Does anyone know how computation requirement scales with an increase in dataset size?

[seunggs]

I couldn't find a discussion about how computation requirement scales with an increase in dataset size. Can anyone shed some light on whether it tends to be linear or not assuming everything is equal?

Thanks!

[SkafteNicki]

By computational requirement do you mean training time or do you mean resources (gpu vram)?

[seunggs]

Any thoughts on this @SkafteNicki?

[seunggs]

I mean training time - for example, something like OpenAI's petaflop-days.