Can I save previous time step input feature in custom layer?

I want to use previous and current time-step input feature togther to improve performance of RNN, if decleare a property named `preFeature`, and update it using function `updatePreFeature`, I have to declare my custom layer handle class, Is it illegall?

Answer 1

[Yash]

Hi Xie,

In my opinion it is perfectly acceptable to declare a custom layer handle class in MATLAB to implement a property like 'preFeature' and an associated function like 'updatePreFeature' for your recurrent neural network (RNN). In fact, using a custom layer handle class is the recommended approach for implementing custom layers and adding extra functionality to your neural network models in MATLAB. This allows you to extend the capabilities of the existing classes and customize them to suit your specific needs.

By using a custom layer handle class, you can incorporate the 'preFeature' and 'updatePreFeature' functions into your RNN model to improve its performance by considering both the previous and current time-step input features.

Hope this helps!

[Xie Shipley]

Thank you Yash, Do you have some idea about this question, CUDNN_STAT​US_EXECUTI​ON_FAILED?