[VKumar]

Originally Posted by smartcat

What is Machine Learning? Same as AI?

Although i don't find myself qualified enough to tell that to you, still giving a try:

Machine learning: When the machine learns from it's past experiences. (Predictive analytics book taught me this)

AI: The outcome of machine learning, when the machine knows what to do in any situation based on the past experiences.

My question to experts:
How this experience is used? Heck, don we have any dedicated thread on T-BHP for this analytics, stats, AI, ML etc? I am sure we must be having experts and gurus from this field too ion board.

[tsk1979]

Originally Posted by Samurai

Actually, the all the prerequisites for ML is indeed covered in engineering mathematics. That is basically statistics, calculus, linear algebra & programming. However, it is taught mindlessly without any exposure to their real world application. Therefore, most students never develop any understanding or interest for it.

Fortunately, we had Lenna
https://en.wikipedia.org/wiki/Lenna

I am surprised she is no longer part of the curriculum. Probably against our culture or something now.

[digitalnirvana]

@smartcat that list accurately portrays the desperate situation people like me are in, who are in the last category. There is a lot of emphasis on reskilling however it will likely be too little too late for folks like us

[locusjag]

Originally Posted by smartcat

What is Machine Learning? Same as AI?

From what I've read on e-books on AI and from my projects at work with ML (though I've been a non-techie Business Analyst):

ML in my opinion is all about having algorithms see a pattern so that they 'learn'. There are two broad types of ML, with further algorithms falling below them -

1) Supervised learning - It's a lot like extrapolation. You have mountains of training data which you tag with an expected outcome, to instruct the computer. Once the training is complete and fresh mountains of data are presented to it, the computer sees the pattern of your tagging and thinks, this is what Jagan did in such a case, so here's what I will assign to this piece of data. (Incidentally, this is how self-driving cars have been trained! I'll touch upon this later.)

2) Unsupervised learning - You bring an alien specie to a co-ed school and let it observe the organisms within the school premises. Give it a day and it'll report that there are mainly two types of organisms (what we know as 'male' and 'female') based on height, weight, external appearances, dresses etc. This is a classic case of unsupervised learning and as you'd have gauged by now, such learning mostly culminates in classification of data-sets.

Is ML different from AI?
No. ML versus what is hailed as AI today is like digging in your backyard with a shovel versus industrial drilling for oil. But at the end of the day, ML is AI.

The term 'AI' was coined way back in the 1950s during a scientific conference in New Hampshire and much has been achieved in AI since then, but we have a lot of brouhaha today about AI as if it's a johnny-come-lately. Tesler's theorem was coined exactly for this social phenomenon wherein people discount whatever's been achieved already in the field of AI as not being part of AI. Or in other words, whatever's not been achieved is popularly termed as AI! This is sadly a popular distortion of AI's history and the Tesler's theorem is a sarcastic commentary of this distortion.

What's new is that Graphical Processing Units (mostly manufactured by NVIDIA) have enabled deep learning and this led to a leap in capabilities of self-driving cars, in facial recognition (used by FB, Google photos etc.) and in other areas. Deep learning is nothing but ML on multiple levels, with the output of one layer being used as a feed for the next level, with the cumulative learning being tremendous. And Deep learning is what is being misconstrued today as a sudden advent of AI.

Tesla's cars, by the way, all enjoy AI which is powered by NVIDIA's GPUs. Self-driving cars have been enabled by supervised learning wherein the cars observe the roads and surroundings, all the while noting the inputs/corrections provided by their drivers. Since the data and types of variables are mind-numbing, mere ML wasn't enough and hence deep learning was required to achieve this. The idea to utilize GPUs for deep learning was hence a landmark breakthrough for AI in recent times.

[Samurai]

Originally Posted by ninjatalli

Will have to disagree. In IT oriented (Computer Science/Computer Engg) engineering, Mathematics comes up in the first semester on a generic basis (common to all engg streams); never after that.

In Karnataka engineering syllabus (VTU), engineering math is taught in 4 semesters. This was true even 30 years when I studied in the yearly scheme, I had math all the four years.

Check the syllabus of Math - IV which happens in 4th semester, and it is common to all branches. Check out what is has, module 4 & 5 has the exact statistics pre-req for ML.

Originally Posted by ninjatalli

P.s. It's been close to 17 years so my response is based on outdated information

I interview around 100 fresh engineers every year since 2004, so my info is not at all outdated.

Originally Posted by locusjag

From what I've read on e-books on AI and from my projects at work with ML (though I've been a non-techie Business Analyst):
.
.
.
Is ML different from AI?
No. ML versus AI is like digging in your backyard with a shovel versus drilling for oil.

As a techie businessman, I have to respectfully disagree.

AI is non-human intelligence. An ability to make decision without human input. This can be achieved either through extremely high amount of logical rules, or machine learning. The first method is too tedious, and is rarely deployed except in very narrow AI applications. But the second method (ML) is more popular since it it way more efficient. Deep learning, Neural networking, Bayesian networking, Reinforcement Learning, etc., are some of the machine learning techniques.

[locusjag]

Originally Posted by Samurai

As a techie businessman, I have to respectfully disagree.

AI is non-human intelligence. An ability to make decision without human input. This can be achieved either through extremely high amount of logical rules, or machine learning. The first method is too tedious, and is rarely deployed except in very narrow AI applications. But the second method (ML) is more popular since it it way more efficient. Deep learning, Neural networking, Bayesian networking, Reinforcement Learning, etc., are some of the machine learning techniques.

If you asked me until 2 hours ago, I'd have said that every single AI algorithm is an ML algorithm and all ML algorithms fall under the 2 categories - supervised and unsupervised learning, based on what i've observed and read earlier. However, I just re-read my stuff and discovered that there is a 3rd category of ML algorithms, which is what you'd mentioned - Reinforcement learning. So, I stand corrected, it's that every ML algorithm falls under 3 categories, the third being reinforcement learning.

Deep learning is just multi-layered ML (enabled by GPUs, as I mentioned earlier) and is synonymous with deep neural networks. Bayesian algorithms seem to lend themselves to both supervised & unsupervised learning. Clustering algorithms are a form of unsupervised learning. Regression is the simplest form of supervised learning, and so on.

[NetfreakBombay]

Originally Posted by ninjatalli

Will have to disagree. In IT oriented (Computer Science/Computer Engg) engineering, Mathematics comes up in the first semester on a generic basis (common to all engg streams); never after that.

I did B Tech form Comp Science in early 90s. Math (Specially Discreet and applied math) was part of syllabus till 5th semester.

[locusjag]

Originally Posted by smartcat

What is Machine Learning? Same as AI?

Originally Posted by Samurai

As a techie businessman, I have to respectfully disagree.

AI is non-human intelligence.... (ML) is more popular since it it way more efficient. Deep learning, Neural networking, Bayesian networking, Reinforcement Learning, etc., are some of the machine learning techniques.

I feel like a disoriented nerd who keeps going back and forth now - sorry, but I wanted to add to my prior posts with a crucial bit that I had missed to help understand AI and where ML fits into the AI scheme of things - AI is considered by scientists to be of two kinds,
1) Artificial narrow intelligence (ANI)
2) Artificial general intelligence (AGI)

ANI includes most forms of AI available today, including all these ML algorithms in the job market.

AGI is what is attained when a computer becomes self-aware. To get here, it is understood that we need exascale computing (wherein computers operate at upwards of 10^18 floating point operations per second). When AGI is attained, we would have reached a singularity, beyond which the normal rules wouldn't apply anymore and all our lives will change in unimaginable ways.

[Samurai]

Originally Posted by locusjag

AGI is what is attained when a computer becomes self-aware.

When you say self-aware, most don't realize what that means. I have give a simpler explanation here.

[NetfreakBombay]

Originally Posted by locusjag

1) Artificial narrow intelligence (ANI)
2) Artificial general intelligence (AGI)

All these are a matter of definition. Only thing that matters is, does the thing work for a given project or not.

For example : "Narmada Singh visited Gujarat Narmada Valley Fertilizers in Mumbai"

1. Narmada is not a river
2. Gujarat is not a state

Deciphering these kinds of things requires a program to understand real world as well (E.g.: River cannot visit an office, but a person can).





Similarly this :

[locusjag]

Originally Posted by NetfreakBombay

All these are a matter of definition. Only thing that matters is, does the thing work for a given project or not.

For example : "Narmada Singh visited Gujarat Narmada Valley Fertilizers in Mumbai"

1. Narmada is not a river
2. Gujarat is not a state

Deciphering these kinds of things requires a program to understand real world as well (E.g.: River cannot visit an office, but a person can).

Your example is a great one to illustrate ANI in the particular field of verbal cognition. However, what I've shared is not just a 'matter of definition'. It's about answering Smartcat's query on what ML is and how it differs from AI, and I believe I've done justice to it.

[NetfreakBombay]

Originally Posted by locusjag

It's about answering Smartcat's query on what ML is and how it differs from AI, and I believe I've done justice to it.

My point is not about definition being correct or not.

In practice, only thing that matters is problem being solved to user's satisfaction (or not). This usually means the same project's pipeline will have a mix of all these approaches.

An example : https://ai.googleblog.com/2018/05/du...versation.html

This system uses things from basic supervised regression (ML) to shallow nets for for character embeddings to deep pipelines to impersonate a person's speech (knocking on the doors of AGI).

[locusjag]

Originally Posted by NetfreakBombay

My point is not about definition being correct or not.

In practice, only thing that matters is problem being solved to user's satisfaction (or not). This usually means the same project's pipeline will have a mix of all these approaches.

An example : https://ai.googleblog.com/2018/05/du...versation.html

This system uses things from basic supervised regression (ML) to shallow nets for for character embeddings to deep pipelines to impersonate a person's speech (knocking on the doors of AGI).

Please stop trying to imply that some example of an ML algorithm answers Smartcat's query - it doesn't.

Here's what I'll say with certainty - none of what you've shared is even remotely close to AGI.

The folks who are inching closest to it are working on the Brainome project, the brain mapping project; since the human brain operates at an exascale, computer scientists are trying to replicate a brain-like ANN so as to achieve AGI. It is also widely known that top secret military projects the world over are also on the way to achieving AGI.

It is generally agreed that we need exascale computing to get to AGI whereas we are all restricted to ANI since we only have mostly Gigascale computing which is prevalent all around. Silicon Valley hasn't found a way to make cheap, faster household-scale processors and we are stuck with processors that compute faster than at a few Ghz. No progress has been made for 3 years and Moore's law is on a 'Pause' for now, as far as I understand the situation.

[Samurai]

Originally Posted by locusjag

I get your point.

Relax. I don't see how he has attacked you. He merely challenged your ideas, and that is not a personal attack.

Looking at the exchange, I can only conclude that you both approaching AI from very different directions. You are seeing from a bird's eye view of a business analyst, so you are by going by definitions and classifications. NetfreakBombay is obviously approaching AI as a techie, so he is looking from the trenches. At that level definitions/classification mean nothing.

So let's not turn this into a fight. - Support Team

[ZedMae]

Originally Posted by locusjag

I feel like a---
AGI is what is attained when a computer becomes self-aware. To get here, it is understood that we need exascale computing (wherein computers operate at upwards of 10^18 floating point operations per second). When AGI is attained, we would have reached a singularity, beyond which the normal rules wouldn't apply anymore and all our lives will change in unimaginable ways.

Hi locusjag, Give a shot at the book - Shadows of the Mind by Roger Penrose. The crux of the book is about 4 different viewpoints about the human brain and Artificial Intelligence. In the interest of brevity and in line with the context here - I will just mention two of them.
One is - All thinking (including conscious awareness) of the brain is mere computation. Very similar to how we calculate 2+2. Since it is mere computation, there will be a time when machines with appropriate computing power can entirely simulate our human brain (including - I repeat - being self aware). All the Singularity guys will come under this tenet I guess.

The other viewpoint (2nd of the 4 in the book) is - certain physical activities of the brain trigger or facilitate self awareness and this cannot be simulated computationally irrespective of the quantum of computing power at one's disposal. The author stands by this viewpoint.

He argues for/against all these viewpoints with lot of advanced physics/ maths. I really struggle to read 10 pages an hour

A synopsis of the book's argument vis-a-vis Singularity/ Strong AI can be found here
https://www.linkedin.com/pulse/quant...d-schoeneburg/

cheers