Adam Jermyn comments on Multi-Component Learning and S-Curves

[Adam Jermyn]

I don’t think so? I think that just means you keep the incorrect initialization around while also learning the correct direction.