Skip to content

7月11号 机器学习（Machine Learning)#

Why ML in an HPC course?#

  • ML(especially DL) is a special serious of application requiring large computing power.
  • ML can be used to guide system optimization
  • Emerging research topic : MLsys()

Preview#

  • Introduction

  • Basic of Machine Learning

  • Typical Network Architectures

CNN

Attention

  • DL Hardware & Software
  • MLSys

ML#

  1. Machine Learning Problems(classification / Regression / Clustering / Dimensionality reduction)
  2. Learning method((un)Supervised learning / transfer learning / reinforcement learning)
  3. Gradient Descent, and learning rate (just like 步长) (Stochastic随机)
  4. About fit.

DL#

  1. Multilayer Perceptions(Input Hidden Output)

  2. Forward / Back Propagation(follow chain rules)

Preview

  1. Layers(Activation Function Loss Function Regularization Dropout Normalization)

  2. Optimizers

  3. Problems in DL(Gradient Vanishment Overfitting Fitting Non‐linear Functions )

  1. Activation Function(sigmoid(容易梯度消失) / tanh / Rectified Linear Unit（可避免梯度消失） )

  1. Loss Function(•Cross Entropy / Mean Square Error)
  2. Regularization(正则化)

  3. Dropout

  4. Batch Normalization(skip math)

  5. Optimizer(link)

SGD; SGD + Momentum; AdaGrad; Adam

CNN#

学长讲了不少，但是不太想记了。主要听听就好

Last update: 2024年1月28日 13:01:36
Created: 2023年7月11日 17:42:57