|
|
Low-Resource Neural Machine Translation
Filo, Denis ; Fajčík, Martin (referee) ; Jon, Josef (advisor)
This thesis deals with neural machine translation (NMT) for low-resource languages. The goal was to evaluate current techniques by using the experiments and suggest their improvements. The translation systems in this thesis used the neural network transformer architecture and were trained by the Marian framework. The selected language pairs were Slovak with Croatian and Slovak with Serbian. The subjects of the experiments were the transfer learning techniques and semi-supervised learning.
|
|
|
Data Mining Case Study in Python
Stoika, Anastasiia ; Burgetová, Ivana (referee) ; Zendulka, Jaroslav (advisor)
This thesis focuses on basic concepts and techniques of the process known as knowledge discovery from data. The goal is to demonstrate available resources in Python, which enable to perform the steps of this process. The thesis addresses several methods and techniques focused on detection of unusual observations, based on clustering and classification. It discusses data mining task for data with the limited amount of inspection resources. This inspection activity should be used to detect unusual transactions of sales of some company that may indicate fraud attempts by some of its salespeople.
|
|
|
Computer Vision with Active Learning
Kolář, Martin ; Machová, Kristína (referee) ; Arridge, Simon (referee) ; Zemčík, Pavel (advisor)
Metody strojového vidění se zdokonalují zlepšením modelů, laděním trénovaných parametrů nebo anotací reprezentativních dat. Tato práce řadou experimentů potvrzuje hypotézu že aktivní učení zvyšuje přesnost těchto modelů. Rozšířením přistupu pseudolabelů o aktivní učení přispívá tato práce přístupem "one-shot-learning" k učení nových kategorií obrazů s použitím algoritmických doporučení, dále online grafickým uživatelským rozhraním pro optimalizaci dilema Exploration/Exploitation pro online tagování, a dvoukrokovým offline binárním přístupem aktivního učení pro zlepšení kvality dat používaných pro snímání fontů. Tím, že demonstruje přínos aktivního učení v těchto přístupech, přispívá tato práce k hypotéze i konkrétním aplikacím strojového vidění.
|
|
|
Computer Vision with Active Learning
Kolář, Martin ; Machová, Kristína (referee) ; Arridge, Simon (referee) ; Zemčík, Pavel (advisor)
Metody strojového vidění se zdokonalují zlepšením modelů, laděním trénovaných parametrů nebo anotací reprezentativních dat. Tato práce řadou experimentů potvrzuje hypotézu že aktivní učení zvyšuje přesnost těchto modelů. Rozšířením přistupu pseudolabelů o aktivní učení přispívá tato práce přístupem "one-shot-learning" k učení nových kategorií obrazů s použitím algoritmických doporučení, dále online grafickým uživatelským rozhraním pro optimalizaci dilema Exploration/Exploitation pro online tagování, a dvoukrokovým offline binárním přístupem aktivního učení pro zlepšení kvality dat používaných pro snímání fontů. Tím, že demonstruje přínos aktivního učení v těchto přístupech, přispívá tato práce k hypotéze i konkrétním aplikacím strojového vidění.
|
|
|
Semi-supervised deep learning in sequence labeling
Páll, Juraj Eduard ; Šabata, Tomáš (advisor) ; Flusser, Martin (referee)
Sequence labeling is a type of machine learning problem that involves as- signing a label to each sequence member. Deep learning has shown good per- formance for this problem. However, one disadvantage of this approach is its requirement of having a large amount of labeled data. Semi-supervised learning mitigates this problem by using cheaper unlabeled data together with labeled data. Currently, usage of semi-supervised deep learning for sequence labeling is limited. Therefore, the focus of this thesis is on the application of semi-super- vised deep learning in sequence labeling. Existing semi-supervised deep learning approaches are examined, and approaches for sequence labeling are proposed. The proposed approaches were implemented and experimentally evaluated on named-entity recognition and part-of-speech tagging tasks.
|
|
|
Semi-Supervised Approach To Train Captcha Letter Position Detetor
Bostik, Ondrej
Common Optical Character Recognition (OCR) methods benefit from the fact, that the text is distributedin images in a predictable pattern. This is not the situation with CAPTCHA systems. UtilizingOCR algorithms to overcome common web anti-abuse CAPTCHA systems is therefore a challengingtask. To train a system to overcome any CAPTCHA scheme, an attacker needs a huge dataset ofannotated images. And for some methods, the attacker needs not only the right answers but also anexact position of the character in the CAPTCHA image.Annotate the positions of the object in an image is a time-consuming task. In this paper, we proposea system, which can help to annotate the position of CAPTCHA character with minimal humaninteraction. After annotating a small sample of targeted CAPTCHA images, a YOLO-based regiondetection deep network is used to search for the characters’ locations.
|
|
|
Semi-Supervised Deep Learning Approach For Breaking Geocaching Captchas
Bostik, Ondrej
For nearly two decades, a substantial part of developed anti-abuse and anti-spam systems for web applications called CAPTCHA is based on imperfections in OCR (Optical Character Recognition) algorithms. But with improvements in Deep Learning in OCR, these systems are now obsolete. More and more systems can now break various text Captchas with great accuracy. Now with sufficient training dataset, almost every text-based Captcha scheme can be broken. The focus of this work is to present an idea of a semi-supervised method for reading text-based Captcha which needs only a small initial dataset. The main part of this article is dealing with the problem of training a deep learning system with only a small sample of target Captcha scheme via transfer learning.
|
|
|
Low-Resource Neural Machine Translation
Filo, Denis ; Fajčík, Martin (referee) ; Jon, Josef (advisor)
This thesis deals with neural machine translation (NMT) for low-resource languages. The goal was to evaluate current techniques by using the experiments and suggest their improvements. The translation systems in this thesis used the neural network transformer architecture and were trained by the Marian framework. The selected language pairs were Slovak with Croatian and Slovak with Serbian. The subjects of the experiments were the transfer learning techniques and semi-supervised learning.
|
|
|
Semi-supervised Learning from Unfavorably Distributed Data
Sochor, Matěj ; Pilát, Martin (advisor) ; Mrázová, Iveta (referee)
Semi-supervised learning (SSL) is a branch of machine learning focusing on using not only labeled data samples, but also unlabeled ones, in an effort to decrease the need for labeled data and thus allow using machine learning even when labeling large amounts of data would be too costly. Despite its quick development in the recent years, there are still issues left to be solved before it can be broadly deployed in practice. One of those issues is class distribution mismatch. It arises when the unlabeled data contains samples not belonging to the classes present in the labeled data. This confuses the training and can even lead to getting a classifier performing worse than a classifier trained on the available data in purely supervised fashion. We designed a filtration method called Unfavorable Data Filtering (UDF) which extracts important features from the data and then uses a similarity-based filter to filter the irrelevant data out according to those features. The filtering happens before any of the SSL training takes places, making UDF usable with any SSL algorithm. To judge its effectiveness, we performed many experiments, mainly on the CIFAR-10 dataset. We found out that UDF is capable of significantly improving the resulting accuracy when compared to not filtering the data, identified basic guidelines...
|
|
|
Semi-supervised deep learning in sequence labeling
Páll, Juraj Eduard ; Šabata, Tomáš (advisor) ; Flusser, Martin (referee)
Sequence labeling is a type of machine learning problem that involves as- signing a label to each sequence member. Deep learning has shown good per- formance for this problem. However, one disadvantage of this approach is its requirement of having a large amount of labeled data. Semi-supervised learning mitigates this problem by using cheaper unlabeled data together with labeled data. Currently, usage of semi-supervised deep learning for sequence labeling is limited. Therefore, the focus of this thesis is on the application of semi-super- vised deep learning in sequence labeling. Existing semi-supervised deep learning approaches are examined, and approaches for sequence labeling are proposed. The proposed approaches were implemented and experimentally evaluated on named-entity recognition and part-of-speech tagging tasks.
|
guest ::
