OK, wasn’t sure if you were meaning it in that context, I saw that operator in other contexts to mean “defined as”.

figured I remembered that correctly…

http://en.wikipedia.org/wiki/List_of_logic_symbols

x := y or x ≡ y means x is defined to be another name for y (but note that ≡ can also mean other things, such as congruence).

Btw, anyone interested in human learning, thinking and understanding, I suggest to look into the field of ‘knowledge management’ and especially ‘tacit knowledge’. It might give a better perspective on what learning and understanding is.

http://en.wikipedia.org/wiki/Knowledge_management

http://en.wikipedia.org/wiki/Tacit_knowledge

My views in relation to strong-AI are very much related to my day-to-day experience in implementing corporate knowledge management systems and how people deal with learning and knowledge. Does that make me ‘biased’ or ‘informed’ ?

Victor.

I intended the definition “becomes” or “becomes equal to”
to support more than one fourth dimension.

http://www.google.com/search?q=define:becomes

However, the fourth dimension of time certainly means
“simple” := “complex”.  So there was no
simple pattern matching solution using a 4D array.

Hans asked, “My views in relation to strong-AI are very much related to my day-to-day experience in implementing corporate knowledge management systems and how people deal with learning and knowledge. Does that make me ‘biased’ or ‘informed’ ?”

Answer: Biased.  Being informed gets you (figuratively speaking) fired.

(That was humor.)

8PLA • NET - Mar 12, 2011:

Answer: Biased.  Being informed gets you (figuratively speaking) fired.

.... that’s why I’m the owner of my company, so I can’t get fired.

I didn’t need your remark to get it

8PLA,

Ok, according to

http://www.google.com/search?q=define:becomes

the meaning of :=  is like the C or C++  meaning of “==”  (as opposed to just “=”)

“=”  meaning *DO IT*,  that is ASSIGN the value, so a=5, means you are TELLING the computer to put 5 into A.

“==” is asking the question… so a==5 returns true if a really is 5, and 0 if a is not 5.

Yes Victor,

You’re perfectly correct.  Not to mention, your logic symbols were valid as a response too.

While := is a variant of == used in Pascal syntax,
in general, it is also the old-school way of writing == .

Why is this important?
If you ever externalize a college degree in computer science
by taking tests, you will most likely see   :=  used in the exam questions.

That’s where I’ve seen it !!!!!!!!!  I knew “:=” was used somewhere… of course.. Pascal !!!  it’s been along time. 

I really enjoyed learning Pascal.  Great beginner language.  Very friendly.

Has anyone here written anything using the library for Pascal called “Turbo Vision” ?  Oh yeah… back in the “MS DOS” days.  Text based GUI.

Pascal was where I learned all about OOP.

Has anyone here written anything using the library for Pascal called “Turbo Vision” ?  Oh yeah… back in the “MS DOS” days.  Text based GUI.

Pascal was where I learned all about OOP

I’ve got a couple of years of delphi development under my belt.

It is part of learning. We may call it as remembering.

Actually, when we talk about learning for strong-AI, remembering is not the only thing necessary, though most of the learning approaches in academic papers are
remembering essentially.

Human person can judge which knowledge is correct and which is not before remember the knowledge.  I believe it’s necessary for learning too.

Victor Shulist - Mar 10, 2011:

What is learning ?

FACT : John’s phone number is 111-2222
QUESTION : What is John’s phone number?
ANSWER: 111-2222

Most people wouldn’t call that really “learning”.... its too simple, but yet, you are told a fact, and then later presented with a “problem” (finding John’s number) and you use the knowledge you gained to answer.

But it’s too simple…. right ?  So what IS learning… 

What *would* be a good example of learning via NLP ?

Please…. let’s only see examples . . .  . ( no philosophy for now )

For humans, we seem to have a very relaxed definition, if a school boy memorizes his times tables, most would say he learned them.  That seems to be the same as the telephone example, but I bet no one would want to admit to a computer ‘learning’ phone numbers.  Its a funny thing, isn’t it ?

The reason I’m wanting examples only, and no philosophy is that we can avoid overloaded terms (words like “thinking” and “understanding” which no one REALLY knows the meaning of, and we have all different definitions of). ..can we arrive at a kind of OPERATIONAL definition ?

I see different possibilities for learning, in increasing grade of sophistication:

1. learning by being told: learning here is really “being programmed”
2. learning from examples: here the teacher provides examples, and the system tries to generalize those examples into a theory which it can the apply to other examples.
3. learning by discovering: here the system should crawl the net (or use its sensors) and gather and self-organize its findings

After an amount of type 1. learning, we can arrive at type 2. learning (e.g. neural nets, inductive logic programming).
It’s not clear yet how to arrive at type 3. learning. Perhaps enough of type 2. learning can eventually lead to forms type 3. learning?

Important skills in learning seem to be classification and generation: to learn a mushroom is inedible, in principle it suffices to be able to partition the set of mushrooms in two sets: edible mushrooms and inedible mushrooms. Learning by being told is then memorizing a list of edible mushrooms. But when a mushroom does not appear in the list of learned mushrooms, the system is clueless (it could be programmed for safety and reply inedible). Learning from examples on the other hand would consist of the system generating an “internal model” of how to recognize edible/inedible mushrooms, testing that model by predicting edibility of hitherto unknown mushrooms and then being given feedback which is used to refine the internal model again. Note how tasks like “refining internal model” also require classification skills: to discern relevant parameters from irrelevant ones, which allows to keep the model at a finite size. It also requires generation skills: to recognize potentially relevant parameters. By combining type 1. learning with type 2. learning, it’s possible to increase efficiency of type 2. learning by constraining the number of parameters to take into account (to help the system see the forrest for the trees).

This is of course still a very general (vague) description: it is applicable to artificial neural nets, genetic algorithms, inductive logic programming, and most likely also other things I have never heard of.

It even applies to evolution itself. The way evolution seems to learn by “refining its internal model” (according to Darwin) is to generate many possibilities and keep only the best ones. I see some important elements:
1. Life is finite -> it relies on procreation to continue to exist. (I don’t know what mechanism would cause life to want to sustain itself)
2. The way evolution selects the best possibilities is to have mechanisms that removes the bad ones (being unable to procreate, e.g. by dying too fast).
3. Evolution allows for generating possibilities (via random genetic mutations).
4. Generation of new possibilities can be rewarded (i.e. orgasm). This may be optional:  it’s not clear to me if this also applies to plant life

Shi - May 1, 2011:

3. learning by discovering: here the system should crawl the net (or use its sensors) and gather and self-organize its findings

This brought to mind NELL (http://rtw.ml.cmu.edu/rtw/).  ‘“Read the Web” is a research project that attempts to create a computer system that learns over time to read the web.’

2. The way evolution selects the best possibilities is to have mechanisms that removes the bad ones (being unable to procreate, e.g. by dying too fast).

Is removal necessary when resources are not a problem? For example think of the “long tail”: if an algorithm is only useful for less than 1% of input, but you have enough computing resources to run it, you don’t lose anything by keeping it. As a result, your system can progress more efficiently than evolution (which can get rid of useful skills or traits that may be needed to deal with future environments).

For example, the Netflix competition winners wrote:

[...] our ensemble approach was robust enough to protect against some of the problems that arise within the system’s individual components. Indeed, the solution we had just submitted on 1 October 2007 was a linear combination of 107 separate sets of predictions, using many variations on the above themes and different tuning parameters. Even so, the biggest improvements in accuracy came from relatively few methods. The lesson here is that having lots of ways to skin this particular cat can be useful for gaining the incremental improvements needed to win competitions, but practically speaking, excellent systems can be built using just a few well-selected strategies.

Evolution might be practical, but in computer science we need not be so constrained :)

Robert Mitchell - Jun 30, 2011:

... is to have mechanisms that removes the bad ones ...

... Is removal necessary when resources are not a problem? ...

Aren’t resources always a problem? One of the first things I was taught in computer science was that computer maths are different from theoretical maths because precision is finite, because resources (energy, memory and computing power) are finite.

It’s a dilemma really - on one hand one cannot gather enough knowledge (create freedom to combine knowledge in new ways), on the other hand one can spend lots of resources on maintaining cruft (“properly trained and pruned trees will yield high quality fruit much earlier in their lives and live significantly longer”).