Computer Vision News - July 2016

Every month, Computer Vision News reviews a challenge related to our field, be it in medical imaging, automotive, robotics or else. If you don’t find time to read challenges, but are interested in the new methods proposed by the scientific community to solve them, you can read our challenge summaries! This month we have chosen to review the ChaLearn Automatic Machine Learning Challenge (AutoML) held by CodaLab , an open-source platform that provides worksheets for conducting computational research and challenges to tackle stimulating computational problems. The website of the challenge is here . The winners have released their codes here . Background A recent KDNuggets poll (KDN, 2015) predicts that within ten years most expert-level predictive analytics/data science tasks will be automated. Domain-specific automated software has been in existence for a while, and the AutoML challenge contributes to the transition towards a fully automated learning suite by testing machine learning code operated without any human intervention. It will push the state-of-the-art in fully automatic machine learning on a wide range of problems taken from real world applications, by solving classification and regression problems from given feature representations, in a fully automated way. The considerable success achieved by machine learning often relies on the intervention of human experts, who need to provide appropriate input in order to allow the system to operate. However, there is an acute need for data scientists who have the ability to transform raw data into useful information: in recent years, this ability has been a key factor in the success of Google’s machine-learning- related services, as admitted by Google’s Chief Scientist Peter Norvig in 2011: “ We don’t have better algorithms. We just have more data ”. Hence, methods that take the human expert out of the loop are called for and AutoML is designed to solve this problem in all aspects of automating the machine learning process, with the notable novelty of the “code submission”: code submitted by the participants is executed automatically on Codalab , allowing to train and test learning machines with datasets unknown to the participants, granting identical conditions to all submissions. 36 Computer Vision News Challenge ChaLearn Automatic Machine Learning “ Within ten years, most expert-level predictive analytics/data science tasks will be automated ” KDN, 2015 Challenge