In this presentation, Viviane takes us on a deep dive into Monty and its sensorimotor modeling system. It provides an overview of its current capabilities and limitations, showcasing the advancements made over three years. We cover object and pose detection, modular structure, learning efficiency, multi-object environments, and real-world sensor integration.
I think the question and discussion around the locality sensitive hashing in this video is interesting as it can help optimize the KD tree search significantly but of-course in high-dimensional spaces. It would be interesting to know more details, if it possible share more around this.
Thanks for your interest @mirroredkube! Here is a folder with some code for stuff I tested to speed up the algorithm: monty_lab/speedup at main · thousandbrainsproject/monty_lab · GitHub
Particularly the KNNSearch.ipynb file (monty_lab/speedup/KNNSearch.ipynb at main · thousandbrainsproject/monty_lab · GitHub) This is where I briefly tested LSH (at the very bottom) but I am not an expert in this so I might have applied it wrong. If you have some expertise in this area we would certainly appreciate any pointers you may have (or even a PR) 
watching this video and correlating other things seen today and some years ago:
1:12:28, where Jeff pins down the crucial necessity of grounding methods to the physical (or another, predictable) linked with today’s posts on generated videos clearly not grounded in physical reality:
Remembering this painful watch (poor team):
of Hayao Miyazaki unable to hide his disappointment with “AI”, calling it "an insult to life itself.
Perhaps, emulating life with polynomials (sorry, for ad absurdum) is an insult to physics, life and human progress.
That is to say, it’s a courageous thing to do the opposite of the mainstream approach, and I deeply appreaciate that.