In many scientific applications, we observe a system in different conditions in which its components may change, rather than in isolation.

The goal is to train a good instance classifier.

Algorithmic stability is a major theme in learning theory, where seminal results have firmly established its close relationship with generalization. Recent research has further highlighted the intricate interplay between stability and additional properties of interest beyond statistical generalization.

For learning Boolean concept classes, we show that the entangled and separable sample complexity are polynomially related .

But nonetheless, they tend to perform well on unseen test data.

While contextual bandit has a mature theory, effectively leveraging different feedback patterns to enhance the pace of learning remains unclear.