Profile cover photo
Profile photo
Max Mikhanosha
357 followers -
Does the needful
Does the needful

357 followers
About
Max's posts

Post has attachment

Post has attachment

Post has attachment

Post has attachment

Post has attachment
Someone speaking reconstructed Proto-Indo-European (the ancestor of most modern languages). Surprising how I can actually recognize a few roots, vidanti (seeing) -> russian videt (seeing), manos is self explanatory.

Avis - hauves - oves - ovsa means sheep in russian/Slavic. You can actually hear verb aghenes or such in the middle, this is from "my heart is on fire or my heart is burning" and same root as Slavic aghon - fire, so more of "my heart is firing for"

Pretty amazing how many commonalities survived from the time first Indo-Europeans emerged in the steppes between Ukraine and Caspian

Post has shared content
“Crown jewel” of natural language processing has been open sourced by Google  #nlp  

...prepackaged deep-learning software designed to understand the relationships between words with no human guidance. Just input a textual data set and let underlying predictive models get to work learning.

Google calls it “an efficient implementation of the continuous bag-of-words and skip-gram architectures for computing vector representations of words.”

Deep learning, Howard explained, is essentially a bigger, badder take on the neural network models...

via +Ward Plunet cc\ +Doyle Groves 

Post has shared content
Putting it into my TODO list to check out their library. The site is http://numenta.org/faq.html license appears to be apache like
Jeff Hawkins is open sourcing his Cortical Learning Algorithm (formerly known as "Hierarchical Temporal Memory", I believe), as NuPIC.

He claims that the mission of his company Numenta is to advance the development of this algorithm.

It will be interesting to see how his approach compares to the one chosen by Ray Kurzweil, who is trying to implement very similar ideas (as laid out in his book "How to build a brain") at Google.

I personally like the Open Source approach.

Post has shared content

Post has attachment
The importance of mixed selectivity in complex cognitive tasks Interesting paper in Nature, its paywalled, but PDF link to supplemental materials in the bottom contains most of it.

http://www.nature.com/nature/journal/vaop/ncurrent/full/nature12160.html

My interest in it is from machine learning perspective, and it seem to point to the direction that deep learning neural networkists have it right, and brain decision making works on the same principle. 

I have not read entire thing yet, but what I'm getting from it is that there're "grandmother neurons" that are connected to a high number of assorted brain regions, that encode high selectivity to only certain combination of inputs, and that firing of these "grandmother neurons" corresponds to monkey making a decision.

So that there assorted ensembles in the brain, detecting features, kind of like the first layers of deep learning networks, and then the "grandmother neuron" acts as the final layer and integrator, triggering on specific combination of the multi-dimensional input.

Wondering if there are single grandmother neuron per decision/task? 

Post has shared content
"...and what with that and the telephone and that dreadful phonograph that bottles up all one says and disgorges at inconvenient times, we will soon be able to do everything by electricity; who knows but some genius will invent something for the especial use of lovers? something, for instance, to carry in their pockets, so when they are far away from each other, and pine for a sound of 'that beloved voice', they will have only to take up this electrical apparatus, and be happy. Ah! blissful lovers of the future!"
Wait while more posts are being loaded