Machine Learning Coursera By Andrew Ng: Week 1-Part 1 — My Journey
This series will be used as a study guide for me or any student taking this class. Basically, a shareable note.
The above link will take you to the course in Coursera. It is free to join.
Let’s get started:
What is Machine Learning?
Machine learning is a subset of AI. And what you think of when somebody mentions AI?
Now I know there is a lot of progress being made in Machine Learning but in true AI, there’s virtually no progress yet! Don’t get your hopes up and don’t expect anything soon. That being said, if there’s a breakthrough in Artificial General Intelligence (AGI), we will have our real-life Vision or really anything and everything that is bounded by physics.
Machine Learning’s definition:
“A computer program is said to learn from experience E with respect to some class of tasks T and performance measure P if its performance at tasks in T, as measured by P, improves with experience E.” by Tom M. Mitchell
Think of an athlete doing Hammer Throw, the better he/she trains that muscle and techniques, the farther away that hammer will go.
Now, that same athlete can’t expect to train his/her muscle and techniques and expect to win at a Chess Tournament for example. The training and the task must be correlated to each other to gain better performance.
A few examples of Machine Learning include:
Database Mining:
Web click Data — essentially companies are trying to figure out how a user is navigating through a website by using machine learning. This will help the companies to try and figure out how to serve and “guide” the users better. — More targeted ads, I’m sure. Urgh.
Medical Records — The professor talks about how if we can turn medical records, which we have a lot of them, into medical knowledge then imagine that possibilities. I think this scenario, what it means is that by using machine learning to dissect medical records, we can find patterns that we, as humans, have not seen before.
Biology — By using machine learning, we can better understand the human genome. Again, possibilities are endless. In a couple of hundred years, we might finally have our own unicorn.
Engineering — As dataset grows, we can possibly predict whether certain structure may collapse or hold example.
Applications That Can’t Be Program By Hand:
Autonomous helicopter — The awesome people at Stanford achieved this in 2008. They made the program “watches” a person flew the helicopter and after a while, that helicopter “learns” how to fly as well. One can also achieve this by “Reinforcement Learning”. Video below.
https://www.youtube.com/watch?v=M-QUkgk3HyE
Handwriting recognition — Everybody styles their writings very differently. Cursive, capitals, and all that. The only way that we can differentiate between millions and millions of different characters is by having a computer do that for us.
Natural Language Processing(NLP, Computer Vision — The ability for a program to “understand” the languages and “see” the images is one of the well-known problems that machine learning can solve.
Self-customizing programs — With millions and potentially billions of users, each of them have their hobby, jobs, friends, etc… We, as programmers and problem solvers, know that the task is futile without machine learning.
