My lay-person's understanding is that it's due to the problem one is trying to solve with a deep learning model: draw a curve through the dimensions which separ

My lay-person's understanding is that it's due to the problem one is trying to solve with a deep learning model: draw a curve through the dimensions which separates "good" from "bad" activation values. The lower resolution the line, the higher likelihood the line will fit sometimes and veer off into erroneous space others imagine trying to draw the blue line on the right using only lego blocks: https://youtu.be/QDX-1M5Nj7s?t=1202 https://youtu.be/QDX-1M5Nj7s?t=1202 discussion: https://news.ycombinator.com/item?id=35405338 https://news.ycombinator.com/item?id=35405338