【閱讀筆記】Detection of time delays and directional interactions
前幾天去廈門開會(DDAP10),全英文演講加之大家口音都略重,說實話聽演講主要靠看ppt,摘出一片聽懂的寫篇部落格紀念一下吧。
11.2 Session-A 13:30-18:00 WICC G201
Time | Speaker | No. | Title |
---|---|---|---|
14:30-15:00 | Wei Lin | ST-07 | Dynamical time series analytics: From networks construction to dynamics prediction |
ABSTRACT
Data-based and model-free accurate identification of intrinsic(固有) time delays and directional interactions.
METHOD
Given a time series , one forms a manifold(流形) based on delay coordinate embedding: , where is the embedding dimension and is a proper time lag.
CME method (detect time delay): Say we are given time series and as well as a set of possible time delays: KaTeX parse error: Unexpected character: '' at position 7: \Gamma̲ = {\tau_1,\tau…. For each candidate time delay , we let and form the manifolds and with and being the respective embedding dimensions. For each point , we find nearest neighbors , which are mapped to the mutual neighbors by the cross map. We then estimate by averaging these mutual neighbors through . Finally, we define the CME score as
It is straightforward to show . The larger the value of , the stronger the driving force from to . In a plot of , if there is a peak at , the time delay from to can be identified as . 可以理解為如果是以延遲作用於,那麼當的情況()類似時,之前的(也就是)的情況()也應該類似(協方差大,相關性強)。
RESULTS
To validate our CME method, we begin with a discrete-time logistic model of two non-identical species:
where , and are the coupling parameters, and and are the intrinsic time delays that we aim to determine from time series. 後面也舉了幾個微分方程的例子。 疑問:他所舉例都是兩個節點的連線,並沒有把方法運用到網路中。