Applied and Computational Engineering
- The Open Access Proceedings Series for Conferences
Proceedings of the 4th International Conference on Computing and Data Science (CONF-CDS 2022)
2022-07-16
The goal of this conference is to bring together researchers and practitioners from academia and industry to highlight the importance of computing technologies and data science as well as establish new collaborations in these areas. The conference looks for significant contributions to computing, data mining, and data science in theoretical and practical aspects.
978-1-915371-19-5 (Print)
978-1-915371-20-1 (Online)
2023-03-22
Alan Wang, University of Auckland
With Asian American population on the rise, Asian Americans have more power than ever in political elections. While qualitative methods have been used to measure this uptrend in Asian American political participation, a thorough and cohesive quantitative study on the topic has yet to be done. Therefore, we utilize machine learning and dictionary-based matching to identify the most accurate numbers for Asian American political participation, categorized by political affiliations, geography, gender, etc. We find that Asian American political participation tends to the Democratic party, is predominantly male, and is proportional to the population size of the individual’s respective state.
The Iterative Closest Point (ICP) algorithm is one of the most important algorithms for ge-ometric alignment of three-dimensional surface registration, which is frequently used in computer vision tasks, including the Simultaneous Localization And Mapping (SLAM) tasks. In this paper, we illustrate the theoretical principles of the ICP algorithm, how it can be used in surface registration tasks, and the traditional taxonomy of the variants of the ICP algorithm. As SLAM is becoming a popular topic, we also introduce a SLAM-oriented taxonomy of the ICP algorithm, based on the characteristics of each type of SLAM task, including whether the SLAM task is online or not and whether the landmarks are present as features in the SLAM task. We make a synthesis of each type of SLAM task by compar-ing several up-to-date research papers and analyzing their implementation details.
Due to the low latency requirements in object detection, numbers of one-stage methods like YOLO and SSD adopt a shared head for both classification and localisation tasks. While the decoupled head used to decouple the subtasks into different heads are getting more popular in one-stage detection because they improve accuracy. In contrast, the computational complexity caused by the decoupled head can’t be ignored. To solve these problems, we propose an integrated knowledge distillation framework for transferring the representation ability of the decoupled head to the original coupled head and contributing to efficient one-stage object detection. It solves the problem that the coupled head is insufficient in handling the conflict of subtasks and avoids the time delay introduced by the coupling head and the increase of network parameters.
Style-based image manipulation is to fuse the types of two arbitrary images, which is a popular task in computer vision. StyleGAN is a sophisticated architecture for generating images of high qualities. The framework allows the generator to operate on a latent space that is disentangled and allows us to do scale-specific manipulation on the semantic information of the generated images. In this paper, the author managed to fuse the styles of two given images on a controllable degree. The resultant images have natural appearances approximating real human portraits. Our method provides qualitative results for style-fusion of two given images, which achieves satisfy. Since StyleGAN offers an unraveled latent space representing disentangled semantics, the author hopes to use it on tasks like GAN inversion and manipulate images in a fine-grained control, which is the future work.
With the rapid development of machine learning and the automotive industry, the industry of autonomous driving continues to grow. At the same time, governments have new regulations on autonomous driving, which tells us that reliable systems have become more and more critical while developing autonomous driving. In this paper, I use three different classifiers, which are Logistic Regression (LR), Random Forest (RF), and neural network (Multilayer Perceptron), to do the traffic sign recognition tasks and set the best parameters for every classifier. I train three classifiers with the best parameters and estimate using cross-value methods. Finally, I compared the performance, which indicates Random Forest Classifier has the best result among the three classifiers.
With the advent of intelligent era, the application of artificial intelligence has penetrated all walks of life in our daily lives and has gradually entered the scope of being understood by the public, such as AlphaGo, which is a human-machine game in Go. Meanwhile, with the increasing complexity of deep neural network algorithms, how to compress and optimize neural network models becomes the key to reducing model storage space and improving model deployment efficiency. This article will complete the INT8 quantization of the model for AlexNet and use the concept of batch normalization (BN) layer to prune the Yolov3 model. Finally, it is proved by experiments that the model quantization makes the image recognition time of the AlexNet about 1/3-1/4 of the original format; and pruning of Yolov3 will significantly reduce the storage space occupied by the model, moreover, make the speed doubled of embedded GPU real-time object detection. However, none of the three experiments reduces the model recognition accuracy. This paper proposes some essential concepts in the main body and applies them to model algorithms and experiments to provide theoretical support and proof for the field of deep learning model compression and optimization. Furthermore, it is possible to explore cutting-edge research directions like Sparse Convolution Net based on the research methods and experimental conclusions of the paper.
People nowadays use internet platforms to exchange ideas, share opinions, and learn online. Huge amounts of data are being poured into social media in the form of tweets, blogs, and updates on articles and items, among other things. The data is all unorganized and unprocessed. It is necessary to arrange and examine it. It takes a long time to analyze and process information using traditional methods and is impossible to analyze each and every sentence. So, there is a need to have a better approach. It can be done through sentimental analysis which extracts the opinion of a user in a piece of text data. This sentimental analysis will predict the polarity of the sentence, whether the given sentence is positive or a negative one. The sentimental analysis can be achieved through three approaches namely lexicon based, machine learning based and hybrid approach. This sentimental analysis is a part of NLP. This project aims to perform sentimental analysis using machine learning techniques and few natural language processing techniques on a product reviews dataset.
Image generation has become a heated research topic in recent years owing to its wide landing scenes and great potential in all walks of life. Especially after the emergence of adversarial neural networks, both the training process and results have been greatly improved compared with previous model methods. This paper focuses on the advantages of directly using Generative Adversarial Nets (GAN) to generate images, as well as its main problems: training instability, pattern collapse, and global correlation, and introduces the strategies and skills of subsequent improved GAN for these problems. Through experiments, we compare the improved network with the original GAN and try to combine the core strategies of these networks. In the experiment, the image quality generated by the combined network is higher.
Electronic medical records have been rolled out in the past decades to facilitate the medical exports’ daily routine. However, the number of electronic medical records increases dramatically, which also causes huge workloads for the front-line clinical workers when they face the writing-up work. In this sense, researchers in the artificial intelligence domain wish to automate this process by constructing a natural language processing system, and medical information extraction is one of the key steps amongst the entire work. In this paper, we focus on medical information extraction from doctor-patient dialogues, and propose a novel encoder-decoder model which incorporates global information into the dialogue windows. The experiment on the MIE dataset suggests our model outperforms the compared baseline models, and achieves the state-of-the-art results, which proves our model’s effectiveness.
Osteoporosis is a medical condition that affects the structure and strength of bones. Osteoporosis is an asymptomatic disease of the bone that affects a significant proportion of the world's elderly, leading to increased fragility of the bone and an increased risk of fracture. This paper's key objective is to provide a critical review of the main artificial intelligence-based systems for detecting populations at risk of osteoporosis or fractures. Skeletal deformities, fractures, twisted knees, inherited bone defects, and other bone disorders affect millions of individuals as a result of a variety of bone disorders. These may help to prevent a variety of possible complications if diagnosed and treated early. We discussed deep neural networks in this paper, including recognition, segmentation, and classification. The architecture and concepts of the deep learning algorithm we used to detect bone density were also discussed. As a result, we’ll use a variety of deep learning algorithms to build a model that can detect a person's bone mass density and recognize any potential threats that have occurred or could occur.
With the advancement in technology, data generated by non-stationary in day-to-day life is massive, continuous and rapid. Many applications such as IoT, transaction systems, network sensors, video surveillance systems, and network intrusion detection systems generate a massive amount of real-time data. The data used in traditional data mining is static in nature, and it can be revised for processing and Analysis. While data in data stream mining is dynamic in nature and it never stops. Besides, the data generated may have a change imbibed in its characteristics over a long/short period of time which is called concept drift. So, analysing such data has huge inbuilt challenges that deal with the dynamism of the characteristics of data itself. This dynamic nature is because of fast and continuous changing data and its enormity. To overcome this limitation, we can use modified clustering techniques that could help us in proper data analysis. Clustering is an effective method used in data mining; but clustering data streams may add some additional challenges such as storage capacity, limited time, one pass and rate of arrival. Furthermore, data streams are fickle in nature and because of this behaviour it needs to be processed as and when it arrives. In addition to that, knowledge about the number of clusters like in K-means clustering is unknown. In view of these characteristics of the data stream, the information or the data generated in the data stream are non-deterministic. Such non-deterministic information contains noise points or outliers, so developing an effective clustering algorithm in a data stream is a crucial task. These methods can work with labelled data in data stream clustering, which has the potential to identify clusters of any shape and noise. The motivation for this research work is using the said algorithm to address and overcome the constraints of the data stream and to dig out the best knowledge from it.
The usage of Information Technologies (IT), as well as data has increased over the years to improve businesses and human life generally thereby increasing productivity. In the health sector, Electronic Health Records (EHR) has helped in collecting demographic medical data which helps healthcare practitioners to provide quality health care. The EHR generates lots of medical data and are analyzed and processed to form big health data and sent to Agencies, Government to influence decisions in improving quality of life and to predict or prevent premature deaths and disease development. In this research work, we propose a model using lattice-based cryptography to encrypt health big data and deploy a decoy model which will be coined in fog computing facility to serve as honeypot or trap machine to attract attackers.
Cyber security is one of the most difficult and fast-growing concerns today's enterprises are focusing on. The practice of reducing potentially damaging and unknown events that pose a danger to cyber security is known as cyber security risk management. The Game Theoretic Approach is a popular cyber security risk or threat management strategy (GTA). This study provides a paradigm for cybersecurity risk or threat handling based on a game-theoretic approach to Fog computing, which will encourage proactive cyber risk management and improve cyber-operational efficiency/effectiveness. The method is written in such a way that the PyQt4 framework acts as a shield for the Fog server, performing inline packet inspection and logging any malicious packets to the console and a database on the server using Snort. The study proposes a Bayesian game model for risk management in the cyber domain.
Strong gravitational lensing is a powerful tool for probing the mass content of the Universe. Both high-resolution imaging and spectroscopy observations are important for identifying and measuring the properties of the lensing object. The strong lensing system J1436+4943 was discovered in the SDSS-IV MaNGA survey and further observed with the FOCAS IFU spectrograph on the Subaru Telescope. We investigate whether comparable properties, e.g., the Einstein radius can be obtained using only MaNGA data cubes. The result shows that the general properties of the lens system can be recovered but the Einstein radius is different from the result with FOCAS IFU data by less than 25%. The MaNGA data cubes are helpful for fast analysis of large samples and to study the statistical properties of gravitational lensing systems.
With the continuous development of stochastic gradient descent algorithms, many efficient momentum algorithms have appeared. Stochastic gradient descent(SGD) is one of the classic algorithms in optimization. Its accelerated version, the SGD algorithm with momentum strategy, has been a hot research topic in recent years. Therefore, this paper will analyze and summarize these series of algorithms, starting with the classical momentum algorithm, and introduce some improved versions of the momentum algorithm. Numerical experiments on real problems will also be done to evaluate the performance of these algorithms. It is proved that the addition of momentum and adaptive learning rate effectively improve the performance of these algorithms. In future research, some cutting-edge momentum algorithms and other basic network should be analyzed.
The existing method does not seem to be a killer application that combines image segmentation and inpainting to do image processing tasks for ordinary people. Therefore, we propose a region-based inpainting method, namely Remover. Remover is a method that can analyze the content of images, perform automatic image segmentation tasks with or without manual intervention, and inpainting the segmented part to achieve unwanted objects appearing to have been removed from an image without affecting the content of the image. With the help of open-source code and technical support. Remover stands as an Ubuntu desktop application. The code is under development and will be available at https://github.com/WPCJATH/remover soon.
Automatic feature extraction and processing of greater data is now possible because of advances in Deep Learning. To pre-train from a wider corpus and comprehend the language feature for sentiment classification work, transformers Generalized Autoregressive Pre-training for Language Understanding and Bidirectional Encoder Representations from Transformers (BERT) have been proposed. These language models learn the context in both ways. In the proposed work, we have examined and tested our text dataset of skin cancer cases using the BERTbase model. When determining whether a patient's symptoms are compatible with cancer or not the model has a 97.3 percent accuracy rate.
Memes have become a new type of internet communication. It has the ability to instantly disseminate anger, offensiveness, and violence. Because of its regional meaning, classifying memes is difficult.. This work presents here a computational model of classifying Tamil memes using convolutional neural networks. Convolutional neural networks have the potential to learn, adapt, and rearrange themselves. As a result, it can extract features automatically applying prior knowledge of existing categories, avoiding the time-consuming feature extraction process used in older methods in images. The basic layer of MobileNet is made up of depth-wise separable filters, also referred to as depth-wise separable convolution. The network structure is another feature that boosts performance. It utilizes very less than computation power while applying transfer learning. This network has reduced parameters and computation cost. Skip connections, or shortcuts, are used by residual neural networks to jump past some layers. Residual connections allow parameter gradients to travel more easily from the output layer to the network's prior layers, allowing for the training of deeper networks. Higher accuracies on more demanding tasks may arise from the greater network depth. AlexNet is a leading architecture for any image identification task, and it could have a lot of applications in the artificial intelligence field of computer vision. In the future, AlexNet may be used for image classification jobs more than CNNs. This work aims to classify Tamil memes using Mobilenet, Resnet and hyper parameter turned AlexNet.
Nowadays, more and more people suffer from heart disease because of stressful life, irregular diet, lack of exercise and other reasons. The population affected by heart disease is also younger than ever. If heart disease can be diagnosed as early as possible, it will be of great help in the treatment of heart disease. Thus, this paper proposes models to predict heart disease based on wide and deep neural network, and the result shows that the current work has maintain good performance. Analysis is also provided in this paper to state factors that can affect performance.
Sometimes people are not supposed to be in a photo for various purposes, but this is usually unavoidable. Therefore, in the postprocessing of the image, it can be solved by removing people from the picture without affecting the coherence and naturalness of the object and background in the photo. We propose a human removal method based on image instance segmentation and image inpainting. Firstly, we send an image into the image instance segmentation algorithm to obtain a mask covering the unwanted parts of the picture. Then we do dilatation on this mask to expand the mask region. Finally, the inpainting algorithm will take the image and the processed mask and produce an inpainted image with no human and feel natural.