Publications
2026 NeurIPS
Test-Time Risk Adaptation with Mixture of Agents
Advances in Neural Information Processing Systems (NeurIPS), 2026
A test-time risk-adaptation method that composes source agents according to deployment-time risk specifications.
Contribution
What this paper adds
- Separates source training from deployment-time risk selection.
- Uses mixture-of-agents behavior to adapt policies locally.
- Studies transfer when risk preferences change after training.