[Freddy]

I posted a comment here, but decided to private message so this thread doesn't descend into silliness again.

[LOCKSUIT]

To late lol, what goes on the internet, stays on the internet! lol!

Why would you explain how using such terms? My 8 step explanation is in English and is the only algorithm that could learn to crawl.

Also would you like if I got a baby hamster, put it in a box with food etc, filmed it with a surveillance camera 24/7 to a 2TB HDD, and narrated it?

Also what do you think their spider robot in the video does when it gets to the wall? What do you think about this/think the creator thinks?
---------------------------------------
---------------------------------------
LOCKSUIT, need I remind you what a babbling mess your early posts on your project were ?

It's only now that you are starting to make sense.

We're not idiots, we've been trying to help you. You don't seem to think anyone else has an idea about how these things work.

Be nice to people and don't insult their intelligence. Otherwise please find some other site to visit. This is getting very tiresome.

I meant that the current field, and people that study it, explain things using terms that are unclear. For example the field calls saving senses near look-alikes "labeling data", and doesn't even call senses/actions/searching/etc similar words but rather hidden vague meanings yet are still able to build theirs. Lastly neural networks are just a wire, the functions lay in the boxes, anything they do can be made into functions, if you don't focus on my English terms and focus on the neural nets that are accomplishing the operations, you will be confused.

The hamster question was serious, I'll eventually do it.

[Freddy]

Well, like I said in my message, if you are going to continue to treat people like idiots, then find another site.

You can hardly say that you made much sense when you first got here.

[LOCKSUIT]

I'm trying to be nice and not cause disturbance. As for silly questions like the wall one, I really want to know what you think, maybe everyone else never thought about, just what happens when it gets to that wall lol! What does the creator think! As for the correct terms, well, you probably know the rundown for making a robot learn to crawl like Mr. G I JO robo AI guy in the starfish vid did lookin all formal lol, and as I explained with my terms, I too know the rundown, but tried to ask you to realize that the terms I used are very simple and don't have any better terms that could replace them. Ex: It generates (full/tweak) & saves & does actions, saves senses after passing rewards, searches highest rank as long as similar and initiates linked actions - all of which you'll never directly hear from anyone else in the field.

[Freddy]

I appreciate that you are trying to get along.

You're right, people use different terms, it doesn't mean they don't understand the process is all I am saying. And one is not necessarily better than the other. If it works for you then good.

Anyway, let's just muddle on.

Did you look at the maze learning robots ?

[kei10]

I meant that the current field, and people that study it, explain things using terms that are unclear. For example the field calls saving senses near look-alikes "labeling data", and doesn't even call senses/actions/searching/etc similar words but rather hidden vague meanings yet are still able to build theirs.

Look who's talking. You walked us the visual path of your algorithm with terms from the brain compartments, additionally providing inconsistent details from the actual information, proclaiming that they're not a theory whatsoever, misleading, and has not elaborate most of the points and terms that doesn't exist (perhaps) -- such as "Attention Resistor Cortex". Korrelan gave up, and I, too.

You really love to contradict yourself.

[LOCKSUIT]

Kei, as much as you may think I have on my hands contradictions, trainwreck problems, unwisely making/finding things up as I go along, I can assure you I steadfastly have plus understand it all under my belt and is safe.

Kei, can you explain another algorithm that would learn to crawl other than the one discussed?

[kei10]

Kei, can you explain another algorithm that would learn to crawl other than the one discussed?

Here we are flogging a dead horse. Since I am no expert in the fields of machine learning, I don't have to (maybe one, here). Others here has done the job here showing you already.

To make a machine learn, it must go through something called generation, or evolution. It requires steps... just like we do. Your algorithm is just like any others that does the same goal for this discussion; Teaching something to walk.



So in general of machine learning in my very narrow and unknowledgeable point of view...

• Simulate preemptive movement to acquire first random data, if estimation unfeasible
• Simulate existing preeminent movement data while recording the simulation statistics
• Estimate next preeminent movement data through evolutionary generations acquired from the statistics
• Repeat 2.

[LOCKSUIT]

See same algorithm. Believe it or not there is actually only 1 algorithm, we're interested in. Despite the field says many. In fact this many is what, just KNNs and regression whatnot thingies, while the algorithm we discussed was learning woh look at that, actions, senses, rewards, search, initiate, recall.

....no Freddy, I didn't click the mazes link, as said already though I do have what we do in a maze/wall, also through own examinations etc.

[kei10]

See same algorithm. Believe it or not there is actually only 1 algorithm, we're interested in.

What are you trying to prove here now?

After calling us "Beginner Blindness", and yet you said "Believe it or not there is actually only 1 algorithm, we're interested in".

That could only logically meant that all of us understood what machine learning is, if any of us manage to create or understand any machine that learns -- since you proclaimed that all machine learning is comprised of same algorithm.

You blatantly contradicted yourself again.

Despite the field says many. In fact this many is what, just KNNs and regression whatnot thingies, while the algorithm we discussed was learning woh look at that, actions, senses, rewards, search, initiate, recall.

In the most simplest evolutionary-type machine learning I've seen for teaching a machine to learn to walk does not involve any memory recalling. There are no searching. Therefore, the algorithms of many kinds of machine learning out there, are completely different.

The different terms used by others also means that the functions within the algorithm works differently as well.

Should one say the terms Bubble Sort and Merge Sort algorithm the same, while it does the exact goal; Sorting data?

Algorithm and its purpose is not identical. Teaching a robot to walk is a "Purpose". And the many different algorithms involved are vast and distinct.

[LOCKSUIT]

Kinda weird reply kei but I'll answer it...

The thread wasn't trying to be exxxxxactttt, but rather if you-really-knOw-how-it's-working, you get me? My 8 step explanation early in was perfect, english terms are clear...

Recalling senses and doing linked actions is done when selection of senses occur by input searching for best ranked sense (linked to best actions) but similar to input - all of them do this to improve, again, yous donnot realize. The KNN i.e. assorting senses/other is technique, one of which is used for that one&only algorithm we discussed i.e. generate>reward>select~ Lastly you say many purposes and algorithms, well, to make a working human requires one sort of "purpose or need or *algorithm", and many optimizations on that single algorithm that is the fundamental one&only as discussed, such as KNN for it (assorting senses), but its quite simple, there's no other way to create any of how I work.

We gotta get more solid-er ground, our replies are all over the place and weird and not goin anywhere etc.

[keghn]

  "Saming algorithm?. The algorithm that runs two thing through it to make them equal. Then you subtract the beginning thing
to the after thing to get the amount of change that occurred?

two thing like 18 and 20.
2 + 18 = 20
20 -2 = 18
1/2 * the amount of change is 1 or 0.5 * 2 = 1.
 So the middle ground where these two number are the same are 19.
18 + 1 = 19
20  - 1 = 19

[LOCKSUIT]

keghn..............omg......................................................that only added to creating a complete quagmire

This thread is like over lol, let the future robot ambassadors analyze this now.

[LOCKSUIT]

Art, I never said you didn't know your aerodynamics nor that you can't fly planes, I was only using it as a example to say what I'm seeing from everyone in the AI field rather.

Firstly, I forgot to say that korrelan explanation (reply #2) never even came close to fully explaining how it learns to move forth. See how mine did below. Plus the 20% he did explain, was so long, unclear, and isn't the best choice of words to use to explain it. Please notice how your explanation was NOT full. For example, yours would rank the tried actions, but when does it do them? When it sees cue! Cue is linked to tried actions.

1) You place the robot on the floor.
2) Turn it on.
3) It starts generating random actions for all motors, and are saved as part1 of a single scene.
4) Sensory input passes rewards and happens to trigger the accelometer reward.
5) This scene of senses (part2) is linked to the actions tried, and ranks them.
6) Now when sensory input enters again, it searches and selects the highest ranked but similar sense and initiates the linked actions.
7) It moves forth.
8) The actions it just selected were tweaked, this continues improvement. (only on a pc can you run ex. a 100 different copies off its body guessing/tweaking at the same time).

The commonest method is to combine both the power of a neural net and genetic algorithm/ evolution.

The only pre-set required is target value (say speed forward). 

The neural net is used to move the legs guided by a servo map, the servo map takes its operational parameters for a simulated G-Nome.

A generic servo/ action map is generated along with forty nine or so (the more the better) copies with slight genetic differences.  Each map is loaded into the bots neural net and the servo map is run.  A ‘fitness’ measure variable is calculated for each map from how well the map reached the target.  The two maps with the highest ‘fitness’ at reaching the target are then ‘bred’/ combined/ averaged and another forty nine copies are produced with slight genetic variations. Rinse and repeat until the highest fitness measure obtained.  This is not a perfect solution but does a pretty good job.

Note: This method leverage's the strengths of neural nets and genetic algorithms to enable a bot to learn a task with no human interaction. 

It is NOT the same as generating random servo actions on a single model and hoping your changes result in a better fitness measure. This is like putting all the components for an engine in a box and shaking it, hoping they will assemble themselves.

Secondly, it should not bore people to explain how the robot learns to crawl, my 8 step explanation was a full explanation and was clear and fun. It doesn't involve circuit stuff.

[Korrelan]