# SGD statistical guarantee

I have a question regard online learning with SGD. Is there a way to give a statistical guarantee that the value obtained after $$n$$ samples deviates at most $$\epsilon$$ from the real value?

Thank you guys.