Gradient-Free Optimization

Definitions

1. Optimization methods that do not rely on computing gradients and instead search using alternatives such as evolutionary strategies, sampling, or black-box evaluation. These methods are useful when gradients are unavailable, unreliable, or too expensive to compute.

   In plain English: Optimization that improves a system without using gradient calculations.

   Example: "They used gradient-free optimization to tune prompt and policy parameters because the reward came from an external simulator with no differentiable path."

Related Terms

• distillation
• Chinchilla Scaling
• Constitutional AI Method
• Context Distillation
• Expert Iteration
• Frozen Embedding
• Goal Conditioned
• Machine Unlearning
• Mesa-Optimization
• Objective Function
• Process Reward Model
• Replay Buffer
• Reward Engineering
• Surrogate Model
• Training Dynamics