Conversing Learning: Actively looking for human assistance to improve Machine Learning tasks

Saulo Domingos de Souza Pedro, Estevam Hruschka

Jun 29, 2013 (modified: Jun 29, 2013) AKBC 2013 submission readers: everyone
  • Decision: reject
  • Abstract: In Machine Learning (ML) systems we often apply techniques to learn about real world problems behavior from data. Traditionally, such data comes from instances of datasets (that represents the target problem) which we want to learn from. This approach has been broadly used in many different application domains such as recommending systems, meteorological prediction, medical diagnosis, etc. The recent years of quick development of communications technology, that made the Internet faster and available, made possible the acquisition of information to feed a growing number of Machine Learning applications and, in addition, brought light to the use of human computation and crowdsourcing approaches commonly applied to problems that are easy for human but difficult for computers. Thus, the Social Web has been the focus of many research in Artificial Intelligence (AI) and Machine Learning. In this work we want to show how we can take advantage from the Social Web to add value to Machine Learning systems which can actively and autonomously ask for web users help to improve learning performance.