Algorithmic curation and model training amplify data bias

Algorithmic approaches for data selection, curation, and model training (such as generating or clustering sequences/structures) can amplify existing sampling biases, further entrenching representation gaps.

Part of Chain