7 ms·
Brain Computation Is Organized via Power-of-Two-Based Permutation Logic
- dkdkdjfjf 10y agoWould anyone here be able to offer some book suggestions for tackling AI from the neuroscience/ biological perspective, as opposed to the computer science/ computational perspective? I want to expand my horizons.
- jonsen 10y agohttps://en.wikipedia.org/wiki/On_Intelligence https://en.wikipedia.org/wiki/On_Intelligence
- abrax3141 10y agoI don't think anyone serious buys this version of the GUT. You'd be spending your time better reading Randy O'Reilly's work (CU Boulder). Or: https://en.m.wikipedia.org/wiki/Rethinking_Innateness https://en.m.wikipedia.org/wiki/Rethinking_Innateness
- abrax3141 10y agoWouldn't that be just "I", and the whole field of cognitive neuroscience?
- kurosawa 10y agoThe following suggestion is not about tackling AI from neuroscience/biological perspective, but one that critiques the computational view of mind/brain - which provides only a limited picture. See the work of prof. Walter Freeman: https://en.wikipedia.org/wiki/Walter_Jackson_Freeman_III https://en.wikipedia.org/wiki/Walter_Jackson_Freeman_III Esp: Reclaiming Cognition - Ed. Nunez and Freeman. Also see the work of Hubert Dreyfus. These works will help you figure out what 'tackling AI from a neuroscience' perspective could actually mean.
- DocSavage 10y agoI would suggest "Principles of Neural Design", a book that tries to explain why things might be engineered the way they are in biology: https://mitpress.mit.edu/neuraldesign%20 https://mitpress.mit.edu/neuraldesign%20
- posterboy 10y ago"explains" and "might" are kind of at odds, here
- kensai 10y agoExcept for the Dopaminergic neurons, among other "subsystems".
- aisofteng 10y agoThat is explicitly mentioned in the article.
- goldenkek 10y agoSome questions: We act like stimuli are exact things but the brain is all about abstraction. In fact, a certain amount of entropy/information enters your beain every day from vision, smell, touch, etc. Its not stored in full fidelity. There arent enough nuerons or energy for that. So what is lost? Well...whatever isnt abstractable, right? You celebrate your birthday. You remember blowing out the cake in pristine vision. But how much of that 'movie reel' inside your head is actually a superposition of abstracted reality vs actual observed reality? All Im asking is, is it even worth it to think of a human being in terms of absolute information when their memories are so sparse from reality. We remember whats important. And the things that are important, weve remembered to remember. But ultimately, the actual holding power of the brain is quite small. The brain is a master of deep abstraction from sparsity. Is it like a quantum system? Where the superposition values of one subcortical system are flattened by another? So that a whole brain belief/memory is a function of the fuzzyness of each subsystem being exacted/wave function flattened by the heuristic correlation/connection between all the shitty data? I think we need to start thinking about sparsity. The universe is sparse when it comes to using 3d space. Holographic principle says physics is so symmetrical that we sparsely use 3d space. Such that our entire universe could take place on 2d space. Max Tegmark gave a talk about how deep learning works so well because of the sparsity of faces or music or voices, the overwhelming symmetry/redundance in physics and nature. And transitively, humans.
- dualogy 10y ago> So what is lost? Well...whatever isnt abstractable, right? Raises a most curious question.. what ever isnt abstractable?
- pmontra 10y agoWhy that pdf in the title? The page is a long html article.
- jessaustin 10y agoThis simple mathematical logic can account for brain computation across the entire evolutionary spectrum, ranging from the simplest neural networks to the most complex. That is, further research will investigate how e.g. a mouse turns in a particular direction in a maze before investigating how high-performing humans perform amazing mental feats. It may all be the same at some level, but we have to walk before we can run.
- MrQuincle 10y agoThere are only a few organizational principles that allow the number of groups (K) to scale beyond the number of neurons (N). Polychronization (http://www.izhikevich.org/publications/spnet.pdf http://www.izhikevich.org/publications/spnet.pdf). This is e.g. different from synfire chains in which the number of groups equals the number of neurons. I don't remember much about the capacity analysis of standard reservoir computing methods (ESN, LSM, ELM), but I thought it was also limited to N. I'm looking forward to more research in this direction! We as humans build up such long-term memories. This must be a driving factor in AI. A lot of prediction tasks hinge on being able to handle this huge temporal scale. Currently I'm trying to see if I can formulate chains of transition matrices in MCMC that allow simultaneous search on multiple scales. I feel like the math from multiple angles is converging, exciting! We're gonna solve this in our lifetimes!
- joggery 10y agoWai H. Tsang: http://www.academia.edu/27961545/The_Fractal_Brain_Theory_-_Chapter_4_-_Binary_Trees http://www.academia.edu/27961545/The_Fractal_Brain_Theory_-_...
- mrfusion 10y agoThis seems pretty hand wavy. I'm not seeing any actionable insights from this.
- kpmcc 10y agoFor real though. We already knew that neurons can do binary logic, they either fire or they don't and neurons higher up the chain fire based on converging inputs. Also what do those plots actually 'show'. The authors claim repeatedly that the plots show the existence of 15 clusters, but without stating p values relating to any sort of classification for many of the plots.
- silos372 10y agoIt seems that by "permutation" they mean what a mathematician calls a "subset", or a computer scientists calls a "bitmask". Essentially they found lots of evidence that animal brains implement demultiplexers?
- abeppu 10y agoI spent way too long staring at figure 1, before realizing that they're not using the word "permutation" in the way that people with exposure to CS or discrete math would use it. They're assigning a "neural clique" (possibly not exactly a clique in the graph theory sense) to each subset in the powerset of inputs, and then ignoring the empty set. _Now_ 2^i - 1 makes sense; ordering does not matter to the authors.
- harryjo 10y agoIt seems like you are saying that the headline is nonsense because the authors of the article are using non-standard meanings for their words. Any system with optional components has a factor of 2^n in its count of configurations -- each optional component might be absent. That's an interesting fact of math, not of any particular system.
- abeppu 10y agoI am not saying it is nonsense -- just that it was hard for me to understand their claim, in part due to a background that I expect many other readers here share.
- dzdt 10y agoThis sets off my BS detector. The approach is to make a "thought experiment" about how neural computation might work mathematically, then look for evidence that the guess is correct. Their thought experiment makes them guess powers of 2 are special, that they should expect to see organization at sizes 1,3,7,15. But when they analyze data, I don't see any comparison saying that 7 is really more prevalent than 6 or 8, or 15 is more prevalent than 14 or 16. Indeed 15 seems to be the largest they looked at. The whole numerology theory driving the headline looks like unlikely, unsupported nonsense to me!