Amount of unknowns in a networks

[bramDelft]

Hi all,

I am using relatively small datasets (1000 time series, with 33 datapoints each) with the following default models:

• SimpleFFN
• DeepAR
• MQCNN
• DeepSSM

Given my small dataset and the sometimes large neural networks, I am worrying whether there are too many unknowns to solve in my network (e.g. weighs, biases) given my dataset. In short; that unknowns >> dataset.

This results in too little data and too many unknowns and will cause less well trained models. Right?

My question is; is there a minimum size of your dataset required for using these models? And is there a way to estimate the amount of unknowns in the listed algorithms?

Thank you in advance!
Bram