A learning thread?

[Baron von Lotsov]

This is the latest in how to make machines intelligent. LSTMs are neural networks that have memory and the ability to forget as well. They solve the problem when using time sequence data, e.g. in speech recognition, stock market predication and so on.



From Wikipedia:

In 2018, Bill Gates called it a "huge milestone in advancing artificial intelligence" when bots developed by OpenAI were able to beat humans in the game of Dota 2.[37] OpenAI Five consists of five independent but coordinated neural networks. Each network is trained by a policy gradient method without supervising teacher and contains a single-layer, 1024-unit Long-Short-Term-Memory that sees the current game state and emits actions through several possible action heads.[37]



In 2018, OpenAI also trained a similar LSTM by policy gradients to control a human-like robot hand that manipulates physical objects with unprecedented dexterity.[38]

This is OpenAI's bot.

This machine was not told how to do this. It figured it out itself.

[Baron von Lotsov]

Just a general point here about Youtube. I see there are two sides to it. There's a lot of spam and marketing on the one side, but then there is the serious side to it, where spam and marketing do not feature. This is where one can get educated, and it is this side of it which really does shine. It is bang up-to-date, so if you are in the technology business that is extremely important, and it is truly international. Above all I get a really positive feeling from those who contribute to it, as per they have very positive motivation and will help each other with technical problems. So you can tune in to the cutting edge as it were, and the other day I was watching something from a Swiss chap who builds and invents technology and then over in the viewers comments there's a lecturer who says thanks very much, I'll use that for my lecture. So far from the conventional idea of trickle down knowledge from great academic institutions, it moves from these Youtuber types to the institutions and then to conventional students paying tuition fees! The Swiss chap refers to this concept of "sitting in the front row" quite a bit, and this is how things seem to pan out. Research and invention is now in the virtual world.  I checked wiki and the thing he was explaining had its first stable release on github in October this year, so that's very new indeed. They are just building the wiki page now for the new internet.

[Baron von Lotsov]

If anyone is interested in Economics, there are some good videos on various countries.



The chap reckons India will be the next country to see an industrial revolution. There's also a good one on the UK's economy. Apparently we are a tax haven for the super-rich European elite. They are the families who have owned most of Europe for the last 500 years or more.



https://www.youtube.com/user/JitaLounge/videos">https://www.youtube.com/user/JitaLounge/videos

[Baron von Lotsov]

Here's how to stop your internet being vulnerable to viruses and the like. This is from a Youtube post, and I believe the chap is telling the truth here! Talk about paranoia.

Cloud computing is vulnerable to external factors and we will never use it at all. My country has two internet systems. The usual global internet network AND THE MILITARY INTRANET NETWORK. For security reasons it is hardwired to be impossible to be connected to the global internet network. So I have 2 set of computers. The 1st one is connected to the global internet and the 2sd one is connected to the military intranet that has all of it's source code HARDWIRED in a maskless ANALOG IC CHIP AND CERAMIC TRANSISTORIZED ANALOG PRINTED CIRCUIT BOARDS functioning together side by side so that if something happens to the IC CHIP system, then the CERAMIC TRANSISTORIZED ANALOG PRINTED CIRCUIT BOARDS will continue to function as it switches off the IC CHIP system. All computer activities are recorded by an ANALOG RECORD ONLY MAGNETIC TAPE SYSTEM and floopy discs and zip drives as well as cassette tapes that are all functioning side-by-side with 4 multiple randuant hard drives. There are 4 multiple redundancies separated individually from each other. If I want to stop and rewind to check and back track to what I have done, then while I am doing that it is still being recorded on the parallel 4 hard drives and when I am finished then when push the bottom RECORD the hardwired program will instruct the magnetic tape recorder to go to where it stop and start recording again FROM THE HARD DRIVE IN A SPEEDED UP RATE UNTIL IT CAN CATCH UP WITH THE REST. And when I turn off the computer, then whatever needs to be recorded from the hard drives will automatically stay on until it is finished and then it deactivates itself. And parallel to all of these 23/7 non-stop record only systems is an ASOM system which stands for Auto-Saved On Microfilm system of which there are 8 systems and parallel with it is an ASOP system which stands for Auto-Saved On Print system. The microfilm are archival quality self-developing and self=stabilizing and the ASP system ues archival mineral ink on archival paper.

[Baron von Lotsov]

Here's a quick lesson for today. I think there are some Youtube links to show demonstrations, but really it is about what a company is doing with robotic actuators. Most robots have joints to them like we have in our arms, and these are known as the axes of a robot, which is the degrees of freedom it has, like you have one at your shoulder and one at your elbow, and they generally move in an arc in one plane with robots. Given between about 4 and 6 of these, then one can make a robotic arm that can do most things. Thinking about this, the ideal spec for these actuators is high torque and speed of rotation, precise positioning with critical damping, and the other issue is backlash. Motors spin typically about 3000rpm+, so you need a fairly high gear ratio to get the torque. You would also like all of this to be nice and compact, so you can pack the biggest motor you can for the space you have, and weight is important since the weight of the arm is dead weight. You will need a fair bit of energy to shift the arm at all before you consider the energy to lift heavy objects.



 So what is the answer?



I guess you might think of small cog driving big cog, but its a bit of an awkward geometry. You could even use hydrolics or pneumatics to do the job, hence meaning you have a lighter arm and the big motor is connected via pipes. The trouble is that fluid is compressible, so your robot might not be that accurate and suffer other glitches. So take a look at this.



https://en.wikipedia.org/wiki/Harmonic_drive">https://en.wikipedia.org/wiki/Harmonic_drive



It's a harmonic drive and solves many of these problems. You can build a very compact motor and gear unit with one of these which gives remarkably precise control. You would therefore link this up to a stepper motor and that would be your robot joint. These days in robotics we use the CAN interface, so it works in the same way as car electrics do, and then we connect the CAN bus to a computer and we're away. The question is then, what performance can this give. It turns out a firm is already making these robots and the spec is 5kg to an accuracy of 0.1mm and an arm reach of about 40cm.



So if you can make a robot with a 0.1mm accuracy then say you want to manufacture something. It might be you have an instrument case and you want to cut the front panel out and drill it and so forth. I think if your mechanical precision was that good then few customers could complain. It can also dace, believe it or not!!!



Here's a page with all the gubbins on, links to demos and everything you could ever possibly want to know about the firm and its robots.



https://www.kickstarter.com/projects/1383636492/the-smallest-servomotor-robotic-arm">https://www.kickstarter.com/projects/13 ... obotic-arm">https://www.kickstarter.com/projects/1383636492/the-smallest-servomotor-robotic-arm



Incidently we're not talking about millions of pounds to build a robot factory. These things are cheap and flexible. I can well foresee people building entire products in their garage whilst snoozing in their bed or in front of the TV. Blessed is the harmonic drive.

[Baron von Lotsov]

I have a problem whose solution is probably somewhere on the internet but I am buggered if I can find it.



Assume that you are conducting an opinion poll with a very simple question such as do you like beer, yes or no.



Then assume that you ask 100 people.



To test the accuracy of the poll you find the square root of the number survey, which in this case is 10.



10 as a percentage of 100 = 10%



So your poll has an accuracy of +/- 10%



Now assume that you ask 1600 people. The square root of 1600 is 40 which is 2.5% of 1600 and thus means that the poll has an accuracy of +/-2.5%. Most pollsters are satisfied with this which is why most polls settle for asking the opinions of about 1600.



The problem is that although this rule (or to be pedantic, hypothesis ) is widely used, I have never found a proof.

The square  root comes from the Gaussian or Normal distribution function. It's like if you measure a length and you have a certain error in your measurement.  Say you take two measurements to get the length though, like if you were chopping a bit out of a 1 m rod in the middle and you had to measure from both ends, then the errors add in quadrature, like 1mm error x 2 = root 2mm

[Borchester]

I have a problem whose solution is probably somewhere on the internet but I am buggered if I can find it.



Assume that you are conducting an opinion poll with a very simple question such as do you like beer, yes or no.



Then assume that you ask 100 people.



To test the accuracy of the poll you find the square root of the number survey, which in this case is 10.



10 as a percentage of 100 = 10%



So your poll has an accuracy of +/- 10%



Now assume that you ask 1600 people. The square root of 1600 is 40 which is 2.5% of 1600 and thus means that the poll has an accuracy of +/-2.5%. Most pollsters are satisfied with this which is why most polls settle for asking the opinions of about 1600.



The problem is that although this rule (or to be pedantic, hypothesis ) is widely used, I have never found a proof.

[Baron von Lotsov]

Here's a practical lesson for you today, or what students call "a practical".



This is like Blue Peter for grown ups. We don't use sticky-back-plastic for this one, but copper tubes, and out of this plumbing we shall have ourselves a mass spectrometer.

[Baron von Lotsov]

Oh yes and after watching that I had a look at spin transfer torque memory and how it works. The beauty of it is that it does not need any power to maintain state. It's considered to be a 100bn dollar industry. This is the firm who invented it.



https://en.wikipedia.org/wiki/Grandis_%28company%29">https://en.wikipedia.org/wiki/Grandis_(company)



People are asking whether they should get the Nobel Prize. hmm.

[Baron von Lotsov]

Here's an exceptionally good video charting the history and future projections of the chip. It clears up a lot of commonly believed bullsht, which I have been very unhappy about for a long time. BBC - you should watch this.

[Baron von Lotsov]

This next video is nothing extraordinary about its treatment of the subject, but the subject itself is very interesting and there are many others you will probably find if you look. It's about a new type of processor chip. Traditionally we had the microprocessor with variants on a standard architecture, like PC processors, and then we had a different type, which were hardware processors of graphics, and is essentially what a "graphics card" is, but now a third type emerges, and this is chip hardware to do AI very fast. So it is like the concept of a graphic processor, but to do AI. As will graphics, the way to accelerate is to process vast amounts of data in parallel, so this works like millions of simple processors working together and each is connected in much the same way as your PC is connected to the internet, but in a chip! The architecture is typically an array of them arranged in a grid, or "cube" as some manufacturers call it. cubes do video, where the third axis is time. Anyhow this is a simple example from the Americans on how understand it in 10 minutes.

[Baron von Lotsov]

I'm finding interesting learning material in many different subjects. I'm kind of picking them on their usefulness in real life.



Here's a youtuber who does quick ten minute lessons on control systems. It's what you would find in a textbook, but much faster and easier to learn thanks to the way he does it.



https://www.youtube.com/user/ControlLectures/videos">https://www.youtube.com/user/ControlLectures/videos



Even if you know it already, it can be helpful to remind yourself.

[Baron von Lotsov]

I have a couple of short videos here on the world's largest radio telescope.



this is the construction phase

https://www.youtube.com/watch?v=7SRV3rnULO0">https://www.youtube.com/watch?v=7SRV3rnULO0



And this is a short clip of it running. It's currently looking for pulsars, and has found a few already. They can be used to tell the time far more accurately than any earth clock.

https://www.youtube.com/watch?v=9eDl0D38r4M">https://www.youtube.com/watch?v=9eDl0D38r4M

[Baron von Lotsov]

Be certain to tell everyone you post links to that they're too stupid to understand it - good tactic, 10/10. I'm sure you'll have a lot of people clicking on your links.. 🤔

I said if, not that you won't. I do not know what unknown people know or like, but I do know that there are a very vocal bunch of people who like to put down all good education because they can't deal with it themselves. So the point it, I'm not pandering to them. I'm putting stuff up for the other type, as in those who do value education and do have an appreciation for those who can teach them more. It's a serious thread, and its aim is to help those who are in this country especially, but also those from abroad, who have no bloody idea which way to turn because a large proportion of their education has been mush. If you have had to learn computing at school then you might be a little shocked to hear what the above chap says. It's the way he thinks which is the valuable thing you can gain from it. So it is like a seed, where if you find value in it you'll naturally want to find out more, and in case it is your thing, he has written some extremely good books. It's like if you want to be a good programmer you need a good teacher. I know of some. Perhaps other people know of others in their subjects. Perhaps they don't. We will see.



By the way, one of my hobbies is to read British job adverts in programming. I find them hilarious a lot of the time. They do it all wrong. This guy will show them the right way.

[Borg Refinery]

Why have you quoted me as posting something I didn't?



Lol. Your post restates your earlier posts but with added condescension... towards folks who tried to play your video, but couldn't because it does not allow embedding.



Be certain to tell everyone you post links to that they're too stupid to understand it - good tactic, 10/10. I'm sure you'll have a lot of people clicking on your links.. 🤔