Deep Learning · MT2 · guide
MT2 numerical practice from direct to mixed
- Start here
- How to solve a mixed numerical without guessing
- Drill 1 Signs and brackets
- Drill 2 Half sum versus mean
- Drill 3 Natural log and probability
- Drill 4 Dot product and bias
- Drill 5 Matrix rows receive inputs
- Drill 6 Chain rule on a tiny loss
- Drill 7 Sigmoid with binary cross entropy
- Drill 8 Activation contrast
- Drill 9 Softmax and true class loss
- Drill 10 Sigmoid neuron gradient
- Drill 11 Full ReLU backpropagation
- Drill 12 Two output sigmoid backpropagation
- Drill 13 A negative ReLU blocks this path
- Drill 14 Batch and sequential SGD
- Drill 15 Incomplete mini batch
- Drill 16 GD variation 1
- Drill 17 GD variation 2
- Drill 18 Momentum variation 1
- Drill 19 Momentum variation 2
- Drill 20 Nesterov variation 1
- Drill 21 Nesterov variation 2
- Drill 22 AdaGrad variation 1
- Drill 23 AdaGrad variation 2
- Drill 24 RMSProp variation 1
- Drill 25 RMSProp variation 2
- Drill 26 Adam variation 1
- Drill 27 Adam variation 2
- Drill 28 Regularized objective
- Drill 29 L1 update on positive and negative weights
- Drill 30 L2 update
- Drill 31 Inverted dropout
- Drill 32 Convolution with a negative coefficient
- Drill 33 Padding stride and channels
- Drill 34 Max and average pooling
- Drill 35 Complete small CNN
- Drill 36 Reverse a spatial formula
- Drill 37 Bottleneck cost with new values
- Drill 38 Residual addition
- Drill 39 Two RNN steps
- Drill 40 Mixed convolution pool dense and loss
- Drill 41 Mixed backpropagation and L2
- Drill 42 Mixed dropout and a momentum update
- Drill 43 Mixed CNN dimensions softmax and loss
- Drill 44 Inception concatenation and parameter count
- Drill 45 Projection shortcut
- Revising solved examples