"we'll get to both" -- amazing. My recommendation (which you've hinted at already) is to draw a flowchart of the typical start to "finish" path. So encoders is at/near the start etc whereby getting to an SDR format then allows for a stack of operations (union, addition etc) to be applied. Need help? Feel free to reach out to me or one of us.
+Jon Clement My inclination was to do encoders, but I know some people are anxious to get into the cortical theory. That's why I opened it up in the comments.
OMG! That was sooooooooo good! It really brought the elusive formulas for SDRs; and the known and often touted "benefits" of SDRs - right home! Excellent work Matt! Thank you so much for all the time and effort you've put in to bringing these concepts to the layperson in an accessible and fun way! Kudos!
For future generations, videos from HTM School will all be "classics" from the time when human beings begin to uncover the mystery of AI! Great job, Matt!
Yes encoders are the most logical next step in the progression imo. Also would bet anything that your SpatialPooling conversation will need to refer back to the setup offered by the encoding step?
I know that I’m a year late to the party, but this is so damn cool. Thank you for these videos. I’ve been looking how to combine lispy frame-style information encoding with neural networks and think this is how to do it.
Great series Matt, really appreciate the effort and keep 'em coming. Two didactic points: 1. I'd find it helpful if you quickly outline where you're going with it when you introduce a new concept, helps establish context. 2. You didn't explain the semantic of a false positive.
doing spatial pooling next sounds great! I hope you get to the sensory-motor processing parts of the neocortex as well, those will be quite fun to learn about. :)
Encoding is interesting because I actually can remember what it means and I'm somewhat fuzzy on how it should be done, and it's interesting how subjective judgement goes into it, even though biologically [I guess] it never does. [Or does it--probably a lot of evolution goes into optimizing a sensory signal.] But I also would like to know something about what Spatial Pooling even is. Maybe do Encoding first but be sure to give us Spatial Pooling.
Being rather new to the several aspects of artificial intelligence, I find this tremendously exciting! I was still stuck with the 'old' Neural Networks models, and I would never have guessed that such leaps forward had already been acheived. I was wondering: how large is the permanent team and how many people are, in general, working on this project?
Thanks a lot, I'm really really impressed by what you and your team do, I've understood just enough from the papers I've read to see how powerful this can be! Also, being headed towards a life in computer science (still studying for the moment), I was wondering if I could ever help the community in anyway...I'd love to, when I know enough!
got a question sir, in the 4th minute, when the w and theta are both 4, then the match we gonna get, cannot be false, this match is gonna be exactly the same array of bits what we have in our set of arrays. could you please correct me if im wrong, I appreciate the content🙏
See discourse.numenta.org/t/how-to-run-htm-school-visualizations/2346. All you need is to start the “HTM Visualization Client” which is a nodejs server.
Noise is random bits in the space that are completely meaningless because they are not a part of the signal. To simulate noise in these videos, I randomly flipped a percentage of the bits for % noise.
Imagine the SDRs that might be created by a sensor like a microphone. The world is full of random noise, like static on a TV picture or fuzz on the radio. Your brain has learned to ignore most of it. In computers, introducing random noise would break everything. But with HTM memory systems it is much different. Your brains works much differently than a computer. Keep watching HTM School and you'll understand!
I'm really asking for a more refined version of this video, which I found very informative, apart from a good definition of Noise - what is means, where it comes from etc. Thanks, Andre. Love HTM School!!!!!!
"we'll get to both" -- amazing. My recommendation (which you've hinted at already) is to draw a flowchart of the typical start to "finish" path. So encoders is at/near the start etc whereby getting to an SDR format then allows for a stack of operations (union, addition etc) to be applied. Need help? Feel free to reach out to me or one of us.
+Jon Clement My inclination was to do encoders, but I know some people are anxious to get into the cortical theory. That's why I opened it up in the comments.
OMG! That was sooooooooo good! It really brought the elusive formulas for SDRs; and the known and often touted "benefits" of SDRs - right home! Excellent work Matt! Thank you so much for all the time and effort you've put in to bringing these concepts to the layperson in an accessible and fun way! Kudos!
+cogmission1 Also the bloopers were awesome too! :-P
+cogmission1 Thanks buddy.
Voting is closed! Encoders will be the next topic. Thanks for all your comments!
For future generations, videos from HTM School will all be "classics" from the time when human beings begin to uncover the mystery of AI! Great job, Matt!
I would argue it's the first in a public space.
Really good presentations Matt. Thank you so much.
These videos are great. Keep up the great work Matt and Numenta et al!
Yes encoders are the most logical next step in the progression imo. Also would bet anything that your SpatialPooling conversation will need to refer back to the setup offered by the encoding step?
I know that I’m a year late to the party, but this is so damn cool. Thank you for these videos.
I’ve been looking how to combine lispy frame-style information encoding with neural networks and think this is how to do it.
Be sure to read: arxiv.org/abs/1503.07469j & arxiv.org/abs/1601.00720
Another brilliant episode Matt, thank you.
My vote is Encoders. I'll watch anything you make though, even if that doesn't win.
Thanks for videos - keep them coming! Spatial pooling next please!
Very much enjoying these videos and looking forward to _encoders_ next! :)
These videos are so great, thank you!
I’m so happy this channel exists
Great series Matt, really appreciate the effort and keep 'em coming. Two didactic points:
1. I'd find it helpful if you quickly outline where you're going with it when you introduce a new concept, helps establish context. 2. You didn't explain the semantic of a false positive.
Nice episode. Encoders will be amazing in the sequence.
Please VOTE HERE: discourse.numenta.org/t/htm-school-episode-4-sdr-sets-unions/455
Loving the bad jokes at the start of the episode :)
Thank you for making this series Matt! Spatial Pooling next!
I vote for spatial pooling...great series, thanks
please keep making new videos !!!!
Great videos! Encoders next please
That was very informative and fun to watch!
These tutorials are amazing, but it would be nice to have a faster paced version of them.
I can’t go any faster.
My preference is Encoders next and Spatial Pooling later.
I think the next one should be about Spatial Pooling because it makes sense to fully understand the theory before diving into how to encode the data.
Great videos! I vote Encoders.
Are there any other uses of unions other than improving performance? Really enjoying the videos so far!
Thanks! There might be, but none I know off the top of my head.
This would benefit with a demonstration involving 1B members as opposed to 100.
Spatial Pooling - this is the essence of the process. encoders are easy
Could you please share the GUI that you are using for SDR behavior demonstration.
discourse.numenta.org/t/how-to-run-htm-school-visualizations/2346
Spatial Pooling would be awesome
doing spatial pooling next sounds great! I hope you get to the sensory-motor processing parts of the neocortex as well, those will be quite fun to learn about. :)
Encoders please!
Great videos. thanks
it would be good to know in what situations you have a stack of sdr. perhaps it is from an encoder?
Make the Encoders the next video. We need to know how to supply SDRs before we will able to do some pooling.
Spatial Pooling!
Encoding is interesting because I actually can remember what it means and I'm somewhat fuzzy on how it should be done, and it's interesting how subjective judgement goes into it, even though biologically [I guess] it never does. [Or does it--probably a lot of evolution goes into optimizing a sensory signal.] But I also would like to know something about what Spatial Pooling even is. Maybe do Encoding first but be sure to give us Spatial Pooling.
Encoders!
SDRs into different sets
Have we seen this before ?
+1 encoders, because I want to understand the process/algorithm from the input, from the "front door" so to speak.
Encoders pls!
Being rather new to the several aspects of artificial intelligence, I find this tremendously exciting! I was still stuck with the 'old' Neural Networks models, and I would never have guessed that such leaps forward had already been acheived.
I was wondering: how large is the permanent team and how many people are, in general, working on this project?
We are a very small team, < 20. But there are hundreds of people in the HTM Community: discourse.numenta.org/
Thanks a lot, I'm really really impressed by what you and your team do, I've understood just enough from the papers I've read to see how powerful this can be!
Also, being headed towards a life in computer science (still studying for the moment), I was wondering if I could ever help the community in anyway...I'd love to, when I know enough!
Join in the conversation in the NuPIC section of our forum! discourse.numenta.org/c/nupic
Done, thank you! So many wonderful things to learn, too, it's going to take a while to be up to date.
Keep up the incredible!
Hahahahah Matt, laughed my ass off at the intro! Love the way you do the videos, keep it up
w00p Encoders!
got a question sir, in the 4th minute, when the w and theta are both 4, then the match we gonna get, cannot be false, this match is gonna be exactly the same array of bits what we have in our set of arrays. could you please correct me if im wrong, I appreciate the content🙏
I mean its not a false positive, the positive what we can get in this situation, HAS to be a true one, because the w and theta have a same value
+1 encoders
Are the slider visualisations available anywhere?
discourse.numenta.org/t/how-to-run-htm-school-visualizations/2346
Thanks!
What is the software you are using to explain sdr
I wrote it, see: discourse.numenta.org/t/how-to-run-htm-school-visualizations/2346
I think it's too late to have my vote mean anything. ;)
Encoders
Encoders :)
Is there any way to access this SDR visualization web site?
+MobileGuru732 Yes, but I've only tested it on Chrome: nupic-community.github.io/sdr-viz/site/index.html
+Matthew Taylor Thank you Matt! Nice to see some SDR implementation is Javascript.
Why can't those examples you use in the video be on a web page? Surely its simple Javascript.
See discourse.numenta.org/t/how-to-run-htm-school-visualizations/2346. All you need is to start the “HTM Visualization Client” which is a nodejs server.
do encoders, first
You talk about Noise but don't say what noise *is* in the context of SDRs (unless I missed it, if so I'm very sorry)
Noise is random bits in the space that are completely meaningless because they are not a part of the signal. To simulate noise in these videos, I randomly flipped a percentage of the bits for % noise.
Bits cannot be random - if they were, computers would be useless. Can you please be more precise.
Imagine the SDRs that might be created by a sensor like a microphone. The world is full of random noise, like static on a TV picture or fuzz on the radio. Your brain has learned to ignore most of it. In computers, introducing random noise would break everything. But with HTM memory systems it is much different. Your brains works much differently than a computer. Keep watching HTM School and you'll understand!
I'm really asking for a more refined version of this video, which I found very informative, apart from a good definition of Noise - what is means, where it comes from etc. Thanks, Andre. Love HTM School!!!!!!
Here are some more resources: arxiv.org/abs/1601.00720 & arxiv.org/abs/1503.07469 & arxiv.org/abs/1602.05925
Encoders !
Encoders please!
Encoders!
Encoders!