Mission

Our mission is to explore and establish a methodology of evolutionary intelligence that produces a highly adaptive, autonomous, and theoretically reliable learning and optimization ability. We focus on nature-inspired computing, more specifically, evolutionary computation, and then combine it with machine learning approaches, called evolutionary intelligence or evolutionary machine learning. Our research covers optimization, knowledge discovery as well as other machine learning domains.

From 2019, --- with the start of our lab --- we explore an evolutionary methodology for "an innovative design and its latent law derivation".


Overview

Apart from specific branches of evolutionary intelligence we have dealt with, we here summarize two main techniques.

Evolutionary Computation (EC), inspired from a biological evolutionary process or Darwinism, is a well-known and powerful optimization technique. EC has been applied to many real-world problems and succeed to find a novel solution e.g. industrial design. So many branches of EC have been developed for the last 40 years. We here seek to develop a machine-learning based EC approach (surrogate-EC technique) that enhances a solution-search capacity boosted by experience, prediction and imitation. For the application, we tackled an optical waveguide design optimization problem with a joint work. We succeed to find a better design candidate that maximally derives a high performance ten times than the previously-known manual design.

Evolutionary rule-based Machine Learning, often referred as Learning Classifier System (LCS), is a paradigm of evolutionary symbolic learning approach. LCSs aims to find maximally general and maximally effective rules.Hence, LCSs are used as an logic optimizer which finds a complex, heterogeneous logic behind a problem, that is, producing human-readable knowledge. In terms of basics, we explore a theoretially reliable and general framework of LCS for a wide range of machine learning domains e.g. data-mining, prediction, and reinforcement leraning schemes. For instance, we derived a leanring theory which enables LCS to theoretically maximize its learning performance with finit learning iterations. For application, we applied LCSs to "Design strategy geneation for Hybrid rocket engine", "Detection of transonic buffet signature" and "Care plan design for nursing home".


to Candidates

We're more than willing to welcome candidates for our laboratory.

Our labpratory is always open to visitors but please make sure to make an appointment by Contact before visiting us.