Toshitake Asabuki, ECL Unit Leader

Please describe your research.

My research uses computational models to investigate brain function. One of the tools I use for this is a so-called recurrent neural network. In the brain, recurrent neural networks abstract and reorganize learned information and experiences into an ‘internal model’ of the external world. Learning occurs by comparing new inputs against this internal model and making predictions, thereby enabling efficient and sophisticated neural information processing. We aim to formalize this mechanism mathematically, and are developing theoretical models to test using experimental data. This will help us better understand the brain’s flexible information processing.

How did you become interested in your current research?

My interest in the brain stemmed from an illness that my mother had. Soon after I began my studies at Waseda University, my mother suffered a cerebral hemorrhage. She regained consciousness, but lost part of her memory. Over time, I witnessed her gradual cognitive recovery through rehabilitation. She began to recall events—not from recent memory—but in the order of her developmental timeline, from early childhood onwards. It was through that experience that I came to see the brain as a kind of system. That sparked my deep interest in neuroscience.

What has been the most interesting recent discovery in your field?

For me, one of the most interesting recent discoveries in theoretical neuroscience is the concept of neural manifolds. This framework suggests that, despite the high dimensionality of neural activity, population dynamics are often confined to neural manifolds, which are low-dimensional subspaces. These manifolds offer a powerful way to interpret trial-to-trial variability. It’s an especially exciting concept because it brings together theory and experiment in a remarkably elegant way, providing deep insights.

My research is important for society because…

It bridges the gap between machine learning and biological brain function, which plays a crucial role in our understanding of how the brain learns and adapts. Using computational models to study synaptic plasticity (strengthening or weakening of connections in response to activity) allows us to interpret complex experimental data and develop models that make new predictions about neural computation. These insights contribute to the development of more effective treatments for neurological disorders and more flexible, adaptive artificial intelligence systems.

What do you hope to achieve with your research in the future?

As I continue my neuroscience research at RIKEN, I’m also looking ahead to future contributions to AI. Current deep-learning systems require vast amounts of data and power. If we can build a mathematical model that mimics the brain’s internal model-building and self-supplementing learning processes, we could develop more efficient AI systems—ones that achieve results with less data and energy consumption.