Hello, I’m @carver– I’m looking forward to getting to know and chat with you fine folks. I’m in Italy for an extended stay, but expect to return to San Francisco in the next year.
I got very excited about HTMs and Numenta quite some time ago. I built an early (perhaps first?) open source implementation of HTMs in Python in 2010. Unfortunately, I wasn’t able to put any time into maintaining it, but I enjoyed it as a way to explore the ideas. I’m happy to see the direction that research in recent years has taken, especially motor integration. Lucky for me, I have an opportunity to put more time into TBP now.
My background is in software, where I can most readily contribute. I have started sending some code submissions in for some basic improvements. I hope to make some deeper contributions soon.
As for what gets me excited, I’m especially interested in learning about the neuroscience. As I look at Monty, I keep coming back to a few big ideas that I would like to pursue and understand. Things like:
- Modeling statistical object behaviors, like a wriggling worm, or a flame
- Hypothesis generation in constant time (I think Monty gets slower with more learned models currently)
- LMs that learn subsets of an object’s morphology/behavior, with a separate language model that ties together LM outputs into a single mental model
- Connections between Monty capabilities and infant/childhood development
- Connections between object identification and SLAM
Each one of those deserves its own thread I think. I’ll start by looking for existing discussions about them. I’m looking forward to contributing together with you all!
