at an instance scale. State-of-the-art performance on the HICO-Det benchmark is achieved by the proposed model, and the effectiveness of the proposed global-and local-scale multimodal learning approach is validated.

Human-object-interaction (HOI) detection is a task that requires the model to detect human-object pairs and recognize the interactions between them. Recent advances in Detection Transformers have had a great impact on the HOI detection research field. However, Transformer-based HOI detection methods only exploit randomized queries for decoding HOI triplets. For the HOI detection model using visual and other modalities of features as explicit queries, the computation complexity of the model can be significantly increased compared to randomized implicit queries. In this paper, a simplified approach for constructing explicit queries with only visual features of instances combined with a set prediction training strategy, used in Transformer-based HOI detection models, is proposed. This paper also proposes a novel method to extract and reduce the dimension of instance visual features, which can benefit the learning process of HOIs. The model proposed, in this paper, achieves state-of-the-art performance on the HICO-Det benchmark and shows a competitive training efficiency compared to other HOI detection methods.

This project aims to address the limitations in existing human-robot interaction systems by developing a real-time facial expression recognition and emotion response system using a Raspberry Pi-based humanoid robot. Unlike traditional systems that rely solely on explicit commands, our system enables the robot to autonomously respond to human emotions, demonstrating a level of emotional intelligence. This approach enhances the naturalness and human-like- ness of robot interactions. Our system processes real-time video feeds captured by a camera to identify user expressions, focusing specifically on sadness and happiness. Upon detecting a sad expression, the robot performs a song and dance routine to uplift the user’s mood and ceases the performance once a happy expression is recognized. Experimental results indicate that the system effectively recognizes emotions and improves users’ emotional states, showcasing its potential applications in therapeutic and interactive environments. The integration of deep learning with robotic control offers a novel approach to enhancing human-robot interaction through emotional intelligence.

Data trading has become a fundamental category of commerce in the current digital era, bringing along substantial opportunities for economic activities. The emergence of blockchain facilitates decentralized data trading, empowering users to maintain complete control over their own data and enabling direct peer-to-peer transactions. However, while users possess complete control over their data product and asset in a decentralized trading environment, they must take responsibility for their behaviors and face the consequences of dishonest behaviors either from themselves or counterparties. Therefore, behavior fairness problems arise, such as the seller refusing to deliver the correct data product after receiving the money and the buyer failing to pay the outstanding balance after receiving the data product. This article addresses these issues by proposing a decentralized fair data trading ecosystem leveraging the capability of cryptography and smart contracts. A novel verifiable proxy re-encryption (VPRE) scheme is designed, which introduces the verifiability of re-encryption keys into the original PRE schemes. The scheme is implemented using smart contracts, which ensures that trading can succeed if and only if the data seller provides a valid re-encryption key and the buyer pays the correct amount of money. Furthermore, experiments are conducted to evaluate the proposed ecosystem in terms of trading fairness, security, and cost. The results show that the proposed scheme can effectively terminate transactions involving fairness violations with affordable costs ranging from 39 276 Gwei to 296 983 Gwei. Without our solution, these unfair transactions will proceed and consume 77 301 Gwei to 754 447 Gwei gas fees.

In the Internet of Vehicles (IoV), federated learning (FL) has been regarded as a promising solution for training distributed models while preserving data privacy, which is a critical requirement for trusted data circulation (TDC). However, due to the limited resources and the dynamic network structure of IoV, improving the data communication efficiency in FL training is still a huge challenge. Compressive sensing (CS) reduces communication overhead by enabling efficient data compression and reconstruction. In this paper, we take CS-enabled FL one step further by introducing an adaptive scheme to enhance communication performance. Specifically, we first select vehicles participating in FL based on moving conditions to maximize system stability and model convergence. Second, inspired by synaptic intelligence (SI) in the field of continuous learning (CL), we prioritize gradient information based on its importance in model performance to optimize the CS compression and reconstruction process. Experimental results show that the proposed scheme can achieve the same accuracy as the classical compression method with less or even less than half of the data, which indicates that it is a practical FL scheme for IoV.

Intellectual Property (IP) is a special type of data that has broad and high trading demands. Existing blockchain-based IP data trading schemes can promote the IP data trading market by removing the dependence on centralized platforms. However, the problem of trading fairness among sellers and buyers is more challenging compared to centralized approaches. This paper addresses the trading fairness problem by representing the data as Non-Fungible Tokens (NFTs) and separating usage rights and ownership. An NFTized IP data trading system is designed and a two-stage fair trading scheme is proposed. They ensure that buyers need not pay additional money if the IP content is not useful for them, and the sellers will not lose the IP ownership until they receive additional money in the second stage of trading. A prototype for the system is realized, and based on it, a set of experiments are carried out to evaluate the performance. The experimental results show the cost is acceptable.

Human-Object Interaction (HOI) detection is a critical topic in the visual understanding field. With the development of deep learning models, the research of HOI detection has been profoundly reshaped. Deep convolutional neural networks increased the object recognition accuracy of static images and induced a detection-based HOI detection stream. The detection-based models resolve the HOI detection problem from a classification perspective. Another stream of HOI detection methods seeks a deeper understanding of the information shown in images, and they are named HOI understanding methods in this survey paper. HOI understanding methods usually acquire external linguistic data to enable the deep models to learn more about the images. Additionally, some of the HOI understanding methods exploit graph neural networks (GNN) to increase the inference accuracy of the model.

Text-to-GraphQL (Text2GraphQL) is a task that converts the user's questions into Graph + QL (Query Language) when a graph database is given. That is a task of semantic parsing that transforms natural language problems into logical expressions, which will bring more efficient direct communication between humans and machines. The existing related work mainly focuses on Text-to-SQL tasks, and there is no available semantic parsing method and data set for the graph database. In order to fill the gaps in this field to serve the medical Human–Robot Interactions (HRI) better, we propose this task and a pipeline solution for the Text2GraphQL task. This solution uses the Adapter pre-trained by “the linking of GraphQL schemas and the corresponding utterances” as an external knowledge introduction plug-in. By inserting the Adapter into the language model, the mapping between logical language and natural language can be introduced faster and more directly to better realize the end-to-end human–machine language translation task. In the study, the proposed Text2GraphQL task model is mainly constructed based on an improved pipeline composed of a Language Model, Pre-trained Adapter plug-in, and Pointer Network. This enables the model to copy objects' tokens from utterances, generate corresponding GraphQL statements for graph database retrieval, and builds an adjustment mechanism to improve the final output. And the experiments have proved that our proposed method has certain competitiveness on the counterpart datasets (Spider, ATIS, GeoQuery, and 39.net) converted from the Text2SQL task, and the proposed method is also practical in medical scenarios.

Diabetes is a disease that actually impacts the capacity of the body to obtain blood glucose, which is usually referred to as blood sugar. At the end of 2019, a new public health problem (COVID-19) emerged. This disease has greatly harmed people with diabetes. Therefore, we intend to make use of data mining algorithms to prevent death and improve the quality of life through the prediction of diabetes. In this paper, four different algorithms have been used to analyze Diabetes from DAT260x Lab01: Logistic, Decision Tree Classifier, Xgboost and SVC. The models are evaluated for which algorithm is much effective. The paper then provides a quick overview of both the set of data and the fieldwork carried out on the subject. In the adjoining step, the dataset and its features are discussed. In addition, the paper explains the four algorithms and virtual environments that have been used to clarify the variables, which have the largest impact on raw data. The findings are obtained by evaluating the confusion matrix applied to the whole selected algorithm. The paper outlines the full observations and conclusions taken based on the results.

In the medical domain, data are often collected over time, evolving from simple to refined categories. The data and the underlying structures of the medical data as to how they have grown to today's complexity can be decomposed into crude forms when data collection starts. For instance, the cancer dataset is labeled either benign or malignant at its simplest or perhaps the earliest form. As medical knowledge advances and/or more data become available, the dataset progresses from binary class to multi-class, having more labels of sub-categories of the disease added. In machine learning, inducing a multi-class model requires more computational power. Model optimization is enforced over the multi-class models for the highest possible accuracy, which of course, is necessary for life-and-death decision making. This model optimization task consumes an extremely long model training time. In this paper, a novel strategy called Group-of-Single-Class prediction (GOSC) coupled with majority voting and model transfer is proposed for achieving maximum accuracy by using only a fraction of the model training time. The main advantage is the ability to achieve an optimized multi-class classification model that has the highest possible accuracy near to the absolute maximum, while the training time could be saved by up to 70%. Experiments on machine learning over liver dataset classification and deep learning over COVID19 lung CT images were tested. Preliminary results suggest the feasibility of this new approach.

This paper describes how to develop diabetes diagnosis through the combined use of the support vector machine, the Decision Tree, Naive Bayes, K-nearest and finally, Random Forest (RF) algorithms. These methods are useful to predict diabetes jointly. The appropriateness of ML-depended techniques to tackle this issue has been revealed. This diabetes diagnosis system using machine-learning algorithms is used to review papers. This project was based on developing python-based code for machine learning algorithms to perform large scales of diabetes analysis. The hardware requirement of machine learning is RAM that is 128 GB DDR4 2133 MHz and 2 TB Hard Disk and needs 512 GB SSD. One standard library is NumPy that uses to support multi-dimensional arrays objects, various components, and matrices. The Random Forest Prediction representing the pictorial visualization of the model and the accuracy for the data analysis using the Random Forest is 76%.

Although bearing fault diagnosis methods based on deep learning are very popular in recent years and a lot of brilliant results have been achieved, they assume that the distribution of training samples is same with test samples. However, the working condition of bearing is variable, and labeling fault tags for all data is time-consuming and laborious. In order to solve the problem of lacking labeled data in cross domain scenario, a novel domain adaptation transfer learning based fault diagnosis method based on deep domain adversarial network is proposed. In this method, a deep convolutional neural network (CNN) is used to extract features from raw vibration signals. Then a discriminator and a classifier are applied to minimize the distribution difference of cross-domain features. Experiments are carried out on three benchmark datasets, and the results show that the accuracy of proposed methods is higher than other existing unsupervised transfer learning methods.

We propose a single-document abstractive summarization system that integrates token relation into a traditional RNN-based encoder-decoder architecture. We employ pointer-wise mutual information to represent the token relation and adopt Graph Convolutional Networks (GCN) to extract token representation from the relation graph. In our experiment on Gigaword, we consider importing two kinds of structural information: token (node) representation from the relation graph. Also, we implement two kinds of GCNs, a spectral-based one and a spatial-based one, to extract structural information. The result shows that the spatial based GCN-enhanced model with node representation outperforms the classical RNN-based encoder-decoder model.

This paper presents a rapid response system architecture for the distributed management of warehouses in logistics by applying the concept of edge computing. A tiered edge node architecture is proposed for the system to process computing tasks of different complexity, and a corresponding rapid response algorithm is introduced. A software-defined simulation is done to evaluate the system performances on rapidness and correctness, from which it can be concluded that all pre-defined emergency cases can be detected and responded to within a relatively short period of time.

Cloud computing is a new commercial model that enables customers to acquire large amounts of virtual resources on demand. Resources including hardware and software can be delivered as services and measured by specific usage of storage, processing, bandwidth, etc. In Cloud computing, task scheduling is a process of mapping cloud tasks to Virtual Machines (VMs). When binding the tasks to VMs, the scheduling strategy has an important influence on the efficiency of datacenter and related energy consumption. Although many traditional scheduling algorithms have been applied in various platforms, they may not work efficiently due to the large number of user requests, the variety of computation resources and complexity of Cloud environment. In this paper, we tackle the task scheduling problem which aims to minimize makespan by Genetic Algorithm (GA). We propose an incremental GA which has adaptive probabilities of crossover and mutation. The mutation and crossover rates change according to generations and also vary between individuals. Large numbers of tasks are randomly generated to simulate various scales of task scheduling problem in Cloud environment. Based on the instance types of Amazon EC2, we implemented virtual machines with different computing capacity on CloudSim. We compared the performance of the adaptive incremental GA with that of Standard GA, Min-Min, Max-Min , Simulated Annealing and Artificial Bee Colony Algorithm in finding the optimal scheme. Experimental results show that the proposed algorithm can achieve feasible solutions which have acceptable makespan with less computation time.

With the continuous development of automation and information technology, large amounts of safety data are produced in the processes of coal production. Most enterprises simply focus on statistics and do not conduct systematic big data analyses. Therefore, it is necessary to study the theory of coal mine safety while using big data systematically. This paper expounds on the changes in coal mine safety that have been driven by big data from three aspects: the connotation, characteristics and research framework. First, the connotation of coal mine safety big data (CMSBD) is redefined by changing the safety entities and methods. Second, the advantages and disadvantages of the big data model are compared from the perspective of feature analysis. Finally, the research paradigm and technical framework of CMSBD are designed. The results show that the management connotation of CMSBD focuses on the role of big data in coal mine safety. Compared with coal mine safety small data (CMSSD), CMSBD has both advantages and disadvantages. Therefore, CMSBD must be combined with a small data method. The research paradigm emphasizes the intersection of the research, the relevance of safety thinking, the importance of safety data analysis, and the fusion of big data with traditional small data models.

Distributed sensing in combination with wireless communication techniques and self-organising deployment approaches enable Wireless Sensors Networks (WSN) to play a crucial role in our daily life. An increasing number of dynamic sensing applications have been developed, and there is a growing demand for middleware to bridge the gap between these applications and a diverse range of underlying operating systems. In this paper, we present a prototype middleware that combines a Mobile Agent-based WSN system in conjunction with the component based LooCI middleware. The system has both merits of agent-based WSN system as well as the benefits of LooCI, which include efficient energy use and providing the possibility of platform independence and component model programming language independence. The feasibility analysis along with potential implementation approaches of this system are proposed in this paper. The direction and value of on-going research based on the research work proposed in this paper is presented as well. Index terms -Mobile Agent, LooCI, WSN, Middleware