Brainstorm

[jfhernandeza]

Brainstorm for the activities of our pipeline. Unsorted Ideas. Any remark about some idea, write it down in the comments section.

• 1. Test with different freezing rates for each one of the models, to empirically find which rate is better. e.g., a freezing rate of 20% means performing transfer learning fixing the first 20% of the layers.

[so77id]

• 2. Data augmentation in most unbalanced classes (test with classical and using GANs).
• 3. Get result of classical networks using transfer learning (Inception, Resnet, Densenet, Inception-Resnet).
• 4. Create ensemble.
     • 1. votation ensemble.
     • 2. Train Fully connected network as a classifier and use the output of pre-trained networks as input.
     • 3. Try with classical classifier like SVM, KNN, etc.
     • 4. Try with another technique in the literature.