2021/08/12 by Thorben Werner, Werner, Thorben
Computer Science · Decision Sciences · #Machine Learning and Algorithms #Advanced Bandit Algorithms Research #Data Stream Mining Techniques
paper · pdf · doi:10.48550/arxiv.2108.05595
Machine Learning requires large amounts of labeled data to fit a model. Many\ndatasets are already publicly available, nevertheless forcing application\npossibilities of machine learning to the domains of those public datasets. The\never-growing penetration of machine learning algorithms in new application\nareas requires solutions for the need for data in those new domains. This\nthesis works on active learning as one possible solution to reduce the amount\nof data that needs to be processed by hand, by processing only those datapoints\nthat specifically benefit the training of a strong model for the task. A newly\nproposed framework for framing the active learning workflow as a reinforcement\nlearning problem is adapted for image classification and a series of three\nexperiments is conducted. Each experiment is evaluated and potential issues\nwith the approach are outlined. Each following experiment then proposes\nimprovements to the framework and evaluates their impact. After the last\nexperiment, a final conclusion is drawn, unfortunately rejecting this work's\nhypothesis and outlining that the proposed framework at the moment is not\ncapable of improving active learning for image classification with a trained\nreinforcement learning agent.\n