2. Why machine learning?
● Cool!
● Solve things which can be solved by human master.
● Discover things which can not be discovered by human master.
● Hail Terminator!
3. Things will be covered:
● Machine learning and CIFAR-10.
● Simple math.
● A shallow deep neural network.
● Data flow inside the neural network.
● Inside the neural network.
● Machine learning terms.
● Neural network zoo.
● Applications.
● Q&A
6. Reinforcement Learning
We can act base on states to
gain rewards. Find the best
actions to get the best result.
https://gym.openai.com/envs/LunarLander-v2
7. Unsupervised Learning
Y = f(X)
We know X. We want to find f to
describe X.
https://github.com/Newmu/dcgan_code
9. Supervised Learning
CIFAR-10
● 32 x 32 x 3 color images.
● 10 classes.
● 50000 training images.
● 10000 test images.
http://www.cs.toronto.edu/~kriz/cifar.html
34. The First Gate of the Shallow NN
W1
of a neural network for CIFAR-10
● shape: 3072 x 100
● sub image: 32 x 32 x 3
● a subimage is a column vector of W1
39. Date sets
● Training set: used to train the model.
● Test set: used to adjust the model during training.
● Validation set: check the accuracy of trained model.
Do Not Blend Them!
40. Batch
Select a few random data for each traning setp because:
● We do not have that much memories!
● We do not have that much computing powers!