Skip to content

This notebook investigates whether multiple CNN models can achieve higher classification accuracy than any individual model.

License

Notifications You must be signed in to change notification settings

GSukr/FashionMNIST_Image_Classification

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

13 Commits
 
 
 
 
 
 
 
 
 
 

Repository files navigation

FashionMNIST_Image_Classification

This approach (see notebook) investigates whether multiple CNN models can achieve higher classification accuracy than any individual model.

Two simple strategies for combining models are examined:

  1. Classification based on the average class probabilities of models
  2. Using the mode class for prediction

Conclusions:

  1. A simple CNN can achieve classification accuracy of over 93%
  2. Combining 3 models improves accuracy around 94.4%
  3. It takes around 16 seconds per epoch using Colaboratory GPU accelerator and Test accuracy does not improve significantly after the first 20 epochs
  4. Combining a few more models trained over 20 epochs may further improve classification accuracy in a resonable amount of time.
  5. Classification accuracy is significantly lower for 4 classes: 'T-shirt/top', 'Pullover', 'Coat', and 'Shirt'.

Opportunities for improvement:

  1. Devise alternate methods for combining models
  2. Increase the diversity of constituent models
  3. Introduce regularization methods that prevent over-fitting beyond 20 epochs
  4. Develop a two-phased approach: Predict using a combination of models in the first phase and use a separate model to re-classify examples predicted as 'T-shirt/top', 'Pullover', 'Coat', or 'Shirt

About

This notebook investigates whether multiple CNN models can achieve higher classification accuracy than any individual model.

Topics

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published