mlsanityData to 7 Oct 2026

Offline Behavioral Data Selection

arXiv 2512.18246 ↗First posted 20 Dec 2025cs.LG cs.AIcited by 0

On the map: Offline RL and imitation learning from data → Offline RL algorithms

Abstract (arXiv)

Behavioral cloning is a widely adopted approach for offline policy learning from expert demonstrations. However, the large scale of offline behavioral datasets often results in computationally intensive training when used in downstream tasks. In this paper, we uncover the striking data saturation in offline behavioral data: policy performance rapidly saturates when trained on a small fraction of the dataset. We attribute this effect to the weak alignment between policy performance and test loss, revealing substantial room for improvement through data selection. To this end, we propose a simple yet effective method, Stepwise Dual Ranking (SDR), which extracts a compact yet informative subset from large-scale offline behavioral datasets. SDR is build on two key principles: (1) stepwise clip, which prioritizes early-stage data; and (2) dual ranking, which selects samples with both high action-value rank and low state-density rank. Extensive experiments and ablation studies on D4RL benchmarks demonstrate that SDR significantly enhances data selection for offline behavioral data.

Key facts. No citing papers in the map yet (data through 7 Oct 2026). Placed in Offline RL algorithms, one of 13 directions in Offline RL and imitation learning from data. Builds on 18 of its 36 references that are in the map. First posted on arXiv 20 Dec 2025.

Builds on

18 of 36 references.

Closest ancestors · 4

Same direction, newest first

  1. 2025State Diversity Matters in Offline Behavior Distillation7 Dec
  2. 2023Cal-QL: Calibrated Offline RL Pre-Training for Efficient Online Fine-Tuning9 Mar
  3. 2020Offline Reinforcement Learning: Tutorial, Review, and Perspectives on Open Problems4 May
  4. 2020D4RL: Datasets for Deep Data-Driven Reinforcement Learning15 Apr

Other references · 14

  1. 2014Adam: A Method for Stochastic Optimizationcited by 36,960
  2. 2018Soft Actor-Critic: Off-Policy Maximum Entropy Deep Reinforcement Learning with a Stochastic Actorcited by 4,635
  3. 2016End to End Learning for Self-Driving Carscited by 1,000
  4. 2018An Empirical Study of Example Forgetting during Deep Neural Network Learningcited by 669
  5. 2021Deep Learning on a Data Diet: Finding Important Examples Early in Trainingcited by 529
  6. 2020Dataset Condensation with Gradient Matchingcited by 510
  7. 2018Behavioral Cloning from Observationcited by 482
  8. 2021GRAD-MATCH: Gradient Matching based Data Subset Selection for Efficient Deep Model Trainingcited by 269
  9. 2022DeepCore: A Comprehensive Library for Coreset Selection in Deep Learningcited by 146
  10. 2021GCR: Gradient Coreset Based Replay Buffer Selection For Continual Learningcited by 95
  11. 2022Active Learning Through a Covering Lenscited by 62
  12. 2022Disentangling Transfer in Continual Reinforcement Learningcited by 31
2 more
  1. 2021Core-set Sampling for Efficient Neural Architecture Searchcited by 23
  2. 2016Learning Transferable Policies for Monocular Reactive MAV Controlcited by 18

Built on by

Not cited yet.

Similar, not cited

Close in wording, no citation either way

  1. 2025Expert or not? assessing data quality in offline reinforcement learning
  2. 2026JumpStart Your Policy Learning with Lessons from 160,000 Training Runs
  3. 2021A Closer Look at Advantage-Filtered Behavioral Cloning in High-Noise Datasets
  4. 2022When Should We Prefer Offline Reinforcement Learning Over Behavioral Cloning?
  5. 2023Unified Off-Policy Learning to Rank: a Reinforcement Learning Perspective
  6. 2021Representation Matters: Offline Pretraining for Sequential Decision Making
  7. 2022Behavior Prior Representation learning for Offline Reinforcement Learning
  8. 2022Improving TD3-BC: Relaxed Policy Constraint for Offline Learning and Stable Online Fine-Tuning