A Survey of Controllable Learning: Methods, Applications, and Challenges in Information Retrieval

Controllable Learning (CL) is a crucial component of trustworthy machine learning that ensures learning models meet predefined targets and adapt to changing requirements without retraining13. It is defined as the ability of a learning system to adapt to various task requirements without requiring retraining, enhancing the reliability and effectiveness of the system. CL is particularly valuable in Information Retrieval (IR) applications, where it enables models to dynamically adjust to different task descriptions, providing personalized and relevant search results without extensive retraining3.

Controllable Learning (CL) adapts to changing requirements in Information Retrieval (IR) systems by allowing learning models to meet predefined targets and adjust to varying task requirements without retraining2. CL techniques, such as user-centric control and platform-mediated control, enable dynamic adjustments to user preferences and environmental changes, ensuring a tailored and effective IR experience3.