Deep Learning Tut Note

Gutterberg version

RNN – TF

1. RNN by tensorflow
   The official tut for all RNN: {Google TF RNN}
2. A kaggle IMDB moive reviews example:
   {toward-datascience} {Colab}
3. Explanation and Colab examples on one-to-more, more-to-one, etc.
   {Weight-and-bias}
4. Generate caption for photos (Reference for potentially coming work)
   A combine of CNN feature extraction and RNN text generation:
   {Wandb}
   

Traditional Algo

1. MARS: mulitvariate adaptive regression splines:
   {wiki}
   It automatically sectionizes by splines and then linearly fit each section.
   The spline function, also know as rectified linear unit (ReLU) or hinge function will sectionize the target function.

MAE v.s. MSE

{Medium} says MAE has less penalty to large errors. MSE will increase with the number of samples.