Monty Progress Update
Sebas, 15 June 2025
This past week I got the chance to meet with Santiago, Mostafa, and Mohamed regarding Monty. Santiago was kind enough to talk to me more about his Master’s thesis where he investigated hierarchy in the Monty model, by stacking learning modules on top of one another. I think my project will somewhat work off of where Santi left off. When I met with Mohamed and Mostafa, I was prepared with a bunch of questions to ask, which helped me further understand the theory behind Monty. Though Mohamed said to focus on the code, I feel that it is helpful for me to understand the theory so that I have something to base my understanding of the code off of. I will listen to Mohamed though, and I will try to get through the tutorials as quickly as possible. I asked Mohamed if I could continue to work with him after the summer ends and he said it would be fine, so I feel that it has released some weight off my shoulders.
Some interesting questions that I will investigate after I get through some of the tutorials is:
- How do the layers in the Sherif and Khalil paper differ from the layers present currently in Monty?
- How is representation mapped/encoded onto each bit in the vectors of the HTM model?
- For example, are bits in a region assigned by a Hebbian/”fire-together, wire-together” mechanism?
- If so, using vision as an example, what is the threshold for considering a feature which would then cause a change in a bit?
- Surely, determining the threshold for a change in a bit would require some careful consideration as there are some things that some humans would notice, while others wouldn’t, and then the divide between conscious and sub-conscious is even further.
However I do think Mohamed has a point, which is to get through the tutorials in order to generate data. I left the code off last time having passed pytest, and having tutorial 1 in progress.
Tags: Brown, Computational-Neuroscience, Models, Monty