1. Machine learning

• 1. Machine learning
  • 1.1. Chapter
    • 1.1.1. Weight
    • 1.1.2. Bias
    • 1.1.3. Cost
  • 1.2. Gradient
    • 1.2.1. minimum of a f(x)
    • 1.2.2. Gradeint
    • 1.2.3. michele nilsen neural networks website
    • 1.2.4. paper
    • 1.2.5. aim is to reduce the cost function
  • 1.3. Backpropagation
  • 1.4. MNIST DATABASE Training materials

1.1. Chapter

• different level have diff orintaion
• split the character and number into many components .

Base on databases wich have thounsansds of number and graph match which teach them .

1.1.1. Weight

1.1.2. Bias

1.1.3. Cost

How good is a network

1.2. Gradient

1.2.1. minimum of a f(x)

• local minumusm
  • caculter the 斜率 to the lower place . test again till the gradient become 0
  • doable
• Global minimum
  • drop several of balls to firgure out the minumum and compare them.
  • crazy hard

1.2.2. Gradeint

“This solve my consider about how to find out the good weight”

• the direction to deep
  • Compute $\nabla$C add a negative before it -$\nabla$C small step add repeat

    “this is what need to add and reduce every times”

  • like W0 should increase 0.1 W2should decrese 0.2

1.2.3. michele nilsen neural networks website

1.2.4. paper

• understanding deep learning requires re-thinking generalization
• A closer lokk at memorization in deep networks

1.2.5. aim is to reduce the cost function

• use -$\nabla$C(weithts and biases) to reduce C()
• The methods mentioned above is an old way of doing this

1.3. Backpropagation

1.4. MNIST DATABASE Training materials