OpenAI goes Hubel & Wiesel, Chapter 2 https://distill.pub/2020/circuits/early-vision/ … (1/)
-
Show this thread
-
Replying to @patrickmineault
I love Jeff Hawkins' (Numenta) "thousands brain theory" for its ingenuity, but this type of research suggests that it is a misunderstanding.
@rhyolight1 reply 0 retweets 2 likes -
Replying to @Plinz @patrickmineault
I don't see what you mean. Nothing new in this article. Did I miss something? These networks have no understanding of reality. The features are mangled throughout and rely on a statistical classification in the top layer. Definitely not how brains do it.
1 reply 0 retweets 2 likes -
Replying to @rhyolight @patrickmineault
They replicate some of the results from Hubel and Wiesel and other neuroscientists, showing similar compositional hierarchies of receptive fields in cortical columns, while the Thousand Brains propose locally complete but differently resolved representations?
1 reply 0 retweets 2 likes -
Replying to @Plinz @patrickmineault
Our theories don't dispute their results. Their results don't disprove our theories. 2 things are missing: movement & richer neuron model.
2 replies 0 retweets 0 likes -
I don't think it will lead to AGI. Still make more sense to me to focus on understanding the brain.
1 reply 0 retweets 0 likes -
I think everyone is wondering how much more of the brain does Numenta have to understand before taking what is known and doing something with it? 1/
2 replies 0 retweets 1 like -
The question is what can we do NOW with the hardware available today? Not much. I'm waiting for a hardware upgrade for HTM to take off, and I finally see it coming.
1 reply 0 retweets 0 likes -
Replying to @rhyolight @SurviveThrive2 and
The idea of objects in reference frames in neocortex is a huge paradigm shift. Big changes take time, and we need to convince more people to look at the AGI problem from our point of view. Part of what I've been doing as community manager for the past 5 years.
2 replies 0 retweets 1 like
I really like this idea. It reminds me of Kanerva's sparse holographic representations, and is reflected in later work by Smolensky. It may not be capturing how the lower visual cortex works, but I think this line of thought has to be pursued.
-
-
Absolutely! The idea of SDRs is the foundation for all this theory. I don't know Smolensky but Kanerva used to work for Jeff.
0 replies 0 retweets 1 likeThanks. Twitter will use this to make your timeline better. UndoUndo
-
Loading seems to be taking a while.
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.