Action Shapley: A Training Data Selection Metric for World Model in Reinforcement Learning

arXiv 26.1 2026 51.7 method

TLDR

Introduces Action Shapley, a metric for selecting training data for world models in RL, with efficient computation and real-world validation.

Reasoning

The paper presents a novel metric and algorithm for training data selection in world models, with strong empirical results across five real-world case studies. However, the abstract lacks details on the theoretical foundations and comparisons to other selection methods beyond ad-hoc baselines.

Read-first score

Read-first score 51.7, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 35.

Recency 8%

100

Uses a gentle age decay so recent papers surface without erasing older foundations. 2026

Methodology quality 25%

Screens visible abstract and analysis fields for experiment, dataset, baseline, metric, and limitation evidence. markers=metric,validation

Topical relevance 42%

Uses existing LLM keyword relevance scores normalized to 0-100. world model,world simulator,generative world model,interactive world model,video world model,world dynamics prediction,model-based reinforcement learning world model

Reproducibility 25%

Screens links and visible text for paper, code, dataset, artifact, and repository signals. pdf=True; code=False; dataset=False; markers=none

Field roles

Frontier

Rank sensitivity

Stability: volatile; rank range: 438.

Keyword Scores

world model

model-based reinforcement learning world model

world dynamics prediction

world simulator

generative world model

interactive world model

video world model

Deep Analysis

Innovations

Action Shapley metric for training data selection in world models
Randomized dynamic algorithm to compute Shapley values with reduced complexity

Methodology

The paper proposes Action Shapley, an agnostic metric for selecting training data for world models in reinforcement learning. To compute it efficiently, a randomized dynamic algorithm is introduced to mitigate the exponential complexity of traditional Shapley value computations. The method is empirically validated across five data-constrained real-world case studies.

Key Results

The randomized dynamic algorithm achieves computational efficiency improvement exceeding 80% compared to conventional exponential time computations. The Action Shapley-based training data selection policy consistently outperforms ad-hoc training data selection.

Resources

Paper