Bit Arrays (Episode 1)

[@zoidbergZA]

fantastically clear introduction lesson, i'm a video game programming student working on a project that is being built from the ground up to be played by neural networks(the non biological kind) and your videos can't come quick enough! thank you very much for the good work!

[@cepi24]

Those videos are amazing, keep on rocking.

[@ivanlaws622]

I haven't done this stuff in 30 years. Good presentation!

[@bradleyfrank7933]

I love these videos. keep them coming.

[@Test_749]

wonderful explanation

[@jonclement]

nicely done and good example of "features". My suggestions: introduce an example from that start that you can slowly develop through these episodes. ie/ refer back to an image. So, a 1 is a black screen, a 0 is white. Show an 8 bit image...compression...etc.

[@godbennett]

Indeed. Albeit, is the claustrum synonymous in " ponrose' " collapse?

[@deepakraya9660]

awesome!, amazing concepts of HTM, Mr.Jeff Hawkins is awesome, numenta is amazing

[@bzzzvzzze]

Love your videos!

[@DeathscreamerS0S]

That dig at Jay Z though

[@haroldturing5958]

In about 8:37-8:38, 4 bits in 256 bit array is not 2%, have you rounded the decimal result?

[@MatthewTaylor-rhyolight]

Yes, I rounded. SDRs in the brain (and in our software) are much larger than this, so there are no big rounding differences like this.

[@haroldturing5958]

@@MatthewTaylor-rhyolight OK, thank you

[@vaigokhale222]

Excellent video man! You rock

[@mdellertson]

Great stuff!

[@dragolov]

Respect!

[@astalavisitor]

sparse representation is 256 bits or 32Bytes of 8bits , right? An index value is 8bits or 1 Byte, exactly 1/32 of space of Sparse Representation of 256bits. So with Compression *we are limited to only 32 indices or less* in order to gain any advantage in,lets say, storing the data? Or could we save 4-bit indices which could increase total index number over 32? I am really noob at this and any advice would be welcome.

[@NumentaTheory]

And SDR can be any number of bits, the lengths you mention are only examples. Realistically, we deal with much larger arrays that I talked about in this video. 40 / 2048 bits active is typical, which is 2% active.

[@astalavisitor]

HTM Open Source yes alright, but that doesn't answer my question. Is the total number of indices bits limited to less than or equal the number of bits the SDR has? lets say SDR with 2048 bits, then the compression indices bits must be

[@NumentaTheory]

I think the next two episodes of HTM School about SDRs will answer your question. This one was really just a primer on SDRs. Have you watched them?

[@astalavisitor]

HTM Open Source ah. Not yet xD

[@mukteshgautam4050]

thanks sir just a doubt how this indices 86,161,206,253 calculated

[@NumentaTheory]

It all depends on what is being encoded.

[@broken_abi6973]

Jay Z, for various reasons xD

[@sharp7j]

Was that a joke calling him not a musician lol?

[@alainghawi9122]

Yes it was or he's planning to kill him :D

[@SCOOTI12]

Llllmmmaaaooo I scrolled all the way down this muhfuh to be like “I GOT YOUR LITTLE JOKE FAM.” And then I remembered this guy just taught me hundreds of years worth of science so I went to episode 2

[@SCOOTI12]

I feel like how I felt listening to “U don’t know.” “I smartened up, opened the market up.” Numenta is the Rocawear of neuroscience. Where the Rocafella brand told Tommy Hillfigger “Fuck off we’ll fund ourselves,” so too did Jeff Hawkins say to traditional research channels. He smartened up, and is now opening a SECOND new market up.

[@Mayhong92]

Is there any special meaning in 32bits? like, The size of instruction should be 32 or 64 bits within the framework of computer architecure(Especially in MIPS architecture).

[@MatthewTaylor-rhyolight]

+강민성 The only special thing about 32 and 64 is that they are powers of 2. :)

[@Mayhong92]

+Matthew Taylor Thanks!

[@AZTECMAN]

I wanted to fact check you on the 8 bit thing around 3:40... according to this table, www.sciencebuddies.org/science-fair-projects/project_ideas/CompSci_ASCII_Table.shtml 'm' is 01101101 which has a value of '109' ; the letter 'n' is 01101110, which is equivalent to '110'... so in this case changing one byte does give us something of semantic value, though admittedly, the whole system is kind of arbitrary at its core.

[@NumentaTheory]

You are right, there is some semantic encoding in some places in the ASCII set. Because they use accumulating bits as they count through the letters in some places. But I'm not sure it is consistent enough throughout the data set to be a good representation.

[@DHorse]

10:15 Array Compression. Hi. I am familiar with this. You store / send an index key with an optional (non binary) value. Q: Is this related to why to GPU's are not a good match for the theory algorithms? ... One more thing Matt... What an awesome place to employ BLOCK CHAIN in AI... 😉😇🤐

[@Mayhong92]

Thank you! First comment~ :)

[@softwaredeveloper6791]

A bit basic, if you'll pardon the pun.

[@lasredchris]

Dense representation