Дисертації з теми "Online deep learning"
Щоб переглянути інші типи публікацій з цієї теми, перейдіть за посиланням: Online deep learning.
Оформте джерело за APA, MLA, Chicago, Harvard та іншими стилями
Ознайомтеся з топ-40 дисертацій для дослідження на тему "Online deep learning".
Біля кожної праці в переліку літератури доступна кнопка «Додати до бібліографії». Скористайтеся нею – і ми автоматично оформимо бібліографічне посилання на обрану працю в потрібному вам стилі цитування: APA, MLA, «Гарвард», «Чикаго», «Ванкувер» тощо.
Також ви можете завантажити повний текст наукової публікації у форматі «.pdf» та прочитати онлайн анотацію до роботи, якщо відповідні параметри наявні в метаданих.
Переглядайте дисертації для різних дисциплін та оформлюйте правильно вашу бібліографію.
1
Fourie, Aidan. ""Online Platform for Deep Learning Education"." Master's thesis, Faculty of Commerce, 2019. http://hdl.handle.net/11427/31381.
Анотація:
My thesis is going to focus on the development of a standalone, web based, machine learning educational platform. This platform will have a specific focus on neural networks. This tool will have the primary intention to provide a theoretical background to the mathematics of neural networks and thereafter to allow users to train their own networks on regression problems of their own creation. This is so as to provide the user with both theoretical, and first-hand, experience in the applications and functions of artificial intelligence. The primary success metric of this project will be how informative it is to the user. The key deliverable will be a fully functional prototype in additional to a written piece inclusive of a literature review and any other relevant findings and conclusions.
2
Guo, Song. "Online Multiple Object Tracking with Cross-Task Synergy." Thesis, The University of Sydney, 2021. https://hdl.handle.net/2123/24681.
Анотація:
Multiple object tracking (MOT) plays an important role in computer vision and has been studied for years. It aims to identify and keep track of objects appear in a given video sequence, which is useful for applications like autonomous driving. Modern online MOT methods usually adopt tracking-by-detection paradigm, where detections of objects are performed first, followed by data association between the found detections to establish identities of tracked targets. Current main challenge of MOT is from the occlusions incurred by interactions among moving objects, where detections may fail to be produced and lead to gaps in tracking. The two stages in the tracking-by-detection naturally shows two possible directions to tackle the problem, and most methods usually focus on one of them, either to improve tracking performance through conducting prediction on targets' new locations in a new frame, or to enhance robustness of embeddings used for similarity metrics in data association. However, occlusion is a shared problem in both of the two tasks, thus improving them individually is less optimal than linking them together to deal with the issue.
To address the problem, a novel unified model is proposed to bring synergy between the two tasks of position prediction and embedding association and its remarkable ability in dealing with occlusions is shown. The two tasks are linked by temporal-aware target attention and distractor attention as well as discriminative embedding aggregation. With attentions generated from identity-specific aggregated embeddings, position prediction focuses more on targets and less on distractors, therefore make more correct predictions. This enables refined embedding extraction under occlusion with the help of attention modules, to produce enriched embeddings for association and aggregation. In this way, the model involves the two tasks with each other to create synergy in between and shows stronger robustness to occlusions. The method is evaluated on MOTChallenge benchmarks to show its state-of-the-art performance. Extensive analyses are conducted to verify the effectiveness of each designed component.
3
Pham, Cuong X. "Advanced techniques for data stream analysis and applications." Thesis, Griffith University, 2023. http://hdl.handle.net/10072/421691.
Анотація:
Deep learning (DL) is one of the most advanced AI techniques that has gained much attention in the last decade and has also been applied in many successful applications such as market stock prediction, object detection, and face recognition. The rapid advances in computational techniques like Graphic Processing Units (GPU) and Tensor Processing Units (TPU) have made it possible to train large deep learning models to obtain high accuracy surpassing human ability in some tasks, e.g., LipNet [9] achieves 93% of accuracy compared with 52% of human to recognize the word from speaker lips movement. Most of the current deep learning research work has focused on designing a deep architecture working in a static environment where the whole training set is known in advance. However, in many real-world applications like predicting financial markets, autonomous cars,
and sensor networks, the data often comes in the form of streams with massive volume, and high velocity has affected the scalability of deep learning models. Learning from such data is called continual, incremental, or online learning.
When learning a deep model in dynamic environments where the data come from streams, the modern deep learning models usually suffer the so-called catastrophic forgetting problem, one of the most challenging issues that have not been solved yet. Catastrophic forgetting occurs when a model learns new knowledge, i.e., new objects or classes, but its performance in the previously learned classes reduces significantly. The cause of catastrophic forgetting in the deep learning model has been identified and is related to the weight-sharing property. In detail, the model updating the corresponding weights to capture
knowledge of the new tasks may push the learned weights of the past tasks away and cause the model performance to degrade. According to the stability-plasticity dilemma [17], if the model weights are too stable, it will not be able to acquire new knowledge,
while a model with high plasticity can have large weight changes leading to significant forgetting of the previously learned patterns. Many approaches have been proposed to tackle this issue, like imposing constraints on weights (regularizations) or rehearsal from experience, but significant research gap still exists.
First, current regularization methods often do not simultaneously consider class imbalance and catastrophic forgetting. Moreover, these methods usually require more memory to store previous versions of the model, which sometimes is not able to hold a copy of a substantial deep model due to memory constraints. Second, existing rehearsal approaches pay little attention to selecting and storing critical instances that help the model to retain as much knowledge of the learned tasks. This study focuses on dealing with these challenges by proposing several novel methods.
We first proposed a new loss function that combines two loss terms to deal with class imbalance data and catastrophic forgetting simultaneously. The former is a modification of a widely used loss function for class imbalance learning, called Focal loss, to handle the exploding gradient (loss goes to NaN) and the ability to learn from highly confident data points. At the same time, the latter is a novel loss term that addresses the catastrophic forgetting within the current mini-batch. In addition, we also propose an online convolution neural network (OCNN) architecture for tabular data that act as a base classifier in an ensemble system (OECNN). Next, we introduce a rehearsal-based method to prevent catastrophic forgetting. In which we select a triplet of instances within each mini-batch to store in the memory buffer. We find that these instances are identified as crucial instances
that can help either remind the model of easy tasks or revise for the hard ones. We also propose a class-wise forgetting detector that monitors the performance of each class encountered so far in a stream. If a class’s performance drops below a predefined
threshold, that class is identified as a forgetting class. Finally, based on the nature of data which often comprises many modalities, we study online multi-modal multi-task (M3T) learning problems. Unlike the traditional methods in stable environments, online M3T learning need to be considered in many scenarios like missing modalities and incremental tasks. We establish the setting for six frequently happened scenarios for M3T. Most of the existing works in M3T fail to run on all of these scenarios. Therefore, we propose a novel M3T deep learning model called UniCNet that can work on all of these scenarios and achieves superior performance compared with state-of-the-art M3T methods.
To conclude, this dissertation contributes to novel computational techniques that deal with catastrophic forgetting problem in continual deep learning.Thesis (PhD Doctorate)
Doctor of Philosophy (PhD)
School of Info & Comm Tech
Science, Environment, Engineering and Technology
Full Text
4
Idani, Arman. "Assessment of individual differences in online social networks using machine learning." Thesis, University of Cambridge, 2017. https://www.repository.cam.ac.uk/handle/1810/270109.
Анотація:
The services that define our personal and professional lives are increasingly accessed through digital devices, which store extensive records of our behaviour. An individual's psychological profile can be accurately assessed using offline behaviour, and I investigate if an automated machine learning system can measure the same psychological factors, only from observing the footprints of online behaviour, without observing any offline behaviour or any direct input from the individual. Prior research shows that psychological traits such as personality can be predicted using these digital footprints, although current state-of-the-art accuracy is below psychometric standards of reliability and self-reports consistently outperform machine-ratings in external validity. I introduce a new machine learning system that is capable of doing five-factor personality assessments, as well as other psychological assessments, from online data as accurately as self-report questionnaires in terms of reliability, internal consistency and external and discriminant validity, and demonstrate that passive psychological assessment can be a realistic option in addition to self-report questionnaires for both research and practice. Achieving this goal is not possible using conventional dimensionality reduction and linear regression models. Here I develop a supervised dimensionality reduction method capable of intelligently selecting only useful parts of data for the relevant prediction at hand which also does not lose variance when eliminating redundancies. In the learning stage, instead of linear regression models, I use an ensemble of decision trees which are able to distinguish scenarios where the same observations on digital data can mean different things for different individuals. This work highlights the interesting idea that similar to how a human expert who is able to assess personality from offline behaviour, an expert machine learning system is able to assess personality from online behaviour. It also demonstrates that big-5 personality are predictors of how predictable users are in social media, with neuroticism having the greatest correlation with unpredictability, while openness having the greatest correlation with predictability.
5
Zhang, Xuan. "Product Defect Discovery and Summarization from Online User Reviews." Diss., Virginia Tech, 2018. http://hdl.handle.net/10919/85581.
Анотація:
Product defects concern various groups of people, such as customers, manufacturers, government officials, etc. Thus, defect-related knowledge and information are essential. In keeping with the growth of social media, online forums, and Internet commerce, people post a vast amount of feedback on products, which forms a good source for the automatic acquisition of knowledge about defects. However, considering the vast volume of online reviews, how to automatically identify critical product defects and summarize the related information from the huge number of user reviews is challenging, even when we target only the negative reviews. As a kind of opinion mining research, existing defect discovery methods mainly focus on how to classify the type of product issues, which is not enough for users. People expect to see defect information in multiple facets, such as product model, component, and symptom, which are necessary to understand the defects and quantify their influence. In addition, people are eager to seek problem resolutions once they spot defects. These challenges cannot be solved by existing aspect-oriented opinion mining models, which seldom consider the defect entities mentioned above. Furthermore, users also want to better capture the semantics of review text, and to summarize product defects more accurately in the form of natural language sentences. However, existing text summarization models including neural networks can hardly generalize to user review summarization due to the lack of labeled data.
In this research, we explore topic models and neural network models for product defect discovery and summarization from user reviews. Firstly, a generative Probabilistic Defect Model (PDM) is proposed, which models the generation process of user reviews from key defect entities including product Model, Component, Symptom, and Incident Date. Using the joint topics in these aspects, which are produced by PDM, people can discover defects which are represented by those entities. Secondly, we devise a Product Defect Latent Dirichlet Allocation (PDLDA) model, which describes how negative reviews are generated from defect elements like Component, Symptom, and Resolution. The interdependency between these entities is modeled by PDLDA as well. PDLDA answers not only what the defects look like, but also how to address them using the crowd wisdom hidden in user reviews. Finally, the problem of how to summarize user reviews more accurately, and better capture the semantics in them, is studied using deep neural networks, especially Hierarchical Encoder-Decoder Models.
For each of the research topics, comprehensive evaluations are conducted to justify the effectiveness and accuracy of the proposed models, on heterogeneous datasets. Further, on the theoretical side, this research contributes to the research stream on product defect discovery, opinion mining, probabilistic graphical models, and deep neural network models. Regarding impact, these techniques will benefit related users such as customers, manufacturers, and government officials.Ph. D.
Product defects concern various groups of people, such as customers, manufacturers, and government officials. Thus, defect-related knowledge and information are essential. In keeping with the growth of social media, online forums, and Internet commerce, people post a vast amount of feedback on products, which forms a good source for the automatic acquisition of knowledge about defects. However, considering the vast volume of online reviews, how to automatically identify critical product defects and summarize the related information from the huge number of user reviews is challenging, even when we target only the negative reviews. People expect to see defect information in multiple facets, such as product model, component, and symptom, which are necessary to understand the defects and quantify their influence. In addition, people are eager to seek problem resolutions once they spot defects. Furthermore, users also want to better summarize product defects more accurately in the form of natural language sentences. These requirements cannot be satisfied by existing methods, which seldom consider the defect entities mentioned above, or hardly generalize to user review summarization. In this research, we develop novel Machine Learning (ML) algorithms for product defect discovery and summarization. Firstly, we study how to identify product defects and their related attributes, such as Product Model, Component, Symptom, and Incident Date. Secondly, we devise a novel algorithm, which can discover product defects and the related Component, Symptom, and Resolution, from online user reviews. This method tells not only what the defects look like, but also how to address them using the crowd wisdom hidden in user reviews. Finally, we address the problem of how to summarize user reviews in the form of natural language sentences using a paraphrase-style method. On the theoretical side, this research contributes to multiple research areas in Natural Language Processing (NLP), Information Retrieval (IR), and Machine Learning. Regarding impact, these techniques will benefit related users such as customers, manufacturers, and government officials.
6
Nguyen, Trang Pham Ngoc. "A privacy preserving online learning framework for medical diagnosis applications." Thesis, Edith Cowan University, Research Online, Perth, Western Australia, 2022. https://ro.ecu.edu.au/theses/2503.
Анотація:
Electronic Health records are an important part of a digital healthcare system. Due to their significance, electronic health records have become a major target for hackers, and hospitals/clinics prefer to keep the records at local sites protected by adequate security measures. This introduces challenges in sharing health records. Sharing health records however, is critical in building an accurate online diagnosis framework. Most local sites have small data sets, and machine learning models developed locally based on small data sets, do not have knowledge about other data sets and learning models used at other sites.
The work in this thesis utilizes the framework of coordinating the blockchain technology and online training mechanism in order to address the concerns of privacy and security in a methodical manner. Specifically, it integrates online learning with a permissioned blockchain network, using transaction metadata to broadcast a part of models while keeping patient health information private. This framework can treat different types of machine learning models using the same distributed dataset. The study also outlines the advantages and drawbacks of using blockchain technology to tackle the privacy-preserving predictive modeling problem and to improve interoperability amongst institutions. This study implements the proposed solutions for skin cancer diagnosis as a representative case and shows promising results in preserving security and providing high detection accuracy. The experimentation was done on ISIC dataset, and the results were 98.57, 99.13, 99.17 and 97,18 in terms of precision, accuracy, F1-score and recall, respectively.
7
Adewopo, Victor A. "Exploring Open Source Intelligence for cyber threat Prediction." University of Cincinnati / OhioLINK, 2021. http://rave.ohiolink.edu/etdc/view?acc_num=ucin162491804723753.
8
Al, Rawashdeh Khaled. "Toward a Hardware-assisted Online Intrusion Detection System Based on Deep Learning Algorithms for Resource-Limited Embedded Systems." University of Cincinnati / OhioLINK, 2018. http://rave.ohiolink.edu/etdc/view?acc_num=ucin1535464571843315.
9
Meyer, Lucas. "Deep Learning en Ligne pour la Simulation Numérique à Grande Échelle." Electronic Thesis or Diss., Université Grenoble Alpes, 2024. http://www.theses.fr/2024GRALM001.
Анотація:
Nombreuses applications d’ingénieries et recherches scientifiques nécessitent la simulation fidèle de phénomènes complexes et dynamiques, transcrits mathématiquement en Équations aux Dérivées Partielles (EDP). Les solutions de ces EDP sont généralement approximées au moyen de solveurs qui effectuent des calculs intenses et génèrent des quantités importantes de données. Les applications requièrent rarement une unique simulation, mais plutôt un ensemble d’exécutions pour différents paramètres afin d’analyser la sensibilité du phénomène ou d’en trouver une configuration optimale. Ces larges ensembles de simulations sont limités par des temps de calcul importants et des capacités de stockage mémoire finies. La problématique du coût de calcul a jusqu’à présent encouragé le développement du calcul haute-performance (HPC) et de techniques de réductions de modèles. Récemment, le succès de l'apprentissage profond a poussé la communauté scientifique à considérer son usage pour accélérer ces ensembles de simulations. Cette thèse s'inscrit dans ce cadre en explorant tout d'abord deux techniques d’apprentissage pour la simulation numérique. La première propose d’utiliser une série de convolutions sur une hiérarchie de graphes pour reproduire le champ de vitesse d’un fluide tel que généré par le solveur à tout pas de temps de la simulation. La seconde hybride des algorithmes de régression avec des techniques classiques de réduction de modèles pour prédire les coefficients de toute nouvelle simulation dans une base réduite obtenue par analyse en composantes principales. Ces deux approches, comme la majorité de celles présentées dans la littérature, sont supervisées. Leur entraînement nécessite de générer a priori de nombreuses simulations. Elles souffrent donc du même problème qui a motivé leur développement : générer un jeu d’entraînement de simulations fidèles à grande échelle est laborieux. Nous proposons un cadre d’apprentissage générique pour l’entraînement de réseaux de neurones artificiels à partir de simulations générées à la volée tirant profit des ressources HPC. Les données sont produites en exécutant simultanément plusieurs instances d’un solveur pour différents paramètres. Le solveur peut lui-même être parallélisé sur plusieurs unités de calcul. Dès qu’un pas de temps est simulé, il est directement transmis pour effectuer l’apprentissage. Aucune donnée générée par le solveur n’est donc sauvegardée sur disque, évitant ainsi les coûteuses opérations d’écriture et de lecture et la nécessité de grands volumes de stockage. L’apprentissage se fait selon une distribution parallèle des données sur plusieurs GPUs. Comme il est désormais en ligne, cela crée un biais dans les données d’entraînement, comparativement à une situation classique où les données sont échantillonnées uniformément sur un ensemble de simulations disponibles a priori. Nous associons alors chaque GPU à une mémoire tampon en charge de mélanger les données produites. Ce formalisme a permis d’améliorer les capacités de généralisation de modèles issus de l’état de l’art, en les exposant à une diversité globale de données simulées plus riches qu’il n’aurait été faisable lors d’un entraînement classique. Des expériences montrent que l’implémentation de la mémoire tampon est cruciale pour garantir un entraînement de qualité à haut débit. Ce cadre d’apprentissage a permis d’entraîner un réseau à reproduire des simulations de diffusion thermique en moins de 2 heures sur 8TB de données générées et traitées in situ, améliorant ainsi les prédictions de 47% par rapport à un entraînement classiqueMany engineering applications and scientific discoveries rely on faithful numerical simulations of complex phenomena. These phenomena are transcribed mathematically into Partial Differential Equation (PDE), whose solution is generally approximated by solvers that perform intensive computation and generate tremendous amounts of data. The applications rarely require only one simulation but rather a large ensemble of runs for different parameters to analyze the sensitivity of the phenomenon or to find an optimal configuration. Those large ensemble runs are limited by computation time and finite memory capacity. The high computational cost has led to the development of high-performance computing (HPC) and surrogate models. Recently, pushed by the success of deep learning in computer vision and natural language processing, the scientific community has considered its use to accelerate numerical simulations. The present thesis follows this approach by first presenting two techniques using machine learning for surrogate models. First, we propose to use a series of convolutions on hierarchical graphs to reproduce the velocity of fluids as generated by solvers at any time of the simulation. Second, we hybridize regression algorithms with classical reduced-order modeling techniques to identify the coefficients of any new simulation in a reduced basis computed by proper orthogonal decomposition. These two approaches, as the majority found in the literature, are supervised. Their training needs to generate a large number of simulations. Thus, they suffer the same problem that motivated their development in the first instance: generating many faithful simulations at scale is laborious. We propose a generic training framework for artificial neural networks that generate data simulations on-the-fly by leveraging HPC resources. Data are produced by running simultaneously several instances of the solver for different parameters. The solver itself can be parallelized over several processing units. As soon as a time step is computed by any simulation, it is streamed for training. No data is ever written on disk, thus overcoming slow input-output operations and alleviating the memory footprint. Training is performed by several GPUs with distributed data-parallelism. Because the training is now online, it induces a bias in the data compared to classical training, for which they are sampled uniformly from an ensemble of simulations available a priori. To mitigate this bias, each GPU is associated with a memory buffer in charge of mixing the incoming simulation data. This framework has improved the generalization capabilities of state-of-the-art architectures by exposing them during training to a richer diversity of data than would have been feasible with classical training. Experiments show the importance of the memory buffer implementation in guaranteeing generalization capabilities and high throughput training. The framework has been used to train a deep surrogate for heat diffusion simulation in less than 2 hours on 8TB of data processed in situ, thus increasing the prediction accuracy by 47% compared to a classical setting
10
Qiao, Zhilei. "Consumer-Centric Innovation for Mobile Apps Empowered by Social Media Analytics." Diss., Virginia Tech, 2018. http://hdl.handle.net/10919/95983.
Анотація:
Due to the rapid development of Internet communication technologies (ICTs), an increasing number of social media platforms exist where consumers can exchange comments online about products and services that businesses offer. The existing literature has demonstrated that online user-generated content can significantly influence consumer behavior and increase sales. However, its impact on organizational operations has been primarily focused on marketing, with other areas understudied. Hence, there is a pressing need to design a research framework that explores the impact of online user-generated content on important organizational operations such as product innovation, customer relationship management, and operations management. Research efforts in this dissertation center on exploring the co-creation value of online consumer reviews, where consumers' demands influence firms' decision-making. The dissertation is composed of three studies. The first study finds empirical evidence that quality signals in online product reviews are predictors of the timing of firms' incremental innovation. Guided by the product differentiation theory, the second study examines how companies' innovation and marketing differentiation strategies influence app performance. The last study proposes a novel text analytics framework to discover different information types from user reviews. The research contributes theoretical and practical insights to consumer-centric innovation and social media analytics literature.PHD
11
Jesse, Edel. "Student Attitudes Toward Use of Massive Open Online Courses." University of Dayton / OhioLINK, 2019. http://rave.ohiolink.edu/etdc/view?acc_num=dayton1573740761560753.
12
Khan, Pour Hamed. "Computational Approaches for Analyzing Social Support in Online Health Communities." Thesis, University of North Texas, 2018. https://digital.library.unt.edu/ark:/67531/metadc1157594/.
Анотація:
Online health communities (OHCs) have become a medium for patients to share their personal experiences and interact with peers on topics related to a disease, medication, side effects, and therapeutic processes. Many studies show that using OHCs regularly decreases mortality and improves patients mental health. As a result of their benefits, OHCs are a popular place for patients to refer to, especially patients with a severe disease, and to receive emotional and informational support. The main reasons for developing OHCs are to present valid and high-quality information and to understand the mechanism of social support in changing patients' mental health. Given the purpose of OHC moderators for developing OHCs applications and the purpose of patients for using OHCs, there is no facility, feature, or sub-application in OHCs to satisfy patient and moderator goals. OHCs are only equipped with a primary search engine that is a keyword-based search tool. In other words, if a patient wants to obtain information about a side-effect, he/she needs to browse many threads in the hope that he/she can find several related comments. In the same way, OHC moderators cannot browse all information which is exchanged among patients to validate their accuracy. Thus, it is critical for OHCs to be equipped with computational tools which are supported by several sophisticated computational models that provide moderators and patients with the collection of messages that they need for making decisions or predictions. We present multiple computational models to alleviate the problem of OHCs in providing specific types of messages in response to the specific moderator and patient needs. Specifically, we focused on proposing computational models for the following tasks: identifying emotional support, which presents OHCs moderators, psychologists, and sociologists with insightful views on the emotional states of individuals and groups, and identifying informational support, which provides patients with an efficient and effective tool for accessing the best-fit messages from a huge amount of patient posts to satisfy their information needs, as well as provides OHC moderators, health-practitioners, nurses, and doctors with an insightful view about the current discussion under the topics of side-effects and therapeutic processes, giving them an opportunity to monitor and validate the exchange of information in OHCs. We proposed hybrid models that combine high-level, abstract features extracted from convolutional neural networks with lexicon-based features and features extracted from long short-term memory networks to capture the semantics of the data. We show that our models, with and without lexicon-based features, outperform strong baselines.
13
Kotsch, Janeen S. "EXPLORING STUDENTS’ EXPERIENCES OF CONCEPT-BASED LEARNING IN AN ASYNCHRONOUS ONLINE PHARMACOLOGY COURSE: AN INTERPRETIVE STUDY." Kent State University / OhioLINK, 2021. http://rave.ohiolink.edu/etdc/view?acc_num=kent161787487052164.
14
Zarei, Koosha. "Fake identity & fake activity detection in online social networks based on transfer learning." Electronic Thesis or Diss., Institut polytechnique de Paris, 2022. http://www.theses.fr/2022IPPAS008.
Анотація:
Les médias sociaux ont permis de connecter un plus grand nombre de personnes dans le monde entier et d'accroître la facilité d'accès à des contenus gratuits, mais ils sont confrontés à des phénomènes critiques tels que les faux contenus, les fausses identités et les fausses activités. La détection de faux contenus sur les médias sociaux est récemment devenue une recherche émergente qui attire une attention considérable. une recherche émergente qui suscite une attention considérable. Dans ce domaine, les fausses identités jouent un rôle important dans la production et la propagation de faux contenus dans les réseaux sociaux en ligne réseaux sociaux en ligne tels que Meta (Facebook), Twitter et Instagram. La principale raison de ce phénomène est que les médias sociaux encouragent les usurpateurs d'identité, les comptes malveillants, les trolls et les robots sociaux àproduire du contenu et interagir avec des humains ou d'autres robots sans tenir compte de la crédibilité du contenu. La détection de faux contenus sur les médias sociaux est récemment devenue une recherche émergente qui attire une attention considérable. une recherche émergente qui suscite une attention considérable. Dans ce domaine, les fausses identités jouent un rôle important dans la production et la propagation de faux contenus dans les réseaux sociaux en ligne réseaux sociaux en ligne tels que Meta (Facebook), Twitter et Instagram. La principale raison de ce phénomène est que les médias sociaux encouragent les usurpateurs d'identité, les comptes malveillants, les trolls et les robots sociaux à produire du contenu et interagir avec des humains ou d'autres bots sans tenir compte de la crédibilité du contenu et inciter les utilisateurs à cliquer et à les partager.L'objectif de cette thèse est d'étudier le problème des fausses identités, des fausses activités et du contenu authentique qu'elles génèrent dans les médias sociaux et de proposer des algorithmes pour classifier le contenu factice. Nous définissons le faux contenu comme un élément d'information vérifiable et faux partagé intentionnellement pour tromper les lecteurs. Je propose différentes approches dans lesquelles j'adapte des modèles avancés de Transfer Learning (TL) et des techniques NLP pour détecter les fausses identités et classer le faux contenu automatiquementWhile Social Media has connected more people around the world and has increased the easeof access to free content, but is dealing with critical phenomena such as fake content, fakeidentities, and fake activities. Fake content detection on social media has recently becomeemerging research that is attracting tremendous attention. In this area, fake identities areplaying an important role in the production and propagation of fake content in Online SocialNetworks such as Meta (Facebook), Twitter, and Instagram. The main reason behind thisis that social media encourages impersonators, malicious accounts, trolls, and social bots toproduce content and interact with humans or other bots without considering the credibilityof the content and entice users to click and share them.In this thesis, I primarily concentrate on impersonators as one of the concerning vari-eties of fake identities. These entities are nefarious fake accounts that intend to disguise alegitimate account by making similar profiles and then striking social media with fake con-tent, which makes it considerably harder to understand which posts are genuinely produced.The recent advancements in Natural Language Processing (NLP), and Transformer-basedLanguage Models (LM) can be adapted to develop automatic methods for many relatedNLP downstream tasks in this area. Language Models and their flexibility to cope withany corpus delivering great results has made this approach very popular. The fake contentclassification can be handled using Pretrained Language Models (PLM) and accurate deeplearning models.The aim of this thesis is to investigate the problem of fake identities, fake activities, andtheir generated ingenuine content in social media and propose algorithms to classify fakecontent. We define fake content as a verifiably false piece of information shared intentionallyto mislead the readers. I propose different approaches in which I adapt advanced TransferLearning (TL) models and NLP techniques to detect fake identities and classify fake contentautomatically
15
Martin, Alice. "Deep learning models and algorithms for sequential data problems : applications to language modelling and uncertainty quantification." Electronic Thesis or Diss., Institut polytechnique de Paris, 2022. http://www.theses.fr/2022IPPAS007.
Анотація:
Dans ce manuscrit de thèse, nous développons de nouveaux algorithmes et modèles pour résoudre les problèmes d'apprentissage profond sur de la donnée séquentielle, en partant des problématiques posées par l'apprentissage des modèles de langage basés sur des réseaux de neurones. Un premier axe de recherche développe de nouveaux modèles génératifs profonds basés sur des méthodes de Monte Carlo Séquentielles (SMC), qui permettent de mieux modéliser la diversité du langage, ou de mieux quantifier l'incertitude pour des problèmes de régression séquentiels. Un deuxième axe de recherche vise à faciliter l'utilisation de techniques de SMC dans le cadre de l'apprentissage profond, en développant un nouvel algorithme de lissage à coût computationnel largement réduit, et qui s'applique à un scope plus large de modèles à espace d'états, notamment aux modèles génératifs profonds. Finalement, un troisième axe de recherche propose le premier algorithme d'apprentissage par renforcement permettant d'apprendre des modèles de langage conditionnels "ex-nihilo" (i.e sans jeu de données supervisé), basé sur un mécanisme de troncation de l'espace d'actions par un modèle de langage pré-entrainéIn this thesis, we develop new models and algorithms to solve deep learning tasks on sequential data problems, with the perspective of tackling the pitfalls of current approaches for learning language models based on neural networks. A first research work develops a new deep generative model for sequential data based on Sequential Monte Carlo Methods, that enables to better model diversity in language modelling tasks, and better quantify uncertainty in sequential regression problems. A second research work aims to facilitate the use of SMC techniques within deep learning architectures, by developing a new online smoothing algorithm with reduced computational cost, and applicable on a wider scope of state-space models, including deep generative models. Finally, a third research work proposes the first reinforcement learning that enables to learn conditional language models from scratch (i.e without supervised datasets), based on a truncation mechanism of the natural language action space with a pretrained language model
16
Trenholm, Sven. "Adaptation of tertiary mathematics instruction to the virtual medium : approaches to assessment practice." Thesis, Loughborough University, 2013. https://dspace.lboro.ac.uk/2134/12561.
Анотація:
Mathematics has been singled out as a challenging discipline to teach fully online (FO). Yet both the demand for and development of FO mathematics courses is increasing with little known about the quality of these courses and many calling for research. Whereas most research has investigated the nature of these courses by examining instructional outputs such as student grades this research seeks the same insight but by examining instructional inputs. Specifically, it seeks to investigate the nature of current assessment practice in FO mathematics courses. To conduct this investigation, deep learning (Marton & S??lj??, 1976a, 1976b) is used as the principle theoretical framework. From the growing body of literature associated with deep learning, two studies are selected to investigate current FO mathematics instructors assessment practices. An additional framework based on empirical findings related to the use of different kinds of feedback is also used. In total, six study measures are used to conduct a mixed methods study in two parts. The target demographic and course context are tertiary instructors from Western nations that teach introductory level mathematics (particularly statistics and calculus). The first study explores current FO mathematics assessment practices using an online survey (n=70) where the majority of participants originate from US higher education institutions. In the second study six of the US survey participants are interviewed about how their assessment practices and approaches used in their FO mathematics courses differ from those used in their face-to-face (F2F) mathematics courses. This study represents the first known attempt to investigate the nature of tertiary FO mathematics instructors assessment practices using appropriate theoretical frameworks. In particular, it investigates mathematics instructors experiences of the affordances and constraints of the FO course context when adapting their F2F practice to this new environment. Findings suggest the FO course context is a challenging environment for instructors to orient their teaching and assessment practice in a way that helps develop students understanding of mathematics. Analysis of interview responses suggests the problem lies with the nature of interactivity provided in the FO course context.
17
Yang, Lixuan. "Structuring of image databases for the suggestion of products for online advertising." Thesis, Paris, CNAM, 2017. http://www.theses.fr/2017CNAM1102/document.
Анотація:
Le sujet de la thèse est l'extraction et la segmentation des vêtements à partir d'images en utilisant des techniques de la vision par ordinateur, de l'apprentissage par ordinateur et de la description d'image, pour la recommandation de manière non intrusive aux utilisateurs des produits similaires provenant d'une base de données de vente. Nous proposons tout d'abord un extracteur d'objets dédié à la segmentation de la robe en combinant les informations locales avec un apprentissage préalable. Un détecteur de personne localises des sites dans l'image qui est probable de contenir l'objet. Ensuite, un processus d'apprentissage intra-image en deux étapes est est développé pour séparer les pixels de l'objet de fond. L'objet est finalement segmenté en utilisant un algorithme de contour actif qui prend en compte la segmentation précédente et injecte des connaissances spécifiques sur la courbure locale dans la fonction énergie. Nous proposons ensuite un nouveau framework pour l'extraction des vêtements généraux en utilisant une procédure d'ajustement globale et locale à trois étapes. Un ensemble de modèles initialises un processus d'extraction d'objet par un alignement global du modèle, suivi d'une recherche locale en minimisant une mesure de l'inadéquation par rapport aux limites potentielles dans le voisinage. Les résultats fournis par chaque modèle sont agrégés, mesuré par un critère d'ajustement globale, pour choisir la segmentation finale. Dans notre dernier travail, nous étendons la sortie d'un réseau de neurones Fully Convolutional Network pour inférer le contexte à partir d'unités locales (superpixels). Pour ce faire, nous optimisons une fonction énergie, qui combine la structure à grande échelle de l'image avec le local structure superpixels, en recherchant dans l'espace de toutes les possibilité d'étiquetage. De plus, nous introduisons une nouvelle base de données RichPicture, constituée de 1000 images pour l'extraction de vêtements à partir d'images de mode. Les méthodes sont validées sur la base de données publiques et se comparent favorablement aux autres méthodes selon toutes les mesures de performance considéréesThe topic of the thesis is the extraction and segmentation of clothing items from still images using techniques from computer vision, machine learning and image description, in view of suggesting non intrusively to the users similar items from a database of retail products. We firstly propose a dedicated object extractor for dress segmentation by combining local information with a prior learning. A person detector is applied to localize sites in the image that are likely to contain the object. Then, an intra-image two-stage learning process is developed to roughly separate foreground pixels from the background. Finally, the object is finely segmented by employing an active contour algorithm that takes into account the previous segmentation and injects specific knowledge about local curvature in the energy function.We then propose a new framework for extracting general deformable clothing items by using a three stage global-local fitting procedure. A set of template initiates an object extraction process by a global alignment of the model, followed by a local search minimizing a measure of the misfit with respect to the potential boundaries in the neighborhood. The results provided by each template are aggregated, with a global fitting criterion, to obtain the final segmentation.In our latest work, we extend the output of a Fully Convolution Neural Network to infer context from local units(superpixels). To achieve this we optimize an energy function,that combines the large scale structure of the image with the locallow-level visual descriptions of superpixels, over the space of all possiblepixel labellings. In addition, we introduce a novel dataset called RichPicture, consisting of 1000 images for clothing extraction from fashion images.The methods are validated on the public database and compares favorably to the other methods according to all the performance measures considered
18
Bergkvist, Alexander, Nils Hedberg, Sebastian Rollino, and Markus Sagen. "Surmize: An Online NLP System for Close-Domain Question-Answering and Summarization." Thesis, Uppsala universitet, Institutionen för informationsteknologi, 2020. http://urn.kb.se/resolve?urn=urn:nbn:se:uu:diva-412247.
Анотація:
The amount of data available and consumed by people globally is growing. To reduce mental fatigue and increase the general ability to gain insight into complex texts or documents, we have developed an application to aid in this task. The application allows users to upload documents and ask domain-specific questions about them using our web application. A summarized version of each document is presented to the user, which could further facilitate their understanding of the document and guide them towards what types of questions could be relevant to ask. Our application allows users flexibility with the types of documents that can be processed, it is publicly available, stores no user data, and uses state-of-the-art models for its summaries and answers. The result is an application that yields near human-level intuition for answering questions in certain isolated cases, such as Wikipedia and news articles, as well as some scientific texts. The application shows a decrease in reliability and its prediction as to the complexity of the subject, the number of words in the document, and grammatical inconsistency in the questions increases. These are all aspects that can be improved further if used in production.Mängden data som är tillgänglig och konsumeras av människor växer globalt. För att minska den mentala trötthet och öka den allmänna förmågan att få insikt i komplexa, massiva texter eller dokument, har vi utvecklat en applikation för att bistå i de uppgifterna. Applikationen tillåter användare att ladda upp dokument och fråga kontextspecifika frågor via vår webbapplikation. En sammanfattad version av varje dokument presenteras till användaren, vilket kan ytterligare förenkla förståelsen av ett dokument och vägleda dem mot vad som kan vara relevanta frågor att ställa. Vår applikation ger användare möjligheten att behandla olika typer av dokument, är tillgänglig för alla, sparar ingen personlig data, och använder de senaste modellerna inom språkbehandling för dess sammanfattningar och svar. Resultatet är en applikation som når en nära mänsklig intuition för vissa domäner och frågor, som exempelvis Wikipedia- och nyhetsartiklar, samt viss vetensaplig text. Noterade undantag för tillämpningen härrör från ämnets komplexitet, grammatiska korrekthet för frågorna och dokumentets längd. Dessa är områden som kan förbättras ytterligare om den används i produktionen.
19
Yang, Lixuan. "Structuring of image databases for the suggestion of products for online advertising." Electronic Thesis or Diss., Paris, CNAM, 2017. http://www.theses.fr/2017CNAM1102.
Анотація:
Le sujet de la thèse est l'extraction et la segmentation des vêtements à partir d'images en utilisant des techniques de la vision par ordinateur, de l'apprentissage par ordinateur et de la description d'image, pour la recommandation de manière non intrusive aux utilisateurs des produits similaires provenant d'une base de données de vente. Nous proposons tout d'abord un extracteur d'objets dédié à la segmentation de la robe en combinant les informations locales avec un apprentissage préalable. Un détecteur de personne localises des sites dans l'image qui est probable de contenir l'objet. Ensuite, un processus d'apprentissage intra-image en deux étapes est est développé pour séparer les pixels de l'objet de fond. L'objet est finalement segmenté en utilisant un algorithme de contour actif qui prend en compte la segmentation précédente et injecte des connaissances spécifiques sur la courbure locale dans la fonction énergie. Nous proposons ensuite un nouveau framework pour l'extraction des vêtements généraux en utilisant une procédure d'ajustement globale et locale à trois étapes. Un ensemble de modèles initialises un processus d'extraction d'objet par un alignement global du modèle, suivi d'une recherche locale en minimisant une mesure de l'inadéquation par rapport aux limites potentielles dans le voisinage. Les résultats fournis par chaque modèle sont agrégés, mesuré par un critère d'ajustement globale, pour choisir la segmentation finale. Dans notre dernier travail, nous étendons la sortie d'un réseau de neurones Fully Convolutional Network pour inférer le contexte à partir d'unités locales (superpixels). Pour ce faire, nous optimisons une fonction énergie, qui combine la structure à grande échelle de l'image avec le local structure superpixels, en recherchant dans l'espace de toutes les possibilité d'étiquetage. De plus, nous introduisons une nouvelle base de données RichPicture, constituée de 1000 images pour l'extraction de vêtements à partir d'images de mode. Les méthodes sont validées sur la base de données publiques et se comparent favorablement aux autres méthodes selon toutes les mesures de performance considéréesThe topic of the thesis is the extraction and segmentation of clothing items from still images using techniques from computer vision, machine learning and image description, in view of suggesting non intrusively to the users similar items from a database of retail products. We firstly propose a dedicated object extractor for dress segmentation by combining local information with a prior learning. A person detector is applied to localize sites in the image that are likely to contain the object. Then, an intra-image two-stage learning process is developed to roughly separate foreground pixels from the background. Finally, the object is finely segmented by employing an active contour algorithm that takes into account the previous segmentation and injects specific knowledge about local curvature in the energy function.We then propose a new framework for extracting general deformable clothing items by using a three stage global-local fitting procedure. A set of template initiates an object extraction process by a global alignment of the model, followed by a local search minimizing a measure of the misfit with respect to the potential boundaries in the neighborhood. The results provided by each template are aggregated, with a global fitting criterion, to obtain the final segmentation.In our latest work, we extend the output of a Fully Convolution Neural Network to infer context from local units(superpixels). To achieve this we optimize an energy function,that combines the large scale structure of the image with the locallow-level visual descriptions of superpixels, over the space of all possiblepixel labellings. In addition, we introduce a novel dataset called RichPicture, consisting of 1000 images for clothing extraction from fashion images.The methods are validated on the public database and compares favorably to the other methods according to all the performance measures considered
20
Simao, Miguel. "Segmentation et reconaissance des gestes pour l'interaction homme-robot cognitive." Thesis, Paris, ENSAM, 2018. http://www.theses.fr/2018ENAM0048/document.
Анотація:
Cette thèse présente un cadre formel pour l'interaction Homme-robot (HRI), qui reconnaître un important lexique de gestes statiques et dynamiques mesurés par des capteurs portatifs. Gestes statiques et dynamiques sont classés séparément grâce à un processus de segmentation. Les tests expérimentaux sur la base de données de gestes UC2017 ont montré une haute précision de classification. La classification pas à pas en ligne utilisant des données brutes est fait avec des réseaux de neurones profonds « Long-Short Term Memory » (LSTM) et à convolution (CNN), et sont plus performants que les modèles statiques entraînés avec des caractéristiques spécialement conçues, au détriment du temps d'entraînement et d'inférence. La classification en ligne des gestes permet une classification prédictive avec réussit. Le rejet des gestes hors vocabulaire est proposé par apprentissage semi-supervisé par un réseau de neurones du type « Auxiliary Conditional Generative Adversarial Networks ». Le réseau propose a atteint une haute précision de rejet de les gestes non entraînés de la base de données UC2018 DualMyoThis thesis presents a human-robot interaction (HRI) framework to classify large vocabularies of static and dynamic hand gestures, captured with wearable sensors. Static and dynamic gestures are classified separately thanks to the segmentation process. Experimental tests on the UC2017 hand gesture dataset showed high accuracy. In online frame-by-frame classification using raw incomplete data, Long Short-Term Memory (LSTM) deep networks and Convolutional Neural Networks (CNN) performed better than static models with specially crafted features at the cost of training and inference time. Online classification of dynamic gestures allows successful predictive classification. The rejection of out-of-vocabulary gestures is proposed to be done through semi-supervised learning of a network in the Auxiliary Conditional Generative Adversarial Networks framework. The proposed network achieved a high accuracy on the rejection of untrained patterns of the UC2018 DualMyo dataset
21
Buttar, Sarpreet Singh. "Applying Artificial Neural Networks to Reduce the Adaptation Space in Self-Adaptive Systems : an exploratory work." Thesis, Linnéuniversitetet, Institutionen för datavetenskap och medieteknik (DM), 2019. http://urn.kb.se/resolve?urn=urn:nbn:se:lnu:diva-87117.
Анотація:
Self-adaptive systems have limited time to adjust their configurations whenever their adaptation goals, i.e., quality requirements, are violated due to some runtime uncertainties. Within the available time, they need to analyze their adaptation space, i.e., a set of configurations, to find the best adaptation option, i.e., configuration, that can achieve their adaptation goals. Existing formal analysis approaches find the best adaptation option by analyzing the entire adaptation space. However, exhaustive analysis requires time and resources and is therefore only efficient when the adaptation space is small. The size of the adaptation space is often in hundreds or thousands, which makes formal analysis approaches inefficient in large-scale self-adaptive systems. In this thesis, we tackle this problem by presenting an online learning approach that enables formal analysis approaches to analyze large adaptation spaces efficiently. The approach integrates with the standard feedback loop and reduces the adaptation space to a subset of adaptation options that are relevant to the current runtime uncertainties. The subset is then analyzed by the formal analysis approaches, which allows them to complete the analysis faster and efficiently within the available time. We evaluate our approach on two different instances of an Internet of Things application. The evaluation shows that our approach dramatically reduces the adaptation space and analysis time without compromising the adaptation goals.
22
RANDAZZO, VINCENZO. "Novel neural approaches to data topology analysis and telemedicine." Doctoral thesis, Politecnico di Torino, 2020. http://hdl.handle.net/11583/2850610.
23
Barty, Karin, and edu au jillj@deakin edu au mikewood@deakin edu au kimg@deakin. "Students' experiences of e-learning at school." Deakin University. School of Education, 2001. http://tux.lib.deakin.edu.au./adt-VDU/public/adt-VDU20040614.145900.
Анотація:
The dissertation describes the experiences of senior secondary students taking an online course for the first time to further their language education. The experiences are presented from the perspective of students, of supervising teachers and the 'virtual' teacher. Issues of importance with younger learners are identified and discussed and guidelines for the conduct of online courses at school level developed. It is proposed that online courses may have a worthwhile place in school education if specific learning needs can be met using this medium.
24
Würfel, Max. "Online advertising revenue forecasting: an interpretable deep learning approach." Master's thesis, 2021. http://hdl.handle.net/10362/122676.
Анотація:
This paper investigates whether publishers’ Google AdSense online advertising revenues can be predicted from peekd’s proprietary database using deep learning methodologies. Peekd is a Berlin (Germany) based data science company, which primarily provides e Retailers with sales and shopper intelligence. I find that using a single deep learning model, AdSense revenues can be predicted across publishers. Additionally, using unsupervised clustering, publishers were grouped and related time series were fed as covariates when making predictions. No performance improvement was found in relation with this technique. Finally, I find that in the short-term, publishers’ AdSense revenues embed similar temporal patterns as web traffic.
25
"Efficient and Online Deep Learning through Model Plasticity and Stability." Doctoral diss., 2020. http://hdl.handle.net/2286/R.I.62959.
Анотація:
abstract: The rapid advancement of Deep Neural Networks (DNNs), computing, and sensing technology has enabled many new applications, such as the self-driving vehicle, the surveillance drone, and the robotic system. Compared to conventional edge devices (e.g. cell phone or smart home devices), these emerging devices are required to deal with much more complicated and dynamic situations in real-time with bounded computation resources. However, there are several challenges, including but not limited to efficiency, real-time adaptation, model stability, and automation of architecture design.
To tackle the challenges mentioned above, model plasticity and stability are leveraged to achieve efficient and online deep learning, especially in the scenario of learning streaming data at the edge:
First, a dynamic training scheme named Continuous Growth and Pruning (CGaP) is proposed to compress the DNNs through growing important parameters and pruning unimportant ones, achieving up to 98.1% reduction in the number of parameters.
Second, this dissertation presents Progressive Segmented Training (PST), which targets catastrophic forgetting problems in continual learning through importance sampling, model segmentation, and memory-assisted balancing. PST achieves state-of-the-art accuracy with 1.5X FLOPs reduction in the complete inference path.
Third, to facilitate online learning in real applications, acquisitive learning (AL) is further proposed to emphasize both knowledge inheritance and acquisition: the majority of the knowledge is first pre-trained in the inherited model and then adapted to acquire new knowledge. The inherited model's stability is monitored by noise injection and the landscape of the loss function, while the acquisition is realized by importance sampling and model segmentation. Compared to a conventional scheme, AL reduces accuracy drop by >10X on CIFAR-100 dataset, with 5X reduction in latency per training image and 150X reduction in training FLOPs.
Finally, this dissertation presents evolutionary neural architecture search in light of model stability (ENAS-S). ENAS-S uses a novel fitness score, which addresses not only the accuracy but also the model stability, to search for an optimal inherited model for the application of continual learning. ENAS-S outperforms hand-designed DNNs when learning from a data stream at the edge.
In summary, in this dissertation, several algorithms exploiting model plasticity and model stability are presented to improve the efficiency and accuracy of deep neural networks, especially for the scenario of continual learning.Dissertation/Thesis
Doctoral Dissertation Electrical Engineering 2020
26
Liou, Yu-Ming, and 劉育銘. "A Deep Learning Approach for Online Guitar Chord Tabs Retrieval." Thesis, 2019. http://ndltd.ncl.edu.tw/handle/a8z793.
Анотація:
碩士國立中山大學
資訊管理學系研究所
107
Chord tabs provide the information for us to play music. There are many chord tabs for guitar on the Internet. However, they differ in their format, and a song may have many chord tabs. Besides, many chord tabs are of poor quality, and they exhibit the quality discrepancy among chord sequences of a given song. Despite the fact that most websites provide the user ratings, which can be used to measure the quality of chord tabs, user ratings are rare for those unpopular songs and new songs. In this research, we proposed an approach to automatically determine the quality of chord tabs. We propose a deep learning model to learn the chord sequence similarity between a chord tab and its pertaining song, and utilize the similarity as an index to distinguish the quality of chord tabs. We utilize the similarity index to perform the relevance analysis on 1000 songs and 3510 chord tabs. In our experiment, we find that there is a positive correlation between the number of visitors for chord tabs and similarity (above 0.12) after transforming the key of each chord tab to the key of the corresponding songs. Comparing with the other methods which simply calculate chord sequence similarity using editing distance, our proposed machine learning approach performs better and add music features to measure the similarity between songs and chord tabs effectively.
27
CHUANG, YU-HAO, and 莊友豪. "Automatic Mobile Online Game Bot Detection Model Based on Deep Learning." Thesis, 2018. http://ndltd.ncl.edu.tw/handle/bxryp2.
Анотація:
碩士國立臺北大學
資訊工程學系
106
The excessive flooding of game bot causes the imbalances in mobile online games and even shortens the life cycle of mobile online games. The random forest algorithm is a general solution to identify game bot through behavioral features. Although the random forest algorithm can detect most game bot exactly, however, there are some players belonging to gray zone that cannot be detected accurately. Therefore, in this paper, we propose a deep learning based game bot detection approach, collecting players’ data and extracting the features to build the multilayer perceptron model as the detection standard. We use different methods to design four sets training parameters, and then choose the best performance training parameters as our deep learning model approach baseline. This approach is implemented on the mobile online game named KANO and the model calculates each data's probability. Then we count every probability’s number and search the data in the middle, through the algorithm to define the detecting bot critical value. The experimental result displays the proposed model has better performance, reducing the error rate from 6.218% to 2.53% and increasing the accuracy from 95.2% to 99.894% as compared with the random forest model in the same players’ data. And the training data’s critical value has very little difference with the testing data’s critical value. Thus our model can detect bot players more accurately and has lower false negative and false positive.
28
WENG, HEONG KAM, and 香靖宥. "Prediction of Online Shopping Consumers' Inter-purchase Time Using Deep Learning." Thesis, 2019. http://ndltd.ncl.edu.tw/handle/6gg4ed.
Анотація:
碩士國立臺北科技大學
經營管理系
107
In the past, it was not easy to predict consumers' behavior, there were many indicators that could not be quantified by scientific or predictive methods, such as emotions, habits, attitudes and values. The rise of big data analysis and performance of computing power, the unpredictable behavior has slowly broken through, consumers' behavior and preference has gradually predictable. There’s an important issue that merchandises want to know when the consumers are going to buy the products. Therefore, the purpose of this study is to predict the inter-purchase time of e-commerce consumers. First, cluster analysis is used to achieve consumer segmentation, who has similar behaviors would be clustered in the same group. Secondly, Recurrent Neural Network predictive model be used to predict to each group. Finally, study compares the predictive model with NeuralNet, CART, and SVM models, found that the deviation of the model is lower than others and effectively predicts the consumer's purchase interval. The application of clustering and predictive model has provided accurate reference and helps the merchandises to understand consumers' preferences, which enables effectively marketed, improve the penetration rate and reduce the cost of advertising.
29
Roy, Bhupendra. "Identifying Deception in Online Reviews: Application of Machine Learning, Deep Learning and Natural Language Processing." Master's thesis, 2020. http://hdl.handle.net/10362/101187.
Анотація:
Dissertation presented as the partial requirement for obtaining a Master's degree in Data Science and Advanced AnalyticsCustomers increasingly rate, review and research products online, (Jansen 2010). Consequently, websites containing consumer reviews are becoming targets of opinion spam. Now-a-days, people are paid money to write fake positive review online, to misguide customer and to augment sales revenue. Alternatively, people are also paid to pose as customers and to post negative fake reviews with the objective to slash competitors. These have caused menace in social media and often resulting in customer being baffled. In this study, we have explored multiple aspects of deception classification. We have explored four kinds of treatments to input i.e., the reviews using Natural Language Processing – lemmatization, stemming, POS tagging and a mix of lemmatization and POS Tagging. Also, we have explored how each of these inputs responds to different machine learning models – Logistic Regression, Naïve Bayes, Support Vector Machine, Random Forest, Extreme Gradient Boosting and Deep Learning Neural Network. We have utilized the gold standard hotel reviews dataset created by (Ott, Choi, et al. 2011) & (Ott, Cardie and Hancock, Negative Deceptive Opinion Spam 2013). Also, we used restaurant reviews dataset and doctors’ reviews dataset used by (Li, et al. 2014). We explored the usability of these models in similar domain as well as across different domains. We trained our model with 75% of hotel reviews dataset and check the accuracy of classification on similar dataset like 25% of unseen hotel reviews and on different domain dataset like unseen restaurant reviews and unseen doctors’ reviews. We perform this to create a robust model which can be applied on same domain and across different domains. Best accuracy for testing dataset of hotels achieved by us was at 91% using Deep Learning Neural Network. Logistic regression, support vector machine and random forest had similar results like neural network. Naïve Bayes also had similar accuracy; however, it had more volatility in cross domain accuracy performance. Accuracy of extreme gradient boosting was weakest among all the models that we explored. Our results are comparable and at times exceeding performance of other researchers’ work. Additionally, we have explored various models (Logistic Regression, Naïve Bayes, Support Vector Machine, Random Forest, Extreme gradient boosting, Neural network) vis a vis various input transformation method using Natural Language Processing (lemmatized unigrams, stemmed, POS tagging and a mix of lemmatization and POS Tagging).
30
Mohawesh, RIM. "Machine learning approaches for fake online reviews detection." Thesis, 2022. https://eprints.utas.edu.au/47578/.
Анотація:
Online reviews have a substantial impact on decision making in various areas of society, predominantly in the arena of buying and selling of goods. The truthfulness of online reviews is critical for both consumers and vendors. Genuine reviews can lead to satisfied customers and success for quality businesses, whereas fake reviews can mislead innocent clients, influence customers’ choices owing to false descriptions and inaccurate sales. Therefore, there is a need for efficient fake review detection models and tools that can help distinguish between fraudulent and legitimate reviews to protect the ecosystem of e-commerce sites, consumers, and companies from these misleading fake reviews.
Although several fake review detection models have been proposed in the literature, there are still some challenges regarding the performance of these models that need to be addressed. For instance, existing studies are highly dependent on linguistic features, but these are not enough to capture the semantic meaning of the reviews, necessary for improving prediction performance. Furthermore, when analysing a fake review text data stream, concept drift may occur where the input and output relationship of text changes over time, affecting model performance. The concept drift problem and its performance impact on fake review detection models has yet to be addressed. Moreover, existing models have only focused upon the review text, reviewer-based features, or both. They have also used a limited number of behavioural and content features that reduce the accuracy of the detection model. Further, these models have used fine-grained features (e.g., words) or coarse-grained features (e.g., documents, sentences, topics) separately, reducing the detection models’ performance.
To address the above discussed research gaps, this research investigates and analyses the performance of different neural network models and advanced pre-trained models for fake review detection. Then, we propose an ensemble model that combines three transformer models to detect fake reviews on semi-real-world datasets. This research also investigates the concept drift problem by detecting the occurrence of concept drift within text data streams of fake reviews, finding a correlation between concept drift (if it exists) and the performance of detection models over time in the real-world data stream. Furthermore, this research proposes a new Interpretable ensemble of multi-view deep learning model (EEMVDLM) that can detect fake reviews based on different feature perspectives and classifiers and provide interpretability of deep learning models. This ensemble model comprises three popular machine learning models: bidirectional long-short-term-memory (Bi-LSTM), convolutional neural network (CNN), and deep neural network (DNN). Additionally, this model provides interpretations to achieve reliable results from deep learning models, which are usually considered as "Black Boxes". For this purpose, the shapley additive explanations (SHAP) method and attention mechanism are used to understand the underlying logic of a model and provide other hints to determine whether it is "unfair".
In summary, this research provides the following contributions: (1) Comprehensive survey that analyses the task of fake review detection by providing the existing approaches, existing feature extraction techniques, challenges, and available datasets. (2) This research investigates and analyses the performance of different neural network models and transformers to demonstrate their effect on fake review detection. Experimental results show that the transformer models perform well with a small dataset for fake review detection. Specifically, the robustly optimised BERT pretraining approach (RoBERTa) achieves the highest accuracy. (3) This research proposes an ensemble of three transformer models to discover the hidden patterns in a sequence of fake reviews and detect them precisely. Experimental results on two semi-real datasets show that the proposed model outperforms the state-of-the-art methods. (4) This research provides an in-depth analysis for detecting concept drift within fake review data streams by using two methods: benchmarking concept drift detection methods and contentbased classification methods. The results demonstrate that there is a strong negative correlation between concept drift and the performance of fake review detection/prediction models, which indicates the difficulty of building more efficient models. (5) This research proposes a new Interpretable ensemble of multi-view deep learning model (EEMVDLM) that can detect fake reviews based on different feature perspectives and classifiers. The experimental results on two real-life datasets present excellent performance and outperformed the state-of-the-art methods. Further, the experimental results prove that our proposed model can provide reasonable interpretations that help users understand why certain reviews are classified as fake or genuine. To the best of our knowledge, this research provides the first study that incorporates advanced pre-trained models, investigates the concept drift problem, and the first Interpretable fake review detection approach. The findings of this thesis contribute to both practical and theoretical applications.
31
Chen, Syuan-Cheng, and 陳軒丞. "Deep Reinforcement Learning based Rate Adaptation for 802.11ac: A Practical Online Approach." Thesis, 2019. http://ndltd.ncl.edu.tw/handle/vm36r3.
Анотація:
碩士國立交通大學
資訊科學與工程研究所
107
As the IEEE 802.11ac becomes the mainstream Wi-Fi standard which introduces several new features, the number of available rate options increases. % due to its new channel bonding and modulation schemes. It challenges the scalability of conventional rate adaptations (RAs). It is because their designs are based on the old rate scope; moreover, many of them are incompliant to commodity Wi-Fi NICs. Our case study shows that two popular 802.11ac RAs, Minstrel-HT and Iwlwifi, fall short of expected performance in some cases due to their non-scalable designs. We thus propose a scalable, intelligent 802.11ac RA solution, called DRL-RA, which takes a deep reinforcement learning (DRL) based approach. The DRL model can guide the RA to reach the best rate by suggesting candidate rates for its probing process based on real-time channel estimation. The key insight is that the model can automatically adapt to environments, and identify a path to the best rate by learning the correlations between rate features, performance, link quality, and channel utilization rate. Its suggested rates are concentrated and precise, thereby being able to locate the best rates with low overhead. We prototype DRL-RA using the Intel NIC driver and TensorFlow with an asynchronous framework across kernel and user spaces. Our experiments show that DRL-RA outperforms the other popular RAs by up to 2.8 times.
32
Lin, Yu-Da, and 林郁達. "Online Video Synopsis: Object Detection and Management Based on Deep Learning and Minimum Collision Trajectory." Thesis, 2017. http://ndltd.ncl.edu.tw/handle/07958314078990257750.
Анотація:
碩士國立臺灣科技大學
電機工程系
105
Video synopsis is a feasible solution to expedite browsing in raw surveillance data and to perform various video analytics tasks. The technique provides a condensed video with reduced spatial or temporal redundancies, without losing the actual content of the source video. However, conventional methods are computationally intensive and time-consuming and also with blinking effect in the resultant video. To overcome these problems, we propose a trajectory-based video synopsis system which can achieve high-performance without object tracking and energy optimization for tube rearrangement. In comparison to existing methods, Spatial-temporal trajectory-based object tube extraction algorithm is performed consistently in keeping tubes continuously to avoid blinking effect. Tube rearrangement based on Minimum Collision Trajectory in spatial-temporal domain is proposed to decide the best temporal position of tubes in synopsis video. Moreover, we integrate the object detection system based on convolutional neural network (CNN) with object tubes, which enables a user quickly locating a specific object. Finally, the proposed system can efficiently generate a condensed video without blinking effect, and its robustness validated with extensive experiments.
33
Huang, Shen-Hang, and 黃慎航. "Online Structural Break Detection for Pairs Trading using Wavelet Transform and Hybrid Deep Learning Model." Thesis, 2019. http://ndltd.ncl.edu.tw/handle/5kkcxr.
Анотація:
碩士國立交通大學
資訊科學與工程研究所
108
With the mature development in the financial market, numerous people study in arbitrage strategies. Pairs trading is one of the common statistical arbitrage strategies. It first supervises two stocks that move similarly and form a stationary equilibrium with certain weights, and then makes arbitrage when the pair deviates from the stable value. The time point that the stationary relationship between two stocks does not exist any longer is called a structural break, and detecting structural breaks is important to pairs trading. There are some traditional methods for this problem, but they are not robust enough to implement in the real world. The purpose of this paper is to precisely detect structural breaks as soon as possible. Therefore, we propose a hybrid wavelet transform deep learning model using both frequency-domain and time-domain features to detect a structural break of a stock pair in Taiwan Stock Exchange Capitalization Weighted Stock Index (TAIEX). We collect the amount of half-year historical tick data for experiments and build a simulation trading system to evaluate the performances of traditional methods and our models in the real condition. The experiment results on performance metrics and simulation trading show that our proposed method successfully not only captures the abnormal signal but also reduces the loss occurred from structural breaks.
34
"Cost-Sensitive Selective Classification and its Applications to Online Fraud Management." Doctoral diss., 2019. http://hdl.handle.net/2286/R.I.53598.
Анотація:
abstract: Fraud is defined as the utilization of deception for illegal gain by hiding the true nature of the activity. While organizations lose around $3.7 trillion in revenue due to financial crimes and fraud worldwide, they can affect all levels of society significantly. In this dissertation, I focus on credit card fraud in online transactions. Every online transaction comes with a fraud risk and it is the merchant's liability to detect and stop fraudulent transactions. Merchants utilize various mechanisms to prevent and manage fraud such as automated fraud detection systems and manual transaction reviews by expert fraud analysts. Many proposed solutions mostly focus on fraud detection accuracy and ignore financial considerations. Also, the highly effective manual review process is overlooked. First, I propose Profit Optimizing Neural Risk Manager (PONRM), a selective classifier that (a) constitutes optimal collaboration between machine learning models and human expertise under industrial constraints, (b) is cost and profit sensitive. I suggest directions on how to characterize fraudulent behavior and assess the risk of a transaction. I show that my framework outperforms cost-sensitive and cost-insensitive baselines on three real-world merchant datasets. While PONRM is able to work with many supervised learners and obtain convincing results, utilizing probability outputs directly from the trained model itself can pose problems, especially in deep learning as softmax output is not a true uncertainty measure. This phenomenon, and the wide and rapid adoption of deep learning by practitioners brought unintended consequences in many situations such as in the infamous case of Google Photos' racist image recognition algorithm; thus, necessitated the utilization of the quantified uncertainty for each prediction. There have been recent efforts towards quantifying uncertainty in conventional deep learning methods (e.g., dropout as Bayesian approximation); however, their optimal use in decision making is often overlooked and understudied. Thus, I present a mixed-integer programming framework for selective classification called MIPSC, that investigates and combines model uncertainty and predictive mean to identify optimal classification and rejection regions. I also extend this framework to cost-sensitive settings (MIPCSC) and focus on the critical real-world problem, online fraud management and show that my approach outperforms industry standard methods significantly for online fraud management in real-world settings.Dissertation/Thesis
Doctoral Dissertation Computer Science 2019
35
Kumari, K., J. P. Singh, Y. K. Dwivedi, and Nripendra P. Rana. "Towards Cyberbullying-free social media in smart cities: a unified multi-modal approach." 2019. http://hdl.handle.net/10454/18116.
Анотація:
YesSmart cities are shifting the presence of people from physical world to cyber world (cyberspace). Along with the facilities for societies, the troubles of physical world, such as bullying, aggression and hate speech, are also taking their presence emphatically in cyberspace. This paper aims to dig the posts of social media to identify the bullying comments containing text as well as image. In this paper, we have proposed a unified representation of text and image together to eliminate the need for separate learning modules for image and text. A single-layer Convolutional Neural Network model is used with a unified representation. The major findings of this research are that the text represented as image is a better model to encode the information. We also found that single-layer Convolutional Neural Network is giving better results with two-dimensional representation. In the current scenario, we have used three layers of text and three layers of a colour image to represent the input that gives a recall of 74% of the bullying class with one layer of Convolutional Neural Network.
Ministry of Electronics and Information Technology (MeitY), Government of India
36
Rodrigues, Nuno Queirós. "O papel da articulação interdisciplinar na regulação do esforço de aprendizagem em ambientes online." Master's thesis, 2018. http://hdl.handle.net/1822/56032.
Анотація:
Dissertação de mestrado em Ciências da Educação (área de especialização em Tecnologia Educativa)A evolução das Tecnologias da Informação e da Comunicação tem vindo a mudar o modo e os meios como acedemos à informação e ao conhecimento. Esta circunstância favoreceu a emergência de novos paradigmas e realçou a importância da aquisição de um conjunto de capacidades, entre as quais destacamos a competência digital e aprender a aprender. Atentas a esta nova realidade, as instituições de Ensino Superior têm procurado aproximar novos públicos, através da oferta crescente de cursos de pósgraduação realizados parcial ou integralmente a distância, proporcionando uma aprendizagem verdadeiramente multidimensional e ubíqua. Neste novo paradigma tecnológico e educativo, os docentes tendem a adotar novos modelos pedagógicos facilitados pelas tecnologias digitais, propondo aos estudantes a realização de tarefas fora do contexto formal de sala de aula. Sabemos, no entanto, que estas atividades são hoje quase sempre realizadas em ambientes online, imersos nos quais os estudantes experienciam múltiplos percursos de aprendizagem, raramente lineares, que devem incluir a leitura crítica, a avaliação e a validação da credibilidade de todas as fontes consultadas. Com efeito, estas atividades exigem dos estudantes de hoje novas competências, atitudes e literacias, e de mais tempo para refletir. Porém, se a proposta de tarefas for realizada de uma forma isolada, concorrente, e não articulada pelos docentes da turma, a potencial elevada simultaneidade de atividades poderá exigir de alguns estudantes um esforço de aprendizagem excessivo, restringindo o tempo necessário para poderem refletir, aprofundar e consolidar as suas aprendizagens. Este estudo procura contribuir para a compreensão de que os estudantes constituem um recurso partilhado pelos docentes da turma, e de que neste sentido, os docentes poderão promover a regulação das suas aprendizagens se conhecerem previamente a calendarização de todas as tarefas propostas pelos seus pares. Apoiados numa metodologia de desenvolvimento com recurso a uma revisão sistemática da literatura e a entrevistas coletivas do tipo focus group com docentes e estudantes do Ensino Superior, confirmamos a relevância do problema e descrevemos uma solução capaz de proporcionar aos docentes da turma o conhecimento em tempo real da calendarização de todas as tarefas propostas. Além disso, apresentamos as perspetivas dos docentes inquiridos sobre este meio comunicante, bem como um conjunto de desafios potencialmente envolvidos na sua implementação.
The evolution of Information and Communication Technologies has been changing the way and the means of how we access information and knowledge. This circumstance favoured the emergence of new paradigms and emphasized the importance of acquiring a set of capacities, among which we highlight digital competence and learn to learn. Aware of this new reality, Higher Education institutions have been seeking to bring new audiences by increasing the number of postgraduate courses implemented partially or entirely at distance, providing a truly multidimensional and ubiquitous learning. In this new technological and educational paradigm, teachers tend to adopt new pedagogical models facilitated by digital technologies, suggesting to students the accomplishment of tasks outside the formal context of the classroom. We know, however, that these activities are nowadays mainly performed in online environments, immersed in which students experience multiple learning pathways, rarely linear, which should include critical reading, evaluation and validation of the credibility of all sources consulted. Indeed, these activities require from today’s students new skills, attitudes and literacies, and more time to reflect. However, if the proposal of tasks is performed in an isolated, competing, and not articulated way by the teachers of the class, the potential high simultaneity of activities may require from some students an excessive learning effort, restraining the time necessary to reflect, deepen, and consolidate their learning. This research aims to contribute to the understanding that students are a resource shared by the teachers of the class, and that in this sense teachers may promote the regulation of their learning if they know in advance the schedule of all tasks proposed by their peers. Supported by a development research methodology using a systematic literature review and focus group interviews with teachers and students of Higher Education, we confirm the relevance of the problem and describe a solution capable of providing class teachers the knowledge of the scheduling of all the proposed tasks. Moreover, we present the perspectives of the teachers interviewed about this communicating medium, as well as a group of challenges potentially involved in its implementation.
37
(6012219), Ayush Jain. "Using Latent Discourse Indicators to identify goodness in online conversations." Thesis, 2020.
Анотація:
In this work, we model latent discourse indicators to classify constructive and collaborative conversations online. Such conversations are considered good as they are rich in content and have a sense of direction to resolve an issue, solve a problem or gain new insights and knowledge. These unique discourse indicators are able to characterize flow of information, sentiment and community structure within discussions. We build a deep relational model that captures these complex discourse behaviors as latent variables and make a global prediction about overall conversation based on these higher level discourse behaviors. DRaiL, a Declarative Deep Relational Learning platform built on PyTorch, is used for our task in which relevant discourse behaviors are formulated as discrete latent variables and scored using a deep model. These variables capture the nuances involved in online conversations and provide the information needed for predicting the presence or absence of collaborative and constructive characterization in the entire conversational thread. We show that the joint modeling of such competing latent behaviors results in a performance improvement over the traditional direct classification methods in which all the raw features are just combined together to predict the final decision. The Yahoo News Annotated Comments Corpus is used as a dataset containing discussions on Yahoo news forums and final labels are annotated based on our precise and restricted definitions of positively labeled conversations. We formulated our annotation guidelines based on a sample set of conversations and resolved any conflict in specific annotation by revisiting those examples again.
38
Singh, Ravinder. "Extracting Human Behaviour and Personality Traits from Social Media." Thesis, 2021. https://vuir.vu.edu.au/42639/.
Анотація:
Online social media has evolved as an integral part of human society. It facilitates collaboration and information flow, and has emerged as a crucial instrument for business and government organizations alike. Online platforms are being used extensively for work, entertainment, collaboration and communication. These positive aspects are, however, overshadowed by their shortcomings. With the constant evolution and expansion of social media platforms, a significant shift has occurred in the way some humans interact with others. Online social media platforms have inadvertently emerged as networking hubs for individuals exhibiting antisocial behaviour (ASB), putting vulnerable groups of people at risk. Online ASB is one of the most common personality disorders seen on these platforms, and is challenging to address due to its complexities.
Human rights are the keystones of sturdy communities. Respect for these rights, based on the values of equality, dignity and appreciation, is vital and an integral part of strong societies. Every individual has a fundamental right to freely participate in all legal activities, including socializing in both the physical and online worlds. ASB, ranging from threatening, aggression, disregard for safety and failure to conform to lawful behaviour, deter such participation and must be dealt with accordingly. Online ASB is the manifestation of everyday sadism and violates the elementary rights (to which all individuals are entitled) of Its victims. Not only does it interfere with social participation, it also forces individuals into anxiety, depression and suicidal ideation. The consequences of online ASB for victims' and families' mental health are often far-reaching, severe and long-lasting, and can even create a social welfare burden. The behaviour can, not only inhibit constructive user participation with social media, it defies the sole purpose of these platforms: to facilitate communication and collaboration at scale. ASB needs to be detected and curtailed, encouraging fair user participation and preventing vulnerable groups of people from falling victim to such behaviour.
Considering the large variety, high contribution speed and high volume of social media data, a manual approach to detecting and classifying online ASB is not a feasible option. Furthermore, a traditional approach based on a pre-defined lexicon and rule-based feature engineering may still fall short of capturing the subtle and latent features of the diverse and enormous volume of social media data. State-of-the-art deep learning, which is a sub-field of machine learning, has produced astonishing results in numerous text classification undertakings, and has outperformed the aforementioned techniques. However, given the complexity associated with implementing deep learning algorithms and their relatively recent development, models based on the technology have significantly been under-utilized when working with online behaviour studies. Specifically, no prior study has undertaken the task of fine-grained and user- generated social media content classification related to online ASB utilizing the deep learning technology.
This thesis introduces a novel three-part framework, based on deep learning, with the objectives of: i) Detecting behaviour and personality traits from online platforms; (ii) Binary detection of online antisocial behaviour and (iii) Multiclass antisocial behaviour detection from social media corpora. A high accuracy classification model is presented proceeded by extensive experimentation with different machine learning and deep learning algorithms, fine tuning of hyper- parameters, and using different feature extraction techniques. Disparate behaviour and personality traits, including ASB and its four variants are detected with a significantly high accuracy from online social media platforms. Along the way, three medium-sized gold standard benchmark data set have been constructed. The proposed approach is seminal and offers a step towards efficient and effective methods of online ASB prevention. The approach and the findings within this thesis are significant and crucial as these lay the groundwork for detecting and eliminating all types of undesirable and unacceptable social behaviour traits from online platforms.
39
Mahale, Gopinath Vasanth. "Algorithm And Architecture Design for Real-time Face Recognition." Thesis, 2016. http://etd.iisc.ac.in/handle/2005/2743.
Анотація:
Face recognition is a field of biometrics that deals with identification of subjects based on features present in the images of their faces. The factors that make face recognition popular and favorite as compared to other biometric methods are easier operation and ability to identify subjects without their knowledge. With these features, face recognition has become an integral part of the present day security systems, targeting a smart and secure world.
There are various factors that de ne the performance of a face recognition system. The most important among them are recognition accuracy of algorithm used and time taken for recognition. Recognition accuracy of the face recognition algorithm gets affected by changes in pose, facial expression and illumination along with occlusions in the images. There have been a number of algorithms proposed to enable recognition under these ambient changes. However, it has been hard to and a single algorithm that can efficiently recognize faces in all the above mentioned conditions. Moreover, achieving real time performance for most of the complex face recognition algorithms on embedded platforms has been a challenge. Real-time performance is highly preferred in critical applications such as identification of crime suspects in public. As available software solutions for FR have significantly large latency in recognizing individuals, they are not suitable for such critical real-time applications. This thesis focuses on real-time aspect of FR, where acceleration of the algorithms is achieved by means of parallel hardware architectures.
The major contributions of this work are as follows. We target to design a face recognition system that can identify at most 30 faces in each frame of video at 15 frames per second, which amounts to 450 recognitions per second. In addition, we target to achieve good recognition accuracy along with scalability in terms of database size and input image resolutions. To design a system with these specifications, as a first step, we explore algorithms in literature and come up with a hybrid face recognition algorithm. This hybrid algorithm shows good recognition accuracy on face images with changes in illumination, pose and expressions, and also with occlusions. In addition the computations in the algorithm are modular in nature which are suitable for real-time realizations through parallel processing.
The face recognition system consists of a face detection module to detect faces in the input image, which is followed by a face recognition module to identify the detected faces. There are well established algorithms and architectures for face detection in literature which can perform detection at 15 frames per second on video frames. Detected faces of different sizes need to be scaled to the size specified by the face recognition module. To meet the real-time constraints, we propose a hardware architecture for real-time bi-cubic convolution interpolation with dynamic scaling factors. To recognize the resized faces in real-time, a scalable parallel pipelined architecture is designed for the hybrid algorithm which can perform 450 recognitions per second on a database containing grayscale images of at most 450 classes on Virtex 6 FPGA. To provide flexibility and programmability, we extend this design to REDEFINE, a multi-core massively parallel reconfigurable architecture. In this design, we come up with FR specific programmable cores termed Scalable Unit for Region Evaluation (SURE) capable of performing modular computations in the hybrid face recognition algorithm. We replicate SUREs in each tile of REDEFINE to construct a face recognition module termed REDEFINE for Face Recognition using SURE Homogeneous Cores (REFRESH).
There is a need to learn new unseen faces on-line in practical face recognition systems. Considering this, for real-time on-line learning of unseen face images, we design tiny processors termed VOP, Processor for Vector Operations. VOPs function as coprocessors to process elements under each tile of REDEFINE to accelerate micro vector operations appearing in the synaptic weight computations. We also explore deep neural networks which operate similar to the processing in human brain and capable of working on very large face databases. We explore the field of Random matrix theory to come up with a solution for synaptic weight initialization in deep neural networks for better classification . In addition, we perform design space exploration of hardware architecture for deep convolution networks and conclude with directions for future work.
40
Mahale, Gopinath Vasanth. "Algorithm And Architecture Design for Real-time Face Recognition." Thesis, 2016. http://etd.iisc.ernet.in/handle/2005/2743.
Анотація:
Face recognition is a field of biometrics that deals with identification of subjects based on features present in the images of their faces. The factors that make face recognition popular and favorite as compared to other biometric methods are easier operation and ability to identify subjects without their knowledge. With these features, face recognition has become an integral part of the present day security systems, targeting a smart and secure world.
There are various factors that de ne the performance of a face recognition system. The most important among them are recognition accuracy of algorithm used and time taken for recognition. Recognition accuracy of the face recognition algorithm gets affected by changes in pose, facial expression and illumination along with occlusions in the images. There have been a number of algorithms proposed to enable recognition under these ambient changes. However, it has been hard to and a single algorithm that can efficiently recognize faces in all the above mentioned conditions. Moreover, achieving real time performance for most of the complex face recognition algorithms on embedded platforms has been a challenge. Real-time performance is highly preferred in critical applications such as identification of crime suspects in public. As available software solutions for FR have significantly large latency in recognizing individuals, they are not suitable for such critical real-time applications. This thesis focuses on real-time aspect of FR, where acceleration of the algorithms is achieved by means of parallel hardware architectures.
The major contributions of this work are as follows. We target to design a face recognition system that can identify at most 30 faces in each frame of video at 15 frames per second, which amounts to 450 recognitions per second. In addition, we target to achieve good recognition accuracy along with scalability in terms of database size and input image resolutions. To design a system with these specifications, as a first step, we explore algorithms in literature and come up with a hybrid face recognition algorithm. This hybrid algorithm shows good recognition accuracy on face images with changes in illumination, pose and expressions, and also with occlusions. In addition the computations in the algorithm are modular in nature which are suitable for real-time realizations through parallel processing.
The face recognition system consists of a face detection module to detect faces in the input image, which is followed by a face recognition module to identify the detected faces. There are well established algorithms and architectures for face detection in literature which can perform detection at 15 frames per second on video frames. Detected faces of different sizes need to be scaled to the size specified by the face recognition module. To meet the real-time constraints, we propose a hardware architecture for real-time bi-cubic convolution interpolation with dynamic scaling factors. To recognize the resized faces in real-time, a scalable parallel pipelined architecture is designed for the hybrid algorithm which can perform 450 recognitions per second on a database containing grayscale images of at most 450 classes on Virtex 6 FPGA. To provide flexibility and programmability, we extend this design to REDEFINE, a multi-core massively parallel reconfigurable architecture. In this design, we come up with FR specific programmable cores termed Scalable Unit for Region Evaluation (SURE) capable of performing modular computations in the hybrid face recognition algorithm. We replicate SUREs in each tile of REDEFINE to construct a face recognition module termed REDEFINE for Face Recognition using SURE Homogeneous Cores (REFRESH).
There is a need to learn new unseen faces on-line in practical face recognition systems. Considering this, for real-time on-line learning of unseen face images, we design tiny processors termed VOP, Processor for Vector Operations. VOPs function as coprocessors to process elements under each tile of REDEFINE to accelerate micro vector operations appearing in the synaptic weight computations. We also explore deep neural networks which operate similar to the processing in human brain and capable of working on very large face databases. We explore the field of Random matrix theory to come up with a solution for synaptic weight initialization in deep neural networks for better classification . In addition, we perform design space exploration of hardware architecture for deep convolution networks and conclude with directions for future work.