教員一覧 List of faculty members

KATORI Yuichi 教授

KATORI, Yuichi Professor

Katori Laboratory page (For details and updates, click here)

Research Vision

I believe that intelligence is not a static set of calculation rules, but rather a phenomenon that emerges within dynamics that change over time.

This research treats neural circuits as high-dimensional nonlinear dynamic systems and aims to elucidate how information representation, integration, prediction, and reorganization occur based on dynamical systems theory. Through concepts such as attractor structure reconstruction, bifurcation phenomena, and chaos dynamics, we aim to provide a unified description of the principles of flexible information processing in the brain.

Furthermore, we aim to extend these dynamic principles to reservoir computing and physical reservoirs, and establish design principles for energy-efficient brain-inspired artificial intelligence (Brain-inspired AI). We strive to systematize intelligence science based on dynamics, bridging theoretical analysis, numerical calculations, and experimental verification.

Research Projects

Our laboratory focuses on elucidating the information processing principles of the brain and applying them to artificial intelligence through theoretical analysis and computer implementation of neural circuit models, based on nonlinear dynamics and dynamical systems theory.

The brain is a nonlinear, dynamic network of numerous nerve cells interacting with each other, and its behavior has a complex structure that includes temporal context. In this research, we will elucidate the mechanisms of dynamic information processing through the construction of mathematical models, bifurcation analysis, and numerical simulations.

In particular, we are working on the theoretical construction and application development of time-series information processing models, especially reservoir computing. We describe concepts such as predictive coding and multisensory integration as dynamic networks, and are also considering applications to physical reservoirs and ultra-low power edge AI.

Main research themes

• Analysis of neural circuit models based on nonlinear dynamical systems
• The theory and applications of reservoir computing
• Predictive coding and dynamic information processing
• Dynamical analysis of expression switching in the prefrontal cortex
• Physical reservoir calculations using biological neural circuits

Research achievements

A multi-sensory speech recognition model integrating dynamic predictive coding and reservoir computation.

We proposed a hierarchical neural network model integrating predictive coding theory and reservoir computing, demonstrating robust performance even in noisy environments in a multisensory speech recognition task. We incorporated a weighting mechanism based on sensory confidence into the dynamic network and clarified how recursive connections extract temporal contextual information. We present a theoretical framework that describes multisensory integration as a "dynamic prediction process."

Demonstration of physical reservoir calculations using biological neural circuits

We measured the multicellular responses of cultured neural networks using optogenetic stimulation and calcium imaging, and analyzed them using a reservoir computation framework. We demonstrated that biological neural circuits with modular structures can achieve speech classification by leveraging short-term memory characteristics, and further demonstrated that their internal dynamics function as a "generalization filter." This result lays the foundation for the implementation of physical reservoirs based on biological neural circuits.

Elucidating the dynamic mechanism of expression switching in the prefrontal cortex model.

We constructed a spiking neural circuit model including dynamic synapses and reproduced the representation switching phenomenon in the prefrontal cortex. We theoretically analyzed the process by which multiple attractor sets are reorganized according to the task context as a branching phenomenon in a dynamical system. We present a framework for describing the neural circuit mechanisms supporting flexible cognitive functions as a reorganization of dynamics.

Proposal of a high-dimensional dynamic reservoir based on a pseudo-billiard system

We proposed a novel reservoir structure utilizing pseudo-billiard dynamics within a hypercube. We introduced a mechanism that controls chaotic time evolution and guarantees echo-state properties. We demonstrate a new principle of time-domain information processing utilizing continuous-time dynamics and suggest its potential for hardware implementation.

Major publications and papers

• Yoshihiro Yonemura, Yuichi Katori, “Dynamic predictive coding with reservoir computing performs noise-robust multi-sensory speech recognition,” Frontiers in Computational Neuroscience, 18, 1464603 (12pages), DOI:10.3389/fncom.2024.1464603, (2024).
• Takuma Sumi, Hideaki Yamamoto, Yuichi Katori, Koki Ito, Satoshi Moriya, Tomohiro Konno, Shigeo Sato, Ayumi Hirano-Iwata, “Biological neurons act as generalization filters in reservoir computing,” The Proceedings of the National Academy of Sciences (PNAS), 120, (25), e2217008120 (10 pages), DOI:10.1073/pnas.2217008120, (2023).
• Yoshihiro Yonemura, Yuichi Katori, “Network Model of Predictive Coding Based on Reservoir Computing for Multi-Modal Processing of Visual and Auditory Signals,” Nonlinear Theory and Its Applications, IEICE, 12, (2), pp.143-156, DOI:10.1587/nolta.12.143, (2021).
• Keita Tokuda, Naoya Fujiwara, Akihito Sudo, Yuichi Katori, “Chaos may enhance expressivity in cerebellar granular layer,” Neural Networks, 136, pp.72-86, DOI:10.1016/j.neunet.2020.12.020, (2021) (*Recipient of the Japan Neural Network Society Paper Award).
• Yuichi Katori, Hakaru Tamukoh, Takashi Morie, “Reservoir Computing Based on Dynamics of Pseudo-Billiard System in Hypercube,” 2019 International Joint Conference on Neural Network (IJCNN2019), N-20372 (8 pages), (2019) (*IJCNN2019 Best Paper Award winner).
• Yuichi Katori, Kazuhiro Sakamoto, Naohiro Saito, Jun Tanji, Hajime Mushiake, Kazuyuki Aihara, “Representational Switching by Dynamical Reorganization of Attractor Structure in a Network Model of the Prefrontal Cortex”, PLoS Computational Biology, 7 (11): e1002266, (2011).
• KATORI Yuichi, Chapter in the book: "Mathematical Models of Neural Circuits," *Handbook of Applied Mathematics*, Asakura Shoten, ISBN: 978-4-254-11141-5, (2013 2013 ）
• Ichiya Aihara, Shigeki Tsuji, Yuichi Katori, Kazuyuki Aihara, Chapter in the book: "Mathematical Modeling of the Brain," *Introduction to Phenomenological Mathematics*, University of Tokyo Press, ISBN: 978-4130629164, (2013 2013 ）