Using the game Minecraft to Understand Deep NN

Course Q&A Deep Learning Specialization Improving Deep Neural Networks: Hyperparameter tun
1

So in the youtube link: My Computer Taught Itself to Play Minecraft - YouTube. I saw very interesting about how Deep Neural Networks are able to make AI smarter. There’s something that i didn’t really understand though. So it says that we received inputs like the rarest ore which is Dirt, Stone, and Iron Ore. Then the output will be the movement, like moving right, left, front, and backward. I want to summarize how it works, and please correct me.

So it received inputs, and since this is a multi-class, we’ll use Softmax Regression. So it says that based on the weights, there are probabilities in the output. And i’m assuming that this youtuber used Hard Max since, when i see that, every time the weight initialize randomly, it only do certain output. Like there are 7 units in the output layer, but only 2 used. So does my summary seems to be true? If not, please do tell me.

Oh and also one more thing is that, i forgot that i’ve learned in this course, but how exactly that the NN knows the Y answer? Thanks.

2

Hi @CourseraFan

The type of machine learning that the youtuber used is called ‘reinforcement learning’, as stated at the very start of the video.

This is a type of machine learning that does use deep learning but is not taught in this specialisation.

If you want to learn more about deep reinforcement learning I suggest you use this:
“Reinforcement Learning Lecture Series 2021”

I have yet to watch those lectures, but they have been made by top AI researchers at a top AI company called Deepmind

Once you understand those lectures you should easily understand how that youtuber’s computer learned to play minecraft

3

Oh ok, thanks for your respond!

1 Like