In this project we propose a deep learning architecture to predict dementia, a disease which affects around 55 million people all over the world and makes them in some cases dependent people. The main aim is to predict the disease in the early stages, in order to start having professional treatment from the beginning, which can improve the quality of life of the patients. Another aim is to analyze how the combination of different modalities, like audio or text, can influence the results obtained by the model. Many research has been done over the different available dementia datasets as well as the classification tasks with audio and text data. To this end, we have used the DementiaBank dataset, which includes audio recordings as well as their transcriptions of healthy people and people with dementia. Different models have been used and tested, including Convolutional Neural Networks for the audio classification, Transformers for the text classification and a combination of both models in a multimodal one. These models have been tested over a test set, obtaining the best results from the text modality, achieving a 90.36% of accuracy on the detection of dementia task.

Certain fields present significant challenges when attempting to train complex Deep Learning architectures, particularly when the available datasets are limited and imbalanced. Real-time object detection in maritime environments using aerial images is a notable example. Although SeaDronesSee is the most extensive and complete dataset for this task, it suffers from significant class imbalance. To address this issue, we present POSEIDON, a data augmentation tool specifically designed for object detection datasets. Our approach generates new training samples by combining objects and samples from the original training set while utilizing the image metadata to make informed decisions. We evaluate our method using YOLOv5 and YOLOv8 and demonstrate its superiority over other balancing techniques, such as error weighting, by an overall improvement of 2.33% and 4.6%, respectively.

No abstract

No abstract