[4I2-OS-1a] AutoML（自動機械学習）

[4I2-OS-1a-01] Reducing search space for high-dimensional Bayesian optimization by Variational Auto-Encoder

〇Yohei Kanzaki¹, Kazuki Ishikawa¹, Ryota Ozaki¹, Masayuki Karasuyama¹, Yu Inatsu¹, Ichiro Takeuchi^2,3 (1. Nagoya Institute of Technology, 2. Nagoya University, 3. RIKEN)

Keywords:hyper-parameter optimization, bayesian optimization, variational auto-encoder, dimensionality reduction

Bayesian optimization is used in various fields, but it is known to not work well in high-dimensional search spaces. One approach to address this problem is to transform the high-dimensional input space into a low-dimensional latent space and perform Bayesian optimization in the latter low-dimensional space. In this study, as one such approach, we propose a new high-dimensional Bayesian optimization method that integrates manifold Bayesian optimization and predictive distribution reconstruction.