In the context of § Machine learning, meta-learning allows a learning system to acquire biases through training, instead of biases being explicitly engineered. This is important because according to the bias-variance trade-off, implicit biases are needed for data-efficient learning.