DrCaptcha: An Interactive Machine Learning Application

Rafael Glikis ; Christos Makris ; Nikos Tsirakis

Computer Science and Information Systems, Tome 18 (2021) no. 3

Résumé

The creation of a Machine Learning system is a typical process that is mostly automated. However, we may address some problems in the during development, such as the over-training on the training set. A technique for eliminating this phenomenon is the assembling of ensembles of models that cooperate to make predictions. Another problem that almost always occurs is the necessity of the human factor in the data preparation process. In this paper, we present DrCaptcha [15], an interactive machine learning system that provides third-party applications with a CAPTCHA service and, at the same time, uses the user's input to train artificial neural networks that can be combined to create a powerful OCR system. A different way to tackle this problem is to use transfer learning, as we did in one of our experiments [33], to retrain models trained on massive datasets and retrain them in a smaller dataset.

Détail

Keywords: machine learning, neural networks, transfer learning, interactive machine learning, ensemble techniques