Course By James Zhou

L4

RNN

limitations of CNN

• fixed input length

LSTM: Long short term memory; a variant of RNN

• easier to retain longrange interactions.
• LSTM application:
  • ehancer/TF prediction
  • ONT base calling

L5 Auto encoder

L6

regularization and optimization for deep learning

• Regularization—prevent overfitting

• • Early stopping
• • L2 regularization (aka weight decay)
• • Multi-task learning; data augmentation
• • Dropout

• Optimization—overcome underfitting

• • SGD, SGD with momentum
• • RMSProp

L9

DL: great power, but poor interpretability

• find important features in individual input examples
  • perturbation 不安；扰乱
  • backpropagate contribution
  • deconvolutional nets

L13 Protein Structure prediction

• amino acid residues
• polypeptides
• Structure level
  • primary: AA seq
  • secondary: local
    • alpha helix, beta sheet ...
  • Tertiary:
  • Quaternary: