Ensemble learning is a machine learning method that combines multiple models. It is often used for classification and regression. Combining multiple models gives higher accuracy than a single model, but the learning time increases by the amount of the generated model. In recent years, with the development of information and communication technology, the data used for learning has increased, and the learning time has also increased. Furthermore, the necessity of the learning process in the edge environment with a small power supply is increasing.

In our lab, we are researching hardware accelerators to speed up and save power in the learning process of the ensemble learning. We are researching both hardware and software viewpoints, such as adapting existing algorithms developed for software to processing in hardware.

As an example, we studied a learning accelerator of random forest, which is a kind of ensemble learning, using FPGA, which is rewritable hardware.  We are currently working on further improvements to this system.