A Team Led by Professor Chun Byung-Gon of Computer Science and Engineering Department of Seoul National University Developed ‘Wind Tunnel’ Technology That Converts Machine Learning Pipelines Into Art

From left Yoo Kyung-in Ph.D. Department of Computer Science and Engineering of College of Engineering of Seoul National University and Professor Chun Byung-gon [Photo provided: News Wire]

College of Engineering of Seoul National University made an announcement on August 9th that a team led by Professor Chun Byung-gon of the Department of Computer Science and Engineering has developed Wind Tunnel a framework for converting machine learning pipelines into neural networks and optimizing it a technology that seeks to realize functions such as human learning capabilities in computers in collaboration with Microsoft.

This achievement is a core technology that takes the advantages of both traditional machine learning techniques and the latest deep learning techniques and is expected to be used for various actual AI applications such as click rate prediction and recommendation systems.

Deep learning is believed as an evolved version of machine learning and is different from machine learning in that it uses a programmable neural network that allows machines to make accurate decisions without human help. Deep learning techniques has been drawn a lot of attention being revealed to be effective in computer vision and natural language processing.

However for table format data used in artificial intelligence applications such as click rate prediction and recommendation systems traditional machine learning techniques such as linear models and GBDT still show better performance.

When using traditional machine learning techniques one machine learning pipeline is usually built by combining multiple machine learning models and data conversion operations and when learning each element that makes up the pipeline is used after it is learned individually.

The professor Chun Byung-gons team has developed a technology that optimizes various components at once by learning each component of the pipeline individually and converting it into an artificial neural network into a Back propagation that operates various components in the order of upper to lower levels as opposed to a propagation in analysis of the neural network.

The Wind Tunnel framework developed using this technology was evaluated to lead to various research to find a compromise between traditional machine learning and deep learning techniques on table-type data along with higher predictive performance compared to existing methods.

Website: : http://cms.wip-news.com/news/articleView.html?idxno=8331