Simulating, Fast and Slow: Learning Policies for Black-Box Optimization

Name: Simulating, Fast and Slow: Learning Policies for Black-Box Optimization
Uploaded: 2026-10-22T10:36:26.725387+00:00
Duration: 728 s
Description: The paper introduces a Reinforcement Learning framework to learn an active learning policy that guides a differentiable surrogate model's training. This policy decides when to call expensive black-box simulators or retrain the surrogate, using its gradients for optimization. The method significantly reduces simulator calls, achieving up to 90% fewer evaluations compared to existing methods.

The paper introduces a Reinforcement Learning framework to learn an active learning policy that guides a differentiable surrogate model's training. This policy decides when to call expensive black-box simulators or retrain the surrogate, using its gradients for optimization. The method significantly reduces simulator calls, achieving up to 90% fewer evaluations compared to existing methods.

Open the full episode