Search results for "Deep Learning"
showing 10 items of 337 documents
Training Deep Neural Networks with Novel Metaheuristic Algorithms for Fatigue Crack Growth Prediction in Aluminum Aircraft Alloys
2022
Fatigue cracks are a major defect in metal alloys, and specifically, their study poses defect evaluation challenges in aluminum aircraft alloys. Existing inline inspection tools exhibit measurement uncertainties. The physical-based methods for crack growth prediction utilize stress analysis models and the crack growth model governed by Paris’ law. These models, when utilized for long-term crack growth prediction, yield sub-optimum solutions and pose several technical limitations to the prediction problems. The metaheuristic optimization algorithms in this study have been conducted in accordance with neural networks to accurately forecast the crack growth rates in aluminum alloys. Through ex…
A Cost-Effective 3D Acquisition and Visualization Framework for Cultural Heritage
2020
Museums and cultural institutions, in general, are in a constant challenge of adding more value to their collections. The attractiveness of assets is practically tightly related to their value obeying the offer and demand law. New digital visualization technologies are found to give more excitements, especially to the younger generation as it is proven by multiple studies. Nowadays, museums around the world are currently trying to promote their collections through new multimedia and digital technologies such as 3D modeling, virtual reality (VR), augmented reality (AR), and serious games. However, the difficulty and the resources required to implement such technologies present a real challen…
Convolutional architectures for virtual screening
2020
Abstract Background A Virtual Screening algorithm has to adapt to the different stages of this process. Early screening needs to ensure that all bioactive compounds are ranked in the first positions despite of the number of false positives, while a second screening round is aimed at increasing the prediction accuracy. Results A novel CNN architecture is presented to this aim, which predicts bioactivity of candidate compounds on CDK1 using a combination of molecular fingerprints as their vector representation, and has been trained suitably to achieve good results as regards both enrichment factor and accuracy in different screening modes (98.55% accuracy in active-only selection, and 98.88% …
Impact of textual data augmentation on linguistic pattern extraction to improve the idiomaticity of extractive summaries
2021
International audience; The present work aims to develop a text summarisation system for financial texts with a focus on the fluidity of the target language. Linguistic analysis shows that the process of writing summaries should take into account not only terminological and collocational extraction, but also a range of linguistic material referred to here as the "support lexicon", that plays an important role in the cognitive organisation of the field. On this basis, this paper highlights the relevance of pre-training the CamemBERT model on a French financial dataset to extend its domainspecific vocabulary and fine-tuning it on extractive summarisation. We then evaluate the impact of textua…
Classification of diabetes-related retinal diseases using a deep learning approach in optical coherence tomography
2019
Background and objectives: Spectral Domain Optical Coherence Tomography (SD-OCT) is a volumetric imaging technique that allows measuring patterns between layers such as small amounts of fluid. Since 2012, automatic medical image analysis performance has steadily increased through the use of deep learning models that automatically learn relevant features for specific tasks, instead of designing visual features manually. Nevertheless, providing insights and interpretation of the predictions made by the model is still a challenge. This paper describes a deep learning model able to detect medically interpretable information in relevant images from a volume to classify diabetes-related retinal d…
Digital Heritage Enrichment through Artificial Intelligence and SemanticWeb Technologies
2019
Art and culture represent substantial ways to transfer the history of humans across civilizations and epochs. Preserving artwork and cultural objects is thus important and the focus of multiple institutions and governments around the world. Digital preservation in cultural heritage represents a cost-effective and reliable long-term preservation and several challenges related to its effectiveness and its reliability have arisen such as metadata enrichment, digital curation, link discoveries, etc. Through this paper, we discuss these challenges and present innovative ways that leverage recent endeavors in artificial intelligence and semantic web technologies to enrich cultural data. Our contr…
Leveraging Uncertainty Estimates to Improve Segmentation Performance in Cardiac MR
2021
International audience; In medical image segmentation, several studies have used Bayesian neural networks to segment and quantify the uncertainty of the images. These studies show that there might be an increased epistemic uncertainty in areas where there are semantically and visually challenging pixels. The uncertain areas of the image can be of a great interest as they can possibly indicate the regions of incorrect segmentation. To leverage the uncertainty information, we propose a segmentation model that incorporates the uncertainty into its learning process. Firstly, we generate the uncertainty estimate (sample variance) using Monte-Carlo dropout during training. Then we incorporate it …
Système de sécurité biométrique multimodal par imagerie, dédié au contrôle d’accès
2019
Research of this thesis consists in setting up efficient and light solutions to answer the problems of securing sensitive products. Motivated by a collaboration with various stakeholders within the Nuc-Track project, the development of a biometric security system, possibly multimodal, will lead to a study on various biometric features such as the face, fingerprints and the vascular network. This thesis will focus on an algorithm and architecture matching, with the aim of minimizing the storage size of the learning models while guaranteeing optimal performances. This will allow it to be stored on a personal support, thus respecting privacy standards.
Application of LSTM architectures for next frame forecasting in Sentinel-1 images time series
2020
L'analyse prédictive permet d'estimer les tendances des évènements futurs. De nos jours, les algorithmes Deep Learning permettent de faire de bonnes prédictions. Cependant, pour chaque type de problème donné, il est nécessaire de choisir l'architecture optimale. Dans cet article, les modèles Stack-LSTM, CNN-LSTM et ConvLSTM sont appliqués à une série temporelle d'images radar sentinel-1, le but étant de prédire la prochaine occurrence dans une séquence. Les résultats expérimentaux évalués à l'aide des indicateurs de performance tels que le RMSE et le MAE, le temps de traitement et l'index de similarité SSIM, montrent que chacune des trois architectures peut produire de bons résultats en fon…
Using Polynomial Loss and Uncertainty Information for Robust Left Atrial and Scar Quantification and Segmentation
2022
Automatic and accurate segmentation of the left atrial (LA) cavity and scar can be helpful for the diagnosis and prognosis of patients with atrial fibrillation. However, automating the segmentation can be difficult due to the poor image quality, variable LA shapes, and small discrete regions of LA scars. In this paper, we proposed a fully-automatic method to segment LA cavity and scar from Late Gadolinium Enhancement (LGE) MRIs. For the loss functions, we propose two different losses for each task. To enhance the segmentation of LA cavity from the multicenter dataset, we present a hybrid loss that leverages Dice loss with a polynomial version of cross-entropy loss (PolyCE). We also utilize …