Summary
2023
Session Number:B4L-3
Session:
Number:B4L-32
Topological Inference of State Space as Effective Goal of Dynamics Learning
Yamada Taiki, Fujiwara Kantaro,
pp.366-369
Publication Date:2023-09-21
Online ISSN:2188-5079
Summary:
Our brain functions rely on the ability to detect the current states of the outside world and infer their evolution. Thus, studying the requirements to achieve dynamics learning is necessary to understand the brain. Reservoir computing (RC) is a recent representative method for deterministic dynamics learning. Although both applicational and theoretical works support the potential of RC, few studies about the achievability of learning in realistic situations exist. This lack of understanding is fatal for further applications, especially when one wants to use the RC framework as a model of the brain. Inspired by previous works, this paper will consider a relaxed but necessary goal of dynamics learning regarding topological inference of state space. This relaxation allows us to consider the achievability of learning with the finite number of learning iterations and the presence of noise in observations. Considered learning goal enables us to treat both deterministic and stochastic observations in a unified way. Therefore, for future works, we expect the development of theories motivated by deterministic and stochastic notions, which facilitate further understanding of our brain.