4Paradigm Holds the World’s First AutoDL Challenge with Google and ChaLearn

2019.03.05

The exploration and breakthrough in automated machine learning (AutoML), the key technology for AI’s industrialization, is considered a bright spot for AI development. Last year, 4Paradigm hosted important AutoML competitions like NeurIPS 2018 AutoML Challenge and PAKDD 2018 Data Competition and this year it is to launch AutoDL Challenge 2019, the world’s first AutoDL competition, together with Google and ChaLearn.

Tu Weiwei, senior algorithmic scientist at 4Paradigm and one of the directors of the organizing committee, said warmly welcomed Google aboard to hold the first AutoDL challenge together. In the past, AutoML contests focused on the application of structured data, but in this challenge participants will be confronted with multiple new issues on structured and unstructured data, which would be more difficult. Today we step up to tap into AutoDL and its advancement to make machines smarter. It is believed in the future, it will bring profound reform to the whole industry, both for traditional machine learning and deep learning. This is the purpose and challenge.

Accordingly, the challenge this time is to solve multiple issues with structured and unstructured data using deep learning, text, voice and image. An array of preformatted public datasets will be provided and a switching database will be established for meta learning. Contestants will need to file the automated deep learning solutions for blind sets in various new test datasets and will be ranked according to the result.

In recent years, the results of deep learning application in voice and image have been observed, but due to factors such as the high threshold of the technology and the shortage of talent have prevented it from large-scale transfer. Therefore, AutoDL technology, which can achieve automated deep learning, has become an important direction for industrial research. 4Paradigm has begun the AutoML exploration and application since its founding in 2014 and has released multiple core technologies including FeatureGo, TemporalGo, Deep Sparse Network and Auto-SSL, which have been used in scenarios like antifraud and customized recommendations, allowing AI apps developed by common developers to sometimes surpass those developed by data scientists.

Based on technological investment and solid experience in AutoML, 4Paradigm joined hands with Google and ChaLearn to launch the AutoDL Challenge 2019 in order to expand the AutoML technological layout and lay a firmer foundation to implement AI industrialization on a greater scale.