Long-short term memory (LSTM) networks are widely used for data-driven time series modeling. Nevertheless, LSTM networks cannot model both the trend and local behavior of the time series, which is especially critical for time series with complicated trend component forms. Additionally, recently proposed neural ordinary differential equations (neural ODEs) allow explicitly modeling time series completely with ODE solutions, theoretically increasing the overall model performance.

Голованев Я.С. (науч. рук. Хватов А.А.) Non-stationary time series modeling with the neural ordinary differential equations // Сборник тезисов докладов конгресса молодых ученых. Электронное издание. – СПб: Университет ИТМО, [2021]. URL: https://kmu.itmo.ru/digests/article/5394