Artificial intelligence : where to begin

[!kx!]

Hi

I am completely new to the area of AI. Can someone direct me where to begin on the subject?

Artificial Neural Networks also come under AI? What if I start with that?
Can someone suggest some project that I can pursue in ANN, as a beginner...?

 

[maverick_starstrider]

Well it's kinda hard to direct if I don't know how much CS/scientific computation you know. Do you want to do support-vector-machine machine learning, or genetic programming or what?

 

[!kx!]

Well, I am an undergrad in Engineering physics... and I don't know much of computation. But, I would like to learn as much as possible during the process.
I am quite good at programming... And as I am just trying to get introduced to the topic, so any type of the project may be helpful (I suppose so...).

 

[Luke1294]

https://www.amazon.com/dp/1558604677/?tag=pfamazon01-20

This book covers a wide range of AI/MI topics, from a slightly more engineering-based perspective- a big theme throughout a majority of the book is autonomous mobile agents. It's a typical survey book in that it treats each subject very lightly, but does give a feel for a wide variety of topics.

Although the book is 11 years old, I highly recommend it.

 

[!kx!]

thanks... the book is fine... but I think going about the subject through some project-work will be easier and faster way to get some feel of it...

 

[maverick_starstrider]

Well, to arbitrarily hand out a coding assignment... Why don't you try designing a genetic programming algorithm that solves the traveling salesman problem. Or a high pass filter by combining circuit elements.

 

[maverick_starstrider]

I cut my teach in AI (which I now, naturally, do none of in my graduate thesis) by helping with computer vision algorithms in MRI imaging. Those are algorithms you can do in just MATLAB.

 

[!kx!]

Thanks a lot...
I will read more about these topics. It may be a good starting point...

 

[mal4mac]

On neural networks (from Microsoft UK's head of research):

http://research.microsoft.com/en-us/um/people/cmbishop/books.htm

 

[rogerharris]

Hi

I am completely new to the area of AI. Can someone direct me where to begin on the subject?

Artificial Neural Networks also come under AI? What if I start with that?
Can someone suggest some project that I can pursue in ANN, as a beginner...?

If you have not done so yet, I would definetlely be reading a summary at the very least of how the brain is thought to work. Otherwise how can you have a real context of what it is they are trying to model ? Ok the Ai introduction books will tell you what bits they tried to model in the past, but i would bet they could be pretty out of date.

the top guns in Ai think their field is at last going to merge with neuroscience in the next decade. Previously the Ai was mostly taking titbits of neural processing, and thinking that these bits would lead to machine intelligence and building computers like this, could just leave the human brain standing as a relic.

None of these projects really did this, so instead of more headscratching, the main players and big money are going back to neuroscience again and saying, ok let's transcribe what is known about the entire brains operation this time, and see if we can model all of it.

 

[rabbitweed]

"..there is little chance that any deep understand of the nature of the mind can come about without our first learning much more about the very basis of physical reality" - Roger Penrose.

Utterly useless in regards to your question, but thought I'd throw that out there:D

 

[maverick_starstrider]

"..there is little chance that any deep understand of the nature of the mind can come about without our first learning much more about the very basis of physical reality" - Roger Penrose.

Utterly useless in regards to your question, but thought I'd throw that out there:D

I actually just commented on this on another post earlier today. From a computational perspective, if one wants to model the brain, numerical stability and round off error (i.e. chaost theory) will dominate basically any simulation LONG before quantum vs. classical error shows its ugly face. Quantum systems as small as 2 particles exhibit decoherence. Even if it was a significant effect it would not be a priority from the computation/simulation perspective.

 

[rabbitweed]

I actually just commented on this on another post earlier today. From a computational perspective, if one wants to model the brain, numerical stability and round off error (i.e. chaost theory) will dominate basically any simulation LONG before quantum vs. classical error shows its ugly face. Quantum systems as small as 2 particles exhibit decoherence. Even if it was a significant effect it would not be a priority from the computation/simulation perspective.

Link the other post please
I'll happily admit I know almost nothing of Neuroscience...though it is food for thought:)

 

[maverick_starstrider]

https://www.physicsforums.com/showthread.php?t=326255 I'm certainly not denying the possibility of quantum effects having an effect on our brain. I'm just pointing out that when we actually model these things on computers (or experimentally) we get lampooned by other sources of error long before quantum effects matter. Plus, as systems grow in size quantum effects DECREASE and chaotic effects INCREASE.

 

[rogerharris]

Although the brain does work in terms of physics we know of, noise based processing and EEG is obviously an electromagnetic field etc...most of that quantum physics neuroscience stuff was an unsuccesfull attempt to inject new age spiritual philosophy into science. Lot of interesting stuff did come out of the attempts though.

The Ai guys are currently very keen to emulate the known physics of neuronal structures, as can be seen to the lengths they have gone to create physical models for the Blue brain project, and have invented all kinds of new patch clamp technology to record cortical collumn activity.

So yes IMHO to keep up with Ai in the next decade you are goingto to have to understand some systems neuroscience as well as accepted neurophysics.

 

[sin2beta]

Well, my advice is minor as I am just starting out, but I was a math and physics undergrad and am now a grad student in Machine Vision/AI. This is what I have learned on my first year for intro material. Although, keep in mind, most of my work is very mathematical and not necessarily what you will see in most AI research (I do a lot of curve matching). One thing I have learned is that there is A LOT of A.I. fields and there are very different things you will need to know for each of the fields. Here are the things that I have found interesting and useful:

Books
--------------
Structure and Interpretation of Computer Programming by Abelson and Sussman (even if you don't use scheme or lisp, it will help you think)
Artificial Intelligence: Norvig (good intro)
Algorithms: Cormen
Anything by Knuth (beware! This stuff is insanely crazy for someone starting out. But the dry humor makes it easy to read none the less. He rates problems by how hard they are. A level 4 (or possibly 5, I forget) are research problems that he warns can take a while. The first level 4 problem was Fermat's last Theorem. I'm hoping that was a joke.)

Lectures
--------------
Lectures for computer science are everywhere and open. It is a good field to find free lectures and textbooks. I'm sure you can find them other places, but iTunes U has most of these.
MIT:
a) SICP lectures by Sussman and Abelson (old but still good, these are on the web but not iTunes
b) Algorithms (on iTunes)
Stanford
a) 106a (if you know a lot can probably be skipped. But if you don't have formal training in programming. I'd still go through it to learn about style a bit better)
b) 106b (beginning abstractions. Coming from physics you will probably need this. I know I did)
c) 107 (awesome!)
d) Machine Learning (I forget the number, but I'm sure this is what you want.)
Berkeley:
The freshman sequence is on iTunes. Scheme, Data Structures and Architecture

Interesting people to look up on youtube or Video.google are Gerald Sussman, Marvin Minsky, John McCarthy, Donald Knuth, and the company Novamente had a good Google talk on artificial general intelligence. There wasn't a lot of substantial information, but it was entertaining.

But the way that I would start is with SICP, and the lectures that go with it (Berkeley's first course, Stanfords 106a/b and MIT's SICP lectures all together. You will come out with familiarity in scheme, C, and Java, which is good) Then I would move on to data structures and algorithms. You may think you are a good programmer, but these courses will help make sure that you are. Best of all, SICP is free online. So, you can't go wrong starting with that. Plus, you get free lectures that go with it.

http://mitpress.mit.edu/sicp/

And since you are a physicist (structure and interpretation of classical mechanics)
http://mitpress.mit.edu/SICM/

But the best advice on how to start I can give is to have fun!

 

[sin2beta]

Oh, and in terms of projects, this is just difficult to tell you. As you start reading a book and learning on your own, you will find stuff that interests you. When you do, try and build a simple version. That's the only thing I can tell people no matter if they are starting or experienced.

Since you are starting, I would still stick with SICP and do some of the homework problems. Scheme/Lisp and Prolog are good languages to learn for artificial intelligence. You will find a lot of source code if you dig around.

Find a university's AI class website. They usually have projects. Do those.

Look at research papers. arXiv has a section on neural and evolutionary programming. http://arxiv.org/list/cs.NE/recent

But I would like to stress that having a project is nice, but do that while studying the fundamentals. It will save you a lot of grief, provide new ideas, and make your work much much better.

 

[!kx!]

Thanks for all that information...

That was really helpful! I will try to keep all that in mind!