Learning to Assign Prediction Tasks to Agents with Capacity Constraints
A new paper on arXiv addresses the problem of assigning prediction tasks to human or AI agents with capacity constraints. The authors develop sequential explore-exploit algorithms that learn agent expertise and optimize assignment policies. Experiments across tabular, image, and text tasks show systematic gains over non-contextual baselines for LLMs and humans.
Key facts
- The paper is titled 'Learning to Assign Prediction Tasks to Agents with Capacity Constraints'.
- It is categorized under Computer Science > Human-Computer Interaction.
- The problem focuses on sequential learning of agent expertise and assignment policies.
- Each agent is constrained to handle a fraction of tasks.
- A general theoretical characterization is provided in terms of agent capacities, expertise differences, and task context.
- The framework uses sequential explore-exploit policy-learning algorithms.
- Experimental results cover tabular, image, and text prediction tasks.
- Gains are demonstrated relative to non-contextual baselines across LLMs and humans.
Entities
Institutions
- arXiv