IM and Neural Nets

Hey,

(mods, apologies in advance if I've got the section wrong here)

I'm not an active IMer but do love to browse BHW, finally registered an acc. You guys are
a very creative and fairly entertaining bunch. Now, myself, I come from a classical CS background - you can usually find me on HN or similar sites - with a fair bit of theoretical math thrown into the mix. Based on my personal observations it is still very rare for our lines of work to cross paths, which I find rather regretful. I'd even claim that most people involved in CS (especially and paradoxically those involved in the whole startup play) display an active disdain for anything SEO related, etc. Either way, I digress.

On to my question, I'm interested in knowing whether there have been successful applications of neural networks, full intelligence based (or partial recommendation based) systems to the problem domain of SEO or IM at large. Yes, I'm aware of Google developing their machine learning infrastructure, but the application domain is rather different.

I see a lot of potential here and am quite surprised to find such an underdeveloped landscape. I believe that the whole IM industry could be transformed, like finance and many others before it. For example, I see potential applications transforming and generating human readable content, recommendation systems advising on a link building strategy, etc, etc

I'd like to start building an open-source general feed forward neural network system and slowly evolve it to potentially tackle various problem domains within IM. Has anybody worked with intelligent systems over here? Would there be interest in this type of thing? Could I use this space as a type of proto-blog, share my findings and progress and get your feedback?

Cheers everybody

Very interesting premise you have there: it could be quite useful.
However, you would probably want to start a thread in the My Journey's section if you're going to share your progress, get feedback etc.

I for one would be delighted to read such a thread.

uristjack said:

Very interesting premise you have there: it could be quite useful.
However, you would probably want to start a thread in the My Journey's section if you're going to share your progress, get feedback etc.

I for one would be delighted to read such a thread.

Click to expand...

Ignore the morons op. Ive always pondered this myself and always wanted to get into ai myself. Combined with I'm it can be an incredibly dangerous money making machine. Interested to hear what you have to offer. Keep in touch. My inbox is open.

Sorry everybody, been quite busy at work, finally got some time to respond to my own thread.

Ignore the morons op. Ive always pondered this myself and always wanted to get into ai myself. Combined with I'm it can be an incredibly dangerous money making machine. Interested to hear what you have to offer. Keep in touch. My inbox is open.

Click to expand...

Absolutely agree with you, it can and will be an absolutely revolutionary device. Don't have enough posts to PM you yet, maybe you can drop me a PM instead?

So, where do we start?

Needles to say, the task at hand is somewhat complex. I think it would serve the thread well if everybody had at least a general understanding of what neural networks (NNs hereafter) are, what is understood by genetic or evolutionary programming, how NNs can be applied to complex problems and
what are some current uses of intelligent algorithms.

As I said in the OP, I'm not a marketer and thus while I know how to write and implement genetic code, I fear I'll mess up (lightly speaking)
the training part as I myself am pretty much a newb when it comes to the finer points of SEO, IM, etc. Naturally I'm a lazy person
and thats the reason why I'd like to do this publicly and on BHW out of all places, I simply don't have the time to learn SEO and whatever else is
required. Traditionally better outlets exists, yet I believe BHW will be instrumental during the learning phase of the algo. Anyhow.

WTF are we building?

Easy, we need an algorithm that can translate gained experiences into knowledge. Better yet, we need a system which we can easily modify,
adjust, train, transfer, etc. Now nature has already bestowed us with said algorithm, through trillions of iterations we have evolved a
general algorithm that is capable of translating our experiences, turning them into knowledge which we can act upon. This algorithm is housed in
our own carbon based cognitive computer we call the "brain".

As cliche as it might sound, the brain is an incredibly limited and incredibly powerful machine at the same time. For example,
biological neuron to neuron signalling (the internal synchronisation of information)
goes through an electrical->chemical->electrical encoding process which is inefficient. Being a biological machine, it is incredibly hard to evolve outside of bounds setup by natural selection.

It has rather limited peak performance, capacity, etc.
Further on, like anything carbon based it eventually falls into decay.

Yet, through trillions of iterations we have evolved the only known, general computing device capable of the
experience->knowledge transfer process. The brain is also incredibly fault tolerant (people who've done mushrooms before will agree), efficiently parallelised, robust and plastic.

From an abstract point of view, neuroevolutionary programming, NNs, genetic programming, etc all focus on mapping
the biological neuroevolutionary process to a software representation. In other words, when constructing NNs we are simply encoding (to the best of our knowledge and ability) what has already been demonstrated to work by nature.

Now of course our resources are limited.

We are also lazy by definition. Crucially we lack time. So when talking about NNs nobody is talking
about mapping the brain 1:1 (although its not such a bad idea and resaerch in this area is certainly serious and ongoing).
Thus what we have to do is create software neurons which can in turn be connected to software or hardware sensors and actuators.

From wikipedia:

Neurons are highly specialized for the processing and transmission of cellular signals. Given the diversity of functions performed by neurons in different parts of the nervous system, there is, as expected, a wide variety in the shape, size, and electrochemical properties of neurons.

Click to expand...

A typical software neural network consists of an interconnected net of neurons (just like the brain), small objects or processes able to processes bits and pieces of information.

This net is typically connected to sensors (a webcam for example) and actuators (e.g. a robotic arm). Once launched a neural network goes through a process of plasticity and iteration. We give it a task, for example, find the best possible SEO strategy for site A.

We give it a task, for example, find the best possible SEO strategy for site A.
Next, we let it gather, or directly feed it, information through its sensors. It could for example launch a scraper and check the site's stats, gather its current rank with a search engine, pick up some analytics data, etc. This information is processed by our NN and finally acted upon through an "actuator". The actuator here can be a simple monitor or a process able to adjust our site accordingly.

Next, results are evaluated by an evaluator process. Based on its output, our learning strategy, fitness scores of individual neurons, etc we kill of a certain population % within our neural net and make "genetic copies" of successful neurons. The whole process is repeated anew. Eventually we will reach a local fitness score that will satisfy our conditions and the neural network can be deployed on site B, C, D and whatever other site we can think of.

This is of course a very basic overview but the same process is applied in many industries to develop intelligent agents. NNs are currently evaluating financial scenarios, trading on the Forex market, producing more efficient encoding algorithms, evaluating political scenarios, learning how to recognise cats from human faces (haha), etc, etc

Well... This has been quite a massive post and I hope I managed to provide at least a general overview of what an NN actually is and what it does. Honestly, I could talk about this for hours and I haven't even touched upon the endgame. The potential singularity. However, its probably not the right time for endless discussion yet. I know you guys are results oriented (as you should be in your profession). I'll need to get you excited first.

I'll be happy to answer any questions and am always happy to discuss and explore potential applications of NNs in the IM domain.

Next post, I'll start building our first, very basic NN and publish it to github.

All the best everyone!

Are you thinking of something like this but much more complex? If you need help with statistical analysis I am your man, lel.

Haha this is exactly it and reminds me of a pacman solver I wrote eons ago (gonna try to find the code, have it lying around somewhere).

Great video. Demonstrates what I've tried to describe in previous post. No less and no more complex really. The process is generally the same. You can see that it employs sensors (haven't watched in full, but probably a sliding window capture), actuators (hardware or software emulator controls) and of course the actual weight based NN. Mario "learns" how to beat the level through a constant process of iteration/adjustment of NN populations.

I'll hit you up man. Drop me a PM please.

This would be super awesome to make haha, something like some automated SEO software that learns its mistakes and next time it would somehow become semi-perfect?

If that could be possible, this would make you a trillionaire most likely haha.

OP, if are starting a Journey then ask a Mod to move the thread to this section:
http://www.blackhatworld.com/blackhat-seo/f167-my-journey-discussions/

Also there are already applications using Neural Nets within the Marketing industry
https://en.wikipedia.org/wiki/Marketing_and_artificial_intelligence
http://lmgtfy.com/?q=Neural+Net+Marketing

W130SN said:

OP, if are starting a Journey then ask a Mod to move the thread to this section:
http://www.blackhatworld.com/blackhat-seo/f167-my-journey-discussions/

Also there are already applications using Neural Nets within the Marketing industry
https://en.wikipedia.org/wiki/Marketing_and_artificial_intelligence
http://lmgtfy.com/?q=Neural+Net+Marketing

Click to expand...

Of course there are applications, but mostly they are using some sort of statistical analysis mixed with predictions. Just imagine some SEO software like GSA SER with this. Limitless possiblites.

dgusic said:

Of course there are applications, but mostly they are using some sort of statistical analysis mixed with predictions. Just imagine some SEO software like GSA SER with this. Limitless possiblites.

Click to expand...

Yes, but how many years development will it take?
Or can OP upgrade the JrVIP Paypal Money Generator in time for the new forum launch?

Should probably be moved to "coding"? How do I contact the powers that are over here?
Not sure what a "Journey" is tbh... This sure ain't no journey son its a state of mind.

That article you linked is pretty worthless tbh.
In your defence, my OP was rather vague. I meant IM specifically, as well as full NN based systems.
As in something akin to DXNN. Something that makes use of benchmarkers, full neural plasticity (through neuromodulation), etc.
If you read through what you linked there none of it (data mining, partial expert systems crucially lacking actuators, mechanisation, etc) qualify as full intelligent systems. And then again, wider marketing is not really something I'm interested in right now.
The "let me google that for you" link is cute though, I give you that.

W130SN said:

Yes, but how many years development will it take?
Or can OP upgrade the JrVIP Paypal Money Generator in time for the new forum launch?

Click to expand...

Good idea + good pitch + good luck + VC funds = TRILLIONS OF DOLLARS IN POCKET

Calm boys. Nobody is talking about an SEO supermachine lol.
Not sure where you all are getting that notion from. Gotta start light and simple.

What sort of system are you thinking of, then? If it's not some SEO supermachine .

Honestly though, I think neural nets would be incredibly powerful in SEO. I'm surprised no-one's tried it before. (Actually I lie - Google has )
Really looking forward to the results. Best of luck!

P.S. Imagine some plugin which would automatically perform on-board SEO, even able to write human-readable content. Now that would be a SEO supermachine.