concept hot and cold learning in category deep learning

appears as: hot d cold learning
Grokking Deep Learning

This is an excerpt from Manning's book Grokking Deep Learning.

The real problem is that even though you know the correct direction to move weight, you don’t know the correct amount. Instead, you pick a fixed one at random (step_amount). Furthermore, this amount has nothing to do with error. Whether error is big or tiny, step_amount is the same. So, hot and cold learning is kind of a bummer. It’s inefficient because you predict three times for each weight update, and step_amount is arbitrary, which can prevent you from learning the correct weight value.