[ML/DL study note] 1-1 History, AGI/StrongAI, ML, DL

Study Source	혼자 공부하는 머신러닝/딥러닝 Chaper 01-1
Study Date	2021/10/30

History of AI

image from 박해선 - 혼자 공부하는 머신러닝+딥러닝 유튜브 강좌

Q. Why First winter has came?

A. So many are there, but important: Lack of computer performance, Moravec's paradox, ...etc. With many reasons, it consequently made funding stop and the... ❄️ Winter has come ❄️

image from Game of thrones

After that, Expert System has developed, second AI Boom, also second AI winter :(

AGI/Strong AI and Week AI

I don't think that i should explane this too much.

We always watched so many AIs in the movie,

most of them are AGI(Artificial general intelligence) or just Strong AI.

It means the computer system that have brains that can be confused with humans own.

But we mostly use, in common is Weak AI.

Such as Personal Assistance, Autonomous Driving, Recommend System... etc

Most of us hope AGI could be here soon, but we should aware of conceit.

We already experienced winters 'cause of this attitude.

Then what is Machine Learning?

Machine Learning (just shortly ML)

is a field that study about the algorithm to find rules without teaching.

ML is deeply related with statistics, and so many algorithms were developed with them.

But nowadays, vary part of ML's development is based on experiences.

Such as Scikit-Learn, TensorFlow, PyTorch.

The book says Scikit-Learn is most represents one,

but Google AI, leading company, is based on tensorflow.

I'm not working in frontmost line of this part, so do not 100% believe my words..
it could be uncorrect especially in this part.

Overline part: I misunderstanded the book. Tensorflow is Open Source DL Library!

It is written in backyard of this part. My bad 🙄

What I want to say is that we have Open Source Library, and algorithms also.

It means we can easily use and adapt in one's product.

image from ICHI.PRO - The best/recent Python Library for ML&DL. (korean)

Deep Learning..? wasn't it just same with ML?

Deep Learning (I'll just call this DL.) is smaller part in ML algorithms.

A ML algorithm that based on Aritificial neural network(ANN), is what we call Deep Learning.

In 1998, Yann Lecun has succeeded to cognize pen writings by ANN.

This is LeNet-5, the first Convolutional Neural Network(CNN). We'll learn this at Ch.8!

In 2012, Geoffrey Hinton's Team won the ImageNet (image classification contest) with overwhelming performance.

It's name is AlexNet, and also used Convolutional Neural Network (CNN).

After this, most of other image classification algorithms used CNN.

For Koreans, Alpha-Go is reallllllly popular to public 'cause of game with Lee-Se-Dol (Korean 바둑 Player).

As you can see here, DL has been preparing for a huge revolutions since early 2010.

What is the reason? What is the motor of this happening?

Well, there are 3 of them.

   - A rich of data

   - Improve of computer performance

   - Innovative algorithms

Maybe beacuse they smelled influences of this Open Source ML Library?

In 2015, Google announced Tensorflow. The most widely used DL Library.

In 2018, Facebook announced there own one, PyTorch.

What they have in common is that they specialize in dealing with ANN, and offers Python API that is easy to use.

---

Well, it must finish here.

Hope to see you again in next post!👏

If you enjoyed or found some errors here,
just contact me in comment, 42 network Slack - kkim, or kwanho0096@gmail.com.