Truncated Importance Sampling

A sampling technique that reduces variance and prevents gradient explosion by truncating importance weights. It is commonly used in offline reinforcement learning.