Summary
A summary and review of machine learning for image data, week 4: tips and tricks.
Well done for completing Week 4 of the course.
This week we looked at:
- things to think about when training models, including:
- dataset size – use of performance curves
- splitting datasets – test and validation sets
- cross-validation
- data augmentation which, for images might include:
- flipping
- rotation
- adding noise
- common challenges and how to tackle them, including:
- the curse of dimensionality and dimensionality reduction
- overfitting and underfitting
- class imbalance
- regularisation, e.g. Lasso regression.
Next week, the final week, we will begin looking into deep learning, and how it differs from the machine learning techniques we have discussed so far.
