[msellout]

It seems like HTMs would be prone to the same over-fitting problems that other types of neural networks have. I tried to find some comments in the Numenta material about this, but didn't see anything about Grok's strategy to avoid over-fitting. Can anyone help me figure this out?

[marmaduke]

I breezed through the PDF on the HTM Cortical Learning Algorithms and there's nothing new since the 80s -- over-fitting is still a problem. Perhaps there's an implicit assumption that because learning is online, over-fitting doesn't happen or isn't important.