Bulletin de philosophie ancienne

Boosting Tree-Assisted Multitask Deep Learning for Small Scientific Datasets

“…The multitask learning technique learns multiple tasks simultaneously with the aim of mutual benefit and has been widely applied in various implementations, including image processing, 41 speech recognition and classification, 42,43 identification of handwritten digits, 44 natural language processing, 45 computer vision, 46 microarray data integration, 47 and drug discovery. [48][49][50][51] Most formulations of multitask learning extract hand-crafted features during the learning process and are based on the assumption that there is a linear relationship between the data and target labels.…”