Black-Box Optimization Lab

AboutAbout the lab

From the design and analysis of evolutionary computation to reinforcement learning and machine learning

The Black-Box Optimization Lab mainly studies the design and analysis of black-box optimization techniques such as evolutionary computation.
Applications to engineering design and optimal control, reinforcement learning, and automated machine learning are also among its research subjects.
The lab is also interested in the design principles and theoretical analysis of evolutionary computation, trustworthy AI technologies based on robust optimization and reinforcement learning, and the use of generative models.
It is a lab of the College of Information Science at the University of Tsukuba and of the Computer Science Program in the graduate school.

ResearchResearch themes

The lab designs and analyzes black-box optimization techniques, centered on evolutionary computation, for solving problems whose objective function cannot be written as a mathematical formula. It is interested in the design principles and theoretical analysis of evolutionary computation, the automation of machine learning and reinforcement learning, trustworthy AI technologies based on robust optimization and reinforcement learning, and the use of generative models.

*Illustrative image

Automatic Ship Berthing↗

In collaboration with a group at Osaka University, the lab works toward the realization of autonomous ships. Using the elemental technology of automatic berthing as an example, it presents the acquisition of maneuvering rules by reinforcement learning and the determination of simulator coefficients by evolutionary computation.

*Illustrative image

Game Level Generation and AI Agents↗

The lab presents its research on generating game stages (levels) with machine learning and evolutionary computation, using Science Birds as an example. The AI agents that play the game are trained by reinforcement learning.

*Illustrative image

Automated Machine Learning (AutoML)↗

The lab works on techniques for searching for the hyperparameters that users set in machine learning and reinforcement learning. Reinforcement learning and evolutionary computation techniques are also widely used in automated machine learning itself.

*Illustrative image

Evolution Strategies (CMA-ES)↗

One of the lab's research topics is evolution strategies, a method for optimizing continuous parameters without using the gradient of the objective function. CMA-ES in particular is a versatile and efficient search method used in many settings, and the lab publishes learning materials on it.

VoicesStudent voices

Ratings and reviews reflect the experiences of individual current or former students, not the lab's official views. Whether a lab has reviews, or how many, says nothing about the lab itself.

Students who take part in an interview may receive a thank-you gift worth ¥1,000, regardless of what they say.