On: What do new Sudoku techniques teach us about real-...

[Buttons840]

Makes me wonder, if we were able to make a fun puzzle out of training machine learning models by hand, would we soon find better training algorithms?

[conv → Buttons840]

Here is another experiment where you can hand-tune the weights of a neural network for a regression problem: https://asdf10.com/nn.html

[milliams → Buttons840]

Here's a fun experiment along those lines: https://koaning.github.io/human-learn/ It's not a game, but it is investigating how having a human in the loop can improve ML models.