·
DataBubble
  • Home
  • Models
  • News
  • Compare
  • Boards
  • Pricing
  • About
  • Contact
Latest
Canva apologizes after its AI tool replaces ‘Palestine’ in designs33m◆China blocks Meta’s $2B Manus deal after months-long probe1h◆OpenAI could be making a phone with AI agents replacing apps1h◆Rebuilding the data stack for AI2h◆Join the new AI Agents Vibe Coding Course from Google and Kaggle2h◆The AI-designed car is taking shape4h◆Meta inks deal for solar power at night, beamed from space5h◆The next phase of the Microsoft OpenAI partnership9h◆From Local to Cluster: A Unified Framework for Causal Discovery with Latent Variables11h◆Consequentialist Objectives and Catastrophe11h◆EgoMAGIC- An Egocentric Video Field Medicine Dataset for Training Perception Algorithms11h◆ReCast: Recasting Learning Signals for Reinforcement Learning in Generative Recommendation11h◆A Probabilistic Framework for Hierarchical Goal Recognition11h◆CNSL-bench: Benchmarking the Sign Language Understanding Capabilities of MLLMs on Chinese National Sign Language11h◆The Shape of Adversarial Influence: Characterizing LLM Latent Spaces with Persistent Homology11h◆Toward Principled LLM Safety Testing: Solving the Jailbreak Oracle Problem11h◆Learning from Natural Language Feedback for Personalized Question Answering11h◆TS-Arena -- A Live Forecast Pre-Registration Platform11h◆Report for NSF Workshop on AI for Electronic Design Automation11h◆Initial results of the Digital Consciousness Model11h◆Canva apologizes after its AI tool replaces ‘Palestine’ in designs33m◆China blocks Meta’s $2B Manus deal after months-long probe1h◆OpenAI could be making a phone with AI agents replacing apps1h◆Rebuilding the data stack for AI2h◆Join the new AI Agents Vibe Coding Course from Google and Kaggle2h◆The AI-designed car is taking shape4h◆Meta inks deal for solar power at night, beamed from space5h◆The next phase of the Microsoft OpenAI partnership9h◆From Local to Cluster: A Unified Framework for Causal Discovery with Latent Variables11h◆Consequentialist Objectives and Catastrophe11h◆EgoMAGIC- An Egocentric Video Field Medicine Dataset for Training Perception Algorithms11h◆ReCast: Recasting Learning Signals for Reinforcement Learning in Generative Recommendation11h◆A Probabilistic Framework for Hierarchical Goal Recognition11h◆CNSL-bench: Benchmarking the Sign Language Understanding Capabilities of MLLMs on Chinese National Sign Language11h◆The Shape of Adversarial Influence: Characterizing LLM Latent Spaces with Persistent Homology11h◆Toward Principled LLM Safety Testing: Solving the Jailbreak Oracle Problem11h◆Learning from Natural Language Feedback for Personalized Question Answering11h◆TS-Arena -- A Live Forecast Pre-Registration Platform11h◆Report for NSF Workshop on AI for Electronic Design Automation11h◆Initial results of the Digital Consciousness Model11h◆
News/Preserve Support, Not Correspondence: Dynamic Routing for Offline Reinforcement Learning
arxiv
PublishedApril 27, 2026 at 4:00 AM

Preserve Support, Not Correspondence: Dynamic Routing for Offline Reinforcement Learning

Source
arxiv.orgfull article ↗
Read on arxiv→
Publisher summary· verbatim

arXiv:2604.22229v1 Announce Type: cross Abstract: One-step offline RL actors are attractive because they avoid backpropagating through long iterative samplers and keep inference cheap, but they still have to improve under a critic without drifting away from actions that the dataset can support. In r

Discussion
Source
↗
arxiv
Read original ↗All from arxiv →

No replies yet. Be first.

Source
↗
arxiv
Read original ↗All from arxiv →

Related coverage

More from ARXIV
arxivFrom Local to Cluster: A Unified Framework for Causal Discovery with Latent Variables11harxivConsequentialist Objectives and Catastrophe11harxivEgoMAGIC- An Egocentric Video Field Medicine Dataset for Training Perception Algorithms11harxivReCast: Recasting Learning Signals for Reinforcement Learning in Generative Recommendation11h
Originally published on arxiv ↗
HomeModelsNews