Pesquise entre vários professores de Inglês...
Postagens
My journey for studying in AI
The first topic my teacher gave me is how to use pressure sensors under two feet to let robot learning to walk without Gyro. So I kept reading all of the paper which written by my laboratory. After I finished reading, I realized there are two shortcomings in those theories. The first thing is if there are too many states, computation will face the curse of dimensionality. And it will compute very slowly even come out with the wrong result. Second, the robot which uses my lab’s theory won’t generate to other situation. If the robot has learned how to walk on the ground successfully, it won’t walk successfully on the sea saw as well. In order to solve these problems, I search many papers to find the answer to the problem. Finally, I found DeepMind have already solved these problems. They used a big deep network to train the robot to play an old game called Atari and it played well, even better than human in some of the games. Everyone was shocked by the company, even Google. That’s why Google bought this company afterward. They don’t use specific states to train the robot. Instead, they train with raw pixels which generated by the screen. If the size of the screen is 100*100 pixels, there would be ten thousands times three, which contain RGB informations, states which equal to thirty thousands. If we use old version of table method, it would face into the curse of dimensionality. DeepMind used deep network wisely to solve this problem.
0
0
Why am I studying in “Double E”?
Einstein once said “The only important thing is not stop questioning.Curiosity has its own reason for exisiting”. Since I was a child, I have been curious about everything. I was wondering how TV works and how computer works and many other amazing machines. I want to figure out why I just put a button then the TV will show me the show. As a result of the curiosity, I decide to study in Electronic Engineering to find the chamber of secret. After four years training, I start to understand how to build these machines. The most interesting thing is I realize how to write a complicated function into a small chip, and make it out just like TSMC, the most famous company in Taiwan does. I have learned not only the secret I want to know but also many other theories. At the beginning, I felt so bored to study in thoes books. Why I should learn so many theories? And what can it be implemented for the rest of my life? When I keep studying in master degree, I gradually found thoes answers. Those process is to make our brain facilitate. If you can’t figure out even with books and teacher’s class. How can it be possible to make you think a great new theory in the future? And I thought the most important thing is mathmatical justification. Even if you come up with a great idea and have a good result, you still can’t convince everyone. They won’t accept your theory until you prove it through mathmatical justification. This is what scientists do in their research.
3
1
Reinforcement Learning.
The first AI programming I learned is Reinforcement Learning.
It can learn different kinds of tasks just like humans do.
The first book I read about this was written by Richard Sutton who was graduated in MIT.
In the beginning,it was like you read a wordless heavenly book.
Even you recognize every word,you still can't figure out the whole theory.
But after I wrote a program by myself,and ran the program one line by one line.
Suddenly, I realized everything and begin to interest in the AI field.
This is first time I know how to transform bunch of numbers into robotic's thought.
3
0
