General Methods

Statistical Preprocessing

<aside> ⚠️ Statistics must be computed on the training data then applied to validation & test sets

example:

❌ zero center using the mean from the entire dataset

✅ zero center using the training set mean

</aside>

This prevents data leakage: revealing information about “true” statistics from the test set

📝 Normalization

📝 Mean Subtraction

📝 PCA

Image Data

• TODO

Text Data

Representations

📝 Word Embeddings