noun
A contextual bandit is an AI model that balances exploration and exploitation to optimize decisions in real time. It learns by testing different options and adjusting based on feedback, like a slot machine that adapts to your preferences.
It works by using past interactions to predict the best action for a given situation, while occasionally trying new options to discover better ones. This approach is widely used in recommendation systems, ad targeting, and personalization.
For developers, it changes how they design feedback loops—requiring real-time data collection and adaptive algorithms. It also reduces reliance on static rules, but demands careful monitoring to avoid reinforcing biases. For example, a streaming service might use a contextual bandit to suggest videos, improving user engagement by 15% over traditional methods.
A shopping app uses a contextual bandit to decide which discount to offer next, learning from past user responses.
In other languages
See also: reinforcement learning