3 ms·
The brain thinks in Lisp parse trees
- richcollins 20y agoHow is a recurrent neural network a "Lisp parse tree"?
- greendestiny 20y agoOr a hierarchy a Lisp parse tree? Its great to see people thinking beyond procedural languages but there is a lot more to computer science than Lisp.
- erdos2 20y agoBesides, the code is in C++, not lisp (unfortunately).
- albertcardona 20y agoIndeed very unfortunate. From numenta's website descriptions of nupic (the C++ HTM implementation), one can immediately shout Greenspun's Tenth Rule. What motivated my wild comparison of HTMs and Lisp is: - HTMs do not distinguish program and data. - HTMs are fractal like a Lisp parse tree: sameness of structure at any level. - HTMs program themselves from input, analogous to Lisp programs' ability to change themselves at run time. Whether the brain actually uses HTMs or not: as far as I know (as a researcher in neuroscience), the HTM structure fits a simplified description of cortex microarchitecture. As Hawkins says, the field of neuromorphic engineering and computation is in its infancy.
- jey 20y agoThe title is not an accurate depiction of Numenta's "Heirarchical Temporal Memories". "HTMs" are not recurrent neural networks either. I don't like how this guy (Jeff Hawkins, founder of Palm and Handspring) claims to have figured out the fundamental algorithm being used in the cortex without any research or data. His book "On Intelligence" is next in my reading queue and maybe I'll be convinced after I read it... not counting on it though.