[4N1-GS-1] Fundamental AI, theory: algorithm:

[4N1-GS-1-05] Creating AutoML Modules with Competition

〇Masaki Onishi¹, Yoshiaki Bando¹, Hirokatsu Kataoka¹, Noriyuki Aoi², Takeyuki Nakae², Takahiro Kusumoto², Masanori Sueoka³, Atsushi Atarashi³ (1. AIST, 2. SIGNATE, 3. NEDO)

Keywords:AutoML, Hyper-parameter optimization, Transfer learning, Competition

Machine learning, such as deep learning, is becoming more complex every year and requires various adjustments. Research on automated machine learning (AutoML), which automates these adjustments and accelerates the use of machine learning, has been reported. Hyperparameter optimization, neural architecture search, meta-learning has been proposed as representative studies of AutoML. Furthermore, transfer learning is also useful as research to accelerate the use of machine learning. On the other hand, the importance of human-in-the-loop machine learning has been pointed out in recent years. In general, humans are often the annotators in human-in-the-loop machine learning, but by involving programmers in the loop, it is expected to accelerate the introduction of machine learning in the field. Competitions are being used for this purpose. We conducted a module competition for hyperparameter optimization in FY2022 and a module competition for pre-training dataset generation in FY2023 with the goal of accelerating the use of machine learning. This paper provides an overview of the contests and their significance, and identifies the findings obtained by conducting the contests.