Browsing by Subject "deep learning"

Now showing 1 - 20 of 37

Access status: Open Access ,
A brief review of recent developments in the integration of deep learning with GIS
(Wydawnictwa AGH, 2022) Mohan, Shyama; Giridhar, M.V.S.S.
The interaction of Deep Learning (DL) methods with Geographical Information System (GIS) provides the opportunity to obtain new insights into environmental processes through the spatial, temporal and spectral resolutions as well as data integration. The two technologies may be connected to form a dynamic system that is incredibly well adapted to the evaluation of environmental conditions through the interrelationships of texture, size, pattern, and process. This perspective has acquired popularity in multiple disciplines. GIS is significantly dependant on processors, particularly for 3D calculations, map rendering, and route calculation whereas DL can process huge amounts of data. DL has received a lot of attention recently as a technology with a plethora of promising results. Furthermore, the growing use of DL methods in a variety of disciplines, including GIS, is evident. This study tries to provide a brief overview of the use of DL methods in GIS. This paper introduces the essential DL concepts relevant to GIS, the majority of which have been published in recent years. This research explores remote sensing applications and technologies in areas such as mapping, hydrological modelling, disaster management, and transportation route planning. Finally, conclusions on contemporary framework methodologies and suggestions for further studies are provided.
Access status: Open Access ,
An evaluation of the capabilities of image-based metal component defect recognition with deep learning techniques
(Wydawnictwa AGH, 2024) Wójcik, Michał P.; Pawlikowski, Kacper; Madej, Łukasz
In the era of Industry 4.0, deploying highly specialised machine learning models trained on unique and often scarce datasets is an attractive solution for advancing automated quality control and minimising production costs. Therefore, the main aim of this research is to evaluate the capabilities of three deep learning models (ResNet-18, ResNet-50 and SE-ResNeXt-101 (32 × 4d)) in the identification of surface defects in forged products. Leveraging advanced photography techniques, including studio lighting and a shadowless box, high-quality images of complex product surfaces were acquired for the training data set. Given the relatively small size of the image dataset, aggressive data augmentation techniques were introduced during the training and evaluation process to ensure robust model generalisation ability. The results obtained demonstrate the significant impact of data augmentation on model performance, highlighting its importance in training and evaluating deep learning models with limited data. This research also emphasises the need for innovative data pre-processing strategies in an efficient and robust machine learning model delivery to the industrial environment.
Access status: Restricted ,
Analiza opinii przy użyciu uczenia głębokiego
(Data obrony: 2018-01-22) Flis, Marcin
Wydział Elektrotechniki, Automatyki, Informatyki i Inżynierii Biomedycznej
Access status: Restricted ,
Application of deep learning approach for identification and classification of scale defects during hot forming process
(Data obrony: 2019-09-19) Furman, Szymon
Wydział Inżynierii Metali i Informatyki Przemysłowej
Access status: Open Access ,
Choroba Alzheimera jako przykład desynchronizacji funkcjonowania i zbiór neurokognitywnych wzorców stanowiących potencjalne źródło zasobów dla rozwoju sztucznej inteligencji
(Wydawnictwa AGH, 2022) Kaszyńska, Anna Aleksandra
The review article focuses on the potential development of Artificial Intelligence by extracting fixed patterns and regularities that enable the improvement and refinement of advanced analyses in the field of artificial neural network learning. Is conducted through the prism of the neurocognitive view of Alzheimer’s disease as a potential set of neurocognitive patterns constituting a potential source of resources for the development of artificial intelligence. It is closely related to encephalography, both used to detect pathological dementia changes, and the analysis of brain activity itself, showing the existence of repeated regularities. These patterns, analogic in the astrophysical Lagrandrean mapping analysis of the galaxy, seem to have the potential to develop Artificial Intelligence. Especially, following the idea of perceiving Alzheimer’s disease as a global functional desynchronisation, global neurodegenerative changes may provide potential resources that, through mathematical and algebraic transformations, to serve as a foundation for the development of Artificial Intelligence.
Access status: Restricted ,
Deep learning : praca z językiem R i biblioteką Keras
(Helion, ©© 2019) Chollet, François; Allaire, J. J.
Access status: Restricted ,
Deep learning : praktyczne wprowadzenie
(Wydawnictwo Helion, ©© 2018) Patterson, Josh; Gibson, Adam
Access status: Restricted ,
Deep learning : systemy uczące się
(Wydawnictwo Naukowe PWN, 2018) Goodfellow, Ian; Bengio, Yoshua; Courville, Aaron
Access status: Restricted ,
Deep learning dla programistów : budowanie aplikacji AI za pomocą fastai i PyTorch
(Helion, ©© 2021) Howard, Jeremy; Gugger, Sylvain
Access status: Open Access ,
Detection of credit card fraud with optimized deep neural network in balanced data condition
(Wydawnictwa AGH, 2024) Shome, Nirupam; Sarkar, Devran Dey; Kashyap, Richik; Laskar, Rabul Hussain
Due to the huge number of financial transactions, it is almost impossible for humans to manually detect fraudulent transactions. In previous work, the datasets are not balanced and the models suffer from overfitting problems. In this paper, we tried to overcome the problems by tuning hyperparameters and balancing the dataset with a hybrid approach using under-sampling and oversampling techniques. In this study, we have observed that these modifications are effective in getting better performance in comparison to the existing models. The MCC score is considered an important parameter in binary classification since it ensures the correct prediction of the majority of positive data instances and negative data instances. So, we emphasize on MCC score and our method achieved an MCC score of 97.09%, which is far more (16 % approx.) than other state-of-the-art methods. In terms of other performance metrics, the result of our proposed model has also improved significantly.
Access status: Restricted ,
Efficient storage of training data for DNN (pytorch-ddnevo) evolution platform
(Data obrony: 2020-09-29) Szeremeta, Grzegorz
Wydział Informatyki, Elektroniki i Telekomunikacji
Access status: Open Access ,
Evolutionary data driven modelling and many objective optimization of non linear noisy data in the blast furnace iron making process
(Wydawnictwa AGH, 2021) Mahanta, Bashista Kumar; Chakraborti, Nirupam
The optimization of process parameters in modern blast furnace operation, where both control and accessing large data set with multiple variables and objectives is a challenging task. To handle such non-linear and noisy data set deep learning techniques have been used in recent time. In this study an evolutionary deep neural network algorithm (EvoDN2) has been applied to derive a data driven model for blast furnace. The optimal front generated from deep neural network is compared against the optimal models developed from bi-objective genetic programming algorithm (BioGP) and evolutionary neural network (EvoNN). The optimization process is applied to all the training models by using constraint based reference vector evolutionary algorithm (cRVEA).
Access status: Open Access ,
Exploring convolutional auto-encoders for representation learning on networks
(Wydawnictwa AGH, 2019) Nerurkar, Pranav Ajeet; Chandane, Madhav; Bhirud, Sunil
A multitude of important real-world or synthetic systems possess network structures. Extending learning techniques such as neural networks to process such non-Euclidean data is therefore an important direction for machine learning re- search. However, this domain has received comparatively low levels of attention until very recently. There is no straight-forward application of machine learning to network data, as machine learning tools are designed for i:i:d data, simple Euclidean data, or grids. To address this challenge, the technical focus of this dissertation is on the use of graph neural networks for network representation learning (NRL), i.e., learning the vector representations of nodes in networks. Learning the vector embeddings of graph-structured data is similar to embedding complex data into low-dimensional geometries. After the embedding process is completed, the drawbacks associated with graph-structured data are overcome. The current inquiry proposes two deep-learning auto-encoder-based approaches for generating node embeddings. The drawbacks in such existing auto-encoder approaches as shallow architectures and excessive parameters are tackled in the proposed architectures by using fully convolutional layers. Extensive experiments are performed on publicly available benchmark network datasets to highlight the validity of this approach.
Access status: Open Access ,
Eye disease segmentation using hybrid neural encoder decoder based Unet hybrid inception
(Wydawnictwa AGH, 2024) Bali, Akanksha; Singh, Kuljeet; Mansotra, Vibhakar
Diabetic retinopathy (DR) is one of the major causes of vision problems worldwide. With proper treatment, early diagnosis of DR can prevent the progression of the disease. In this paper, we present a combinative method using U-Net with a modified Inception architecture for the diagnosis of both the diseases. The proposed method is based on deep neural architecture formalising encoder decoder modelling with convolutional architectures namely Inception and Residual Connection. The performance of the proposed model was validated on the IDRid 2019 contest dataset. Experiments demonstrate that the modified Inception deep feature extractor improves DR classification with a classification accuracy of 99.34% in IDRid across classes with comparison to Resnet. The paper Benchmark tests the dataset with proposed model of Hybrid Dense-ED-UHI: Encoder Decoder based U-Net Hybrid Inception model with 15 fold cross validation. The paper in details discusses the various metrics of the proposed model with various visualisation and multifield validations.
Access status: Restricted ,
Głębokie uczenie z TensorFlow : od regresji liniowej po uczenie przez wzmacnianie
(Helion, ©© 2020) Ramsundar, Bharath; Bosagh Zadeh, Reza
Access status: Restricted ,
Implementacja i ewaluacja algorytmów deep reinforcement learning
(Data obrony: 2017-01-19) Opoka, Grzegorz
Wydział Elektrotechniki, Automatyki, Informatyki i Inżynierii Biomedycznej
Access status: Open Access ,
Learning-free deep features for multispectral palm-print classification
(Wydawnictwa AGH, 2023) Aounallah, Asma; Meraoumia, Abdallah; Bendjenna, Hakim
The feature-extraction step is a major and crucial step in analyzing and understanding raw data, as it has a considerable impact on system accuracy. Despite the very acceptable results that have been obtained by many handcrafted methods, these can unfortunately have difficulty representing features in the cases of large databases or with strongly correlated samples. In this context, we attempt to examine the discriminability of texture features by proposing a novel, simple, and lightweight method for deep feature extraction to characterize the discriminative power of different textures. We evaluated the performance of our method by using a palm print-based biometric system, and the experimental results (using the CASIA multispectral palm--print database) demonstrate the superiority of the proposed method over the latest handcrafted and deep methods.
Access status: Restricted ,
Metody inteligencji komputerowej w analizie danych geofizycznych
(Data obrony: 2018-01-29) Stachura, Gabriel
Wydział Geologii, Geofizyki i Ochrony Środowiska
W pracy przedstawiono model statystyczny ośrodka geologicznego, na którym dokonano klasyfikacji litologicznej na podstawie danych profilowań otworowych. Model utworzono z wykorzystaniem metod sztucznej inteligencji komputerowej. Przeanalizowano 3 odmienne typy sieci neuronowych: klasyczne (ANN), sieci wektorów nośnych (SVM) oraz głębokiego uczenia. Głębokie uczenie jest najbardziej nowatorską metodą maszynowego uczenia, użyteczną zwłaszcza w przypadku pracy z dużymi bazami danych. Metody testowano w dwóch odmiennych środowiskach obliczeniowych – pakiecie R oraz środowisku STATISTICA. Uzyskane rezultaty na poziomie ponad 80% poprawności klasyfikacji pozwalają stwierdzić, iż opracowany model z dość dobrą precyzją oddaje rzeczywisty stan górotworu. Ze względu na dynamiczny rozwój tego rodzaju metod eksploracji danych, należy spodziewać się ich większej dokładności oraz szerszego zastosowania.
Access status: Open Access ,
Pre-trained Deep Neural Network using Sparse Autoencoders and Scattering Wavelet Transform for musical genre recognition
(Wydawnictwa AGH, 2015) Kleć, Mariusz; Koržinek, Danijel
Research described in this paper tries to combine the approach of Deep Neural Networks (DNN) with the novel audio features extracted using the Scattering Wavelet Transform (SWT) for classifying musical genres. The SWT uses a sequence of Wavelet Transforms to compute the modulation spectrum coefficients of multiple orders, which has already shown to be promising for this task. The DNN in this work uses pre-trained layers using Sparse Autoencoders (SAE). Data obtained from the Creative Commons website jamendo.com is used to boost the well-known GTZAN database, which is a standard benchmark for this task. The final classifier is tested using a 10-fold cross validation to achieve results similar to other state-of-the-art approaches.
Access status: Restricted ,
Próba zastosowania systemów głębokiego uczenia do rozpoznawania emocji z obrazów twarzy
(Data obrony: 2018-01-19) Mróz, Paweł
Wydział Elektrotechniki, Automatyki, Informatyki i Inżynierii Biomedycznej