An international research group of Prof Shimazaki developed theories to advance the artificial neural network model used in machine learning researches.
Researchers in neuroscience and machine learning fields use the standard model of artificial neural networks in which many neurons interact via digital signals. However, when the number of neurons is large, its behaviour becomes intractable due to high computational costs. Therefore, many researchers proposed different methods to reduce the costs to approximately capture its behaviour.
Using the mathematical framework called information geometry, Dr Aguilera, Dr Moosavi, and Prof Shimazaki constructed a framework that unifies existing approximation methods, and further developed a method to predict the system’s states more accurately. The proposed method outperforms the previous methods, and works well when the system exhibits versatile patterns as seen in biological systems. The framework significantly advances the speed and accuracy of machine learning technology, and allows for analyzing large-scale data.
Miguel Aguilera, S. Amin Moosavi, Hideaki Shimazaki. (2021) A unifying framework for mean field theories of asymmetric kinetic Ising systems. Nature Communications. https://doi.org/10.1038/s41467-021-20890-5