网络结构设计时的结构要点
2019-02-16 本文已影响0人
提着木剑走天下
- How did I preprocess my data? How will that affect the model?
- How to choose optimizer? Evaluate its pros and cons.
- Overfitting and Underfitting:
- if overfitting, adding dropout or max pooling can help. You can report can describe how dropout or max pooling improved the model.
- if underfitting, you can report how adding more convolution or fully connected layers helped.
- How to decide the number and type of layers?
- How to tune the hyperparameters? Sound knowledge of Linear Algebra, Probability & Statistics and Multivariate Calculus can help you a lot in narrowing down your choices.
- How to train and evaluate the model? What is the metric? How do I set the benchmark?
