https://petewarden.com/2017/12/14/how-many-images-do-you-need-to-train-a-neural-network/

https://petewarden.com/2017/12/14/how-many-images-do-you-need-to-train-a-neural-network/

 

This is what I believe in:  “You need as much data as the number of degrees of freedom in your model (parameters)”

 

If the type of data does not change much, we can get away with a smaller one.

 

Training helps reduce it. But it has its own caveats

 

 

http://fastml.com/how-much-data-is-enough/

https://jakevdp.github.io/blog/2015/07/06/model-complexity-myth/