importance sampling
Importance sampling is a statistical technique for making estimates more efficient when the events you care about are extremely rare. Instead of sampling uniformly, you deliberately over-sample from the rare, interesting cases. Then you reweigh results to correct the bias, yielding accurate estimates with far fewer samples than naive random sampling.
The papers all use importance sampling to tackle rare but critical events in artificial intelligence. One applies it to credit assignment in long reasoning tasks, another to detecting catastrophic failures in language models, a third to measuring safety risks after fine-tuning, and a fourth to skipping unnecessary calculations during training. Each method achieves significant efficiency gains by focusing computation on what matters most.
- Papers
- 5
- Released code
- 0
- First seen
- Sept 2026
- Latest
- Oct 2026
5 papers in the last two months, against 0 in the two before.
Start here
Two papers to read first, in the order they came out- 1Weight perturbations efficiently estimate extremely rare failures in language-model agents
Iterative unalignment estimates rare event probabilities in stochastic agent trajectories via gradient-based importance sampling.
Sept 2026 - 2Training language agents across turns and tokens improves task performance
Multiscale Policy Optimization improves long-horizon LLM agents by aligning credit assignment and policy updates at the turn level.
Oct 2026
How fast it is moving
Often studied with
advantage estimationIn policy optimization, a method that calculates the relative value of actions using group statistics or baselines to guide learning.1 togetheragentic reinforcement learningTraining approach that lets LLM-based agents learn multi-step tool use and planning via reinforcement learning, often with dense or per-tool rewards.1 togethercredit assignmentDetermining which actions or steps in a sequence contributed to a final outcome, often for reinforcement learning.1 togetherThe papers
Most central to this idea first, not most recent.
- Independentcs.AI
Training language agents across turns and tokens improves task performance
Multiscale Policy Optimization improves long-horizon LLM agents by aligning credit assignment and policy updates at the turn level.
Oct 2026
- Research Labcs.LG
Reversing safety fine-tuning helps estimate failures too rare to count
Safety fine-tuning's rare failures can be estimated by amplifying the fine-tuning signal, sampling one amplified batch, and extrapolating from token probabilities.
Max Planck Institute for Security and Privacy (MPI-SP), CISPA Helmholtz Center for Information Security · Oct 2026
- Independentcs.LG
Weight perturbations efficiently estimate extremely rare failures in language-model agents
Iterative unalignment estimates rare event probabilities in stochastic agent trajectories via gradient-based importance sampling.
Sept 2026
- Chinese Techcs.LG
Expert-guided decoding restores learning signals when a model’s reasoning attempts fail
Policy-Compatible Reinforcement Learning recovers verified training examples for hard problems by guiding a model with expert solutions while limiting how far its behavior shifts.
University of Science and Technology of China, Tencent · Oct 2026
- Big Techcs.CL
Skipping selected attention gradients speeds long-context fine-tuning while preserving dense inference
LazyFlash speeds up long-context fine-tuning by making the backward pass block-sparse while keeping the forward pass dense.
Microsoft · Oct 2026
Concepts are extracted from each paper and reused across the corpus, so this page grows on its own as the desk reads.