Predicting facial beauty with convolutional neural networks
Πρόβλεψη ελκυστικότητας προσώπου με χρήση συνελικτικών νευρωνικών δικτύων
As the science of Machine Learning evolves, the need to develop models that mimic the human crisis increases. One area on which many researchers focus is the analysis of optical signals with the aim of drawing conclusions in accordance with human logic, with the most important tool for this analysis being Convolutional Neural Networks. The subject of this dissertation is the prediction of facial beauty using Convolutional Neural Networks and the impact that discriminatory training data may have on the extracted predictions. For this purpose, multiple models were developed with differences in the training data and the results were analyzed based on the introduced bias in each case. According to the results of the study, it becomes clear that model training using biased data can yield unreliable results and lead to erroneous conclusions.