Automating Multi-element Subspace Exploration via Reinforcement Learning

Yi Sun, YinXiao Liu, Wang Zhongyao, BaoLong Niu

Abstract:

Machine learning enters many aspects of our lives
and brings us great convenience. However, building an
effective machine learning model for a specific task requires
not only expertise but also a lot of time and resources. In order
to solve this problem, more and more research projects focus
on automated machine learning (AutoML). In this paper, we
propose an algorithm that can simultaneously optimize the
space of multiple datasets, multiple models, and multiple
hyperparameters. We call this an automating multi-element
subspace exploration algorithm. We first formalize this
problem as a reinforcement learning problem and then we
define the state, action and well-designed reward function in
reinforcement learning system. In addition, we use some skills
and experience to accelerate the entire optimization process.
Finally, our experimental results on multiple tasks
demonstrate that our method is effective.