Optimization
For some basic knowledge you can see Deep Learning.
Adaptive Learning Rate¶
Saddle points can be very challenging -- Saddles
- To detect saddle points
No.1 Overfit you training data -- you model has the flexibility to get there.
Regularization¶
- Parameter Regularization
- \(\ell_1\) forces some variables to be zero to preserve sparsity.
- Maximum a posteriori (MAP) estimation
- Structural Regularization
- Go for simpler models and slowly go more and more complex.
- User tasks specific models
Co-adaptation¶
- Dropout: multiply the output of a hidden layer with a mask of 0s and 1s
- Backward: multiply the weights by \(1-p_i\).
- Stop co-adaptation and learn ensemble
- Other variations
- Gaussian dropout: multiply with a Gaussian with mean 1
- Swapout: Allow skip connections to happen (?)
Multimodal optimization¶
- Biggest challenge
- Pretraining
最后更新:
2024-04-09