[1Win4] Poster session 1

[1Win4-01] Phase Estimation Using a Delayed Coordinate System for Rhythm Phenomena with Hidden States

〇Koichiro Yawata¹, Norihisa Namura¹, Yuzuru Kato², Hiroya Nakao¹ (1.Institute of Science Tokyo, 2.Future University Hakodate)

Keywords:Dynamical system, Physics-informed machine learning

Rhythm phenomena are observed in a wide range of fields, from biology to engineering, in the real world. They are modeled as limit cycle oscillators, and phase reduction theory is useful for understanding the synchronization phenomenon of interacting limit cycle oscillators. In phase reduction theory, the state of a limit cycle oscillator is expressed in terms of a single variable called the asymptotic phase, and many methods have been proposed for estimating this asymptotic phase in a data-driven manner. Phase autoencoder is a method for estimating the asymptotic phase using an autoencoder in a data-driven manner. In previous studies, including phase autoencoder, it was assumed that all states of the system could be observed, but in reality, systems have hidden states and only a limited number of states can be observed. In this study, we extended phase autoencoder to be able to use the delay coordinate system of the observable states for limit cycle oscillators with hidden states, and made it possible to estimate the phase of limit cycle oscillators with hidden states.