deep learning

1. concepts

• neuron (input -> output)
• input, weight, bias      u=∑w*x+b
• activation function     y = f(u)
• cost/loss function
• overfitting: statistically correspond too closely

2. type of activation functions

• sigmoid
• relu
• softmax (convert to probabilities that sum up to 1)

3. tensorflow

• tensor: n-dimension vector
• inception v3, a neural network that's pretrained for image classification
• top layer - right most output layer
• bottleneck layer - layer just before the top layer

reference
1. 25 must know concepts
2. quickly setup google’s tensorflow image recognition
3. deep Learning with tensorflow: part 2 — image classification