What is the proper definition of "layerDelays" for layer recurrent networks ?

1 次查看（过去 30 天）
Nicolas M.
Nicolas M. 2019-1-27
Hi,
I'm having a bit of an issue with the definition of some parameters of the layrecnet function, notably "layerDelays". The documentation of layrecnet refers this parameter as a row vector of increasing 0 or positive delays. However, my understanding of recurrent networks is that each neuron N at state (t) is fed the output the same neuron at state (t-1). Thus, a neuron at state (t+3) for example, has ben fed with its (t+2), which itself has been fed with (t+1) and so on. This creates a recurrence, which can be illustrated as the following:
rnn.png
However, I'm not really sure how the "tap delay" affects this processing. I suppose that, the above scheme, would be the result of using layerDelays=1. And that using a row vector like [1:2] would mean that neurons at state (t) are fed their output at state (t-1) but also (t-2) as two separate inputs, which could bei llustrated like this:
rnn2.png
Is my understanding of this parameter correct?
Is there any detailed reference document that gives further informations regarding layer reccurent networks in matlab ?
Thank you in advance.

请先登录，再回答此问题。

回答（0 个）

类别

在 Help Center 和 File Exchange 中查找有关 Sequence and Numeric Feature Data Workflows 的更多信息

标签

产品


版本

R2018b

Community Treasure Hunt

Find the treasures in MATLAB Central and discover how the community can help you!

Start Hunting!

Translated by