The autonomous lab is described in the following slides:

• Slides.

Follows a short list of freely available datasets to play with. You can find many more online.

• mit67
• cub200
• flowers102
• sdogs
• caltech101
• food101
• mit67
• textures
• wood

Tips

• Properly introduce the data you work with
• Properly split the data into train, test and validation (if needed)
• Don’t use datasets for which achieving a 95% of accuracy is straight-forward (NO MNIST)
• Use and show loss and/or accuracy plots in your discussion
• Apply techniques (e.g., regularization, data augmentation) for a reason and properly motivate it. Dont do it just for the sake of doing stuff.
• Dont try to use all possible methods and techniques. Focus on certain aspects, try to understand and interpret their behavior
• When using a plot, stare at it for a while. Try to reason what can be understood from it. Make yourself related questions and act in consequence.
• To properly analyze and understand a model, train until overfit. Visualize both accuracy and loss curves.