Artificial Intelligence and Machine Learning– Explained (for idiots like me)

[Deleted member 118]

Probably one of the easiest articles I’ve read explaining Artificial Intelligence and Machine Learning

Steve Blank Artificial Intelligence and Machine Learning– Explained

Artificial Intelligence is a once-in-a lifetime commercial and defense game changer (download a PDF of this article here) Hundreds of billions in public and private capital is being invested in Art…

steveblank.com

 

[Slymeat]

Probably one of the easiest articles I’ve read explaining Artificial Intelligence and Machine Learning

Steve Blank Artificial Intelligence and Machine Learning– Explained

Artificial Intelligence is a once-in-a lifetime commercial and defense game changer (download a PDF of this article here) Hundreds of billions in public and private capital is being invested in Art…

steveblank.com

Thanks for sharing @Rocket577, that article does contain some relevant stuff.

I believe part of the problem in the term AI is the general misconception people have of the word “Intelligence”.

’Doing something fast’, ‘doing a complex thing’, ‘recalling information’ are not, in and of themselves, intelligence.

I view intelligence as being able to arrive at a reasonable conclusion given a set of circumstances. And I ascribe a higher degree of intelligence when the reasoning is based on more obscure inputs and especially from inputs that have never been encountered before.

In this respect, the following sentence (taken from the article) is the closest, and possibly ONLY, AI that I currently accept:

”SNNs are good at unsupervised learning – e.g. detecting patterns in unlabeled data streams.”

This is as close as it gets to an intelligent person seeing something new, using past experiences with similar things, reasoning to a satisfactory conclusion, and remembering that experience to enhance future decision making.

And isn’t it grand that this is the realm of Akida!

Using elements that mimick neurons and synapses, to build pathways to derive sensible results from inputs that can be quite dissimilar but have some redeeming similar features. Doing this with minimal power. Ignoring distractions and importantly working out when nothing is needed to be done. Now that is pretty much artificial intelligence in my eyes.

The ability to learn in an unsupervised manner is a true measure of intelligence, whether it be just intelligence or Artificial Intelligence.

Almost everything else is just an illusion of intelligence. Not worthless; just not intelligence.

 

[Slymeat]

I stumbled across a couple of quite short, but brilliant, you tube clips that explain Neural Networks and Spiking Neural Networks in quite an easy to follow approach. Amazingly so in fact.

This first one is a 10 minute video that explains Neural Networks and how they learn. It nicely demonstrates how the maths involved is quite simple, there‘s just a lot of it and every neuron gets involved for each input it receives. The large number of multiplications and additions involved is why GPUs have traditionally been involved—GPUs are number crunching machines originally designed for graphics processing, as in displaying pictures on a computer screen.

Why Neural Networks can learn anything (almost)

This next 3 minute video shows a visualisation of a Spiking Neural Network. I think it very nicely demonstrates the concept of sparcity, and how this can save energy/effort. Data seems quite calamitous at the first layer but from there, only activated neurons fire and the data transmitted to subsequent layers is greatly filtered and reduced.

Spiking Neural Network Visualization

I hope this clears up a lot of the confusion behind these concepts.

These videos don‘t cover everything, but they are a great way to get to understand a lot of the seemingly difficult diagrams and mathematical formula that accompany discussions on neural networks.

 

[Slymeat]

I forgot to highlight a critical element involved in a Spiking Neuromorphic process, and that is time.

Artificial Neurons in a conventional NN simply translate state to a rate by multiplying the input with a weight, incorporating an offset, and passing it through a non-linear filter.

Spiking Neuromorphic processes also consider time as an input.

The concept of Long Short Term Memory (LSTM) extends time to include recent history.

And also a SNN does no multiplications. Only a memory lookup when a spike occurs and summations which are computationally orders of magnitude easier than multiplications.

 

[stuart888]

My recommendation for the "Aha Moment" of understanding helps with code samples! This guy DigitalSreeni of Youtube is fantastic, and has about 300 videos covering it all.

It is really hard to grasp AI and ML without understanding the math basics like Linear Regression. He starts with that here, but ends up doing actual coding of a problem in each video.