Auto-reservoir neural network for multistep-ahead prediction

Speaker:Rui Liu (South China University of Technology)

Date:October 27 Wed. 2:00pm

Venue:  ZOOM ID：675471697

Abstract:It is a challenging task to accurately predict the future states of a short-term time -series. The major difficulty to solve such a task is the lack of the information, which typically results in the failure of most existing approaches due to the overiting problem of the small sample size. To address this issue, we recently proposed a computing framework: Auto-Reservoir Neural Network (ARNN), to efficiently make the multi-step-ahead prediction based on a short-term high-dimensional time-series. Different from traditional reservoir computing whose reservoir is an external dynamical system irelevant to the target system, ARNN directly transforms the observed high-dimensional dynamics as its reservoir, which maps the high- dimensinal/spatial data to the future temporal values of a target variable based on a spatiotemporal information (STI) transformation. The application in predicting the tipping points of biological systems will also be referred in this talk.