# reinforcement learning

#reinforcement learning#large language models#prompt selection#kalman filter

Kalman Meets Curriculum: Efficient Dynamic Prompt Selection for Adaptive RL Finetuning

Haodong Zhu, Yangyang Ren, Yanjing Li +4

The paper introduces Kalman-Guided Prompt Selection (KGPS), a method that treats prompt difficulty as a dynamic state estimated with a Kalman filter to adaptively choose prompts du…

cs.IR2026

Hierarchical Latent Reasoning for LLM-based Recommendation

Peiyu Hu, Siying Gu, Weihai Lu +8

The paper introduces HiLaR, a framework that uses hierarchical latent reasoning and layer-aware reinforcement optimization to improve recommendation performance of large language m…

#recommendation systems#large language models#latent reasoning#hierarchical representation

#reinforcement learning#self-distillation#contrastive learning#large language models

Contrastive Reinforced Policy Optimization via Privileged Self-Distillation

Xingjian Wu, Junlin Liu, Xingchen Liu +6

The paper introduces Contrastive Reinforced Policy Optimization (CRPO), a method that frames on‑policy self‑distillation for large language models as a contrastive learning problem…

#3d bin packing#online packing#reinforcement learning#candidate generation

Operationally Guided Placement-Aware Learning for Industrial Online 3D Bin Packing

Dheeraj Poolavaram, Aanchal Rajesh Chugh, Sebastian Dorn

The paper presents OPAL, a framework that improves industrial online 3D bin packing by using an operationally guided empty-maximal-space generator and a placement-aware learning po…

#scene text spotting#visual token grounding#multimodal large language models#reinforcement learning

One Patch Is Enough: Reinforcement-Optimized Visual Token Grounding for MLLM-Based Scene Text Spotting

Rui Tang, Wentao Yang, Peirong Zhang +4

The paper introduces a vision-centric framework called SPaTS that uses a single visual token per text instance and reinforcement learning to improve scene text spotting accuracy an…

#memory-augmented agents#large language models#experience reconstruction#reinforcement learning

MemHarness: Memory Is Reconstructed, Not Replayed

Rong Wu, Daocheng Fu, Licheng Wen +10

The paper introduces MemHarness, a framework that lets large language model agents reconstruct and adapt retrieved past experiences to the current context instead of replaying them…

#large language models#reinforcement learning#high-performance computing#reward optimization

HARGO: Heterogeneity-Aware Reward-Guided Optimization for RL Post-Training of LLMs on HPC Tasks

Tiangang Li, Xiangbo Tian

The paper introduces HARGO, a reinforcement‑learning post‑training method that weights responses by confidence and reward contrast to better align large language models with divers…

#reinforcement learning#teacher‑student distillation#multi‑teacher aggregation#reasoning models

Beyond the Best Teacher: Expanding and Compressing the Reasoning Solution Manifold

Songshuo Lu, Zhi Chen, Yaohua Tang

The paper proposes an expand‑then‑compress framework that builds a diverse set of RL‑trained teacher models and then distills them into a single student model, improving reasoning,…

#hvac control#multi-zone vav#reinforcement learning#large language models

Exact Action Values Are Not Enough: Rollout-Verified Reinforcement Fine-Tuning of a Reasoning Model for Multi-Zone VAV Control

Takumi Shioda, Kohei Terashima, Tatsuo Nagai

The paper investigates using a reasoning language model and TD3-guided reinforcement fine‑tuning to control multi‑zone variable‑air‑volume HVAC systems without building‑specific tr…

#vision-language models#tool use#self-judging#reinforcement learning

FaithEyes: Towards Faithful Tool Use via Multi-Agent Process-Image Verification

Haoqing Wang, Xingrun Xing, Wei Xia +2

FaithEyes proposes a multi‑agent framework where a vision‑language model judges its own tool calls to ensure they are useful, improving both accuracy and tool faithfulness on visua…

#test-time compute#self-verification#reinforcement learning#adaptive inference

SVR: Self-Verifying Refinement via Joint Verdict-Confidence Reinforcement Learning for Adaptive Test-Time Compute

Hongyu Chen, Liang Lin, Guangrun Wang

The paper proposes Self‑Verifying Refinement (SVR), a reinforcement‑learning framework that lets language models decide when to stop refining answers by using their own correctness…

cs.CL2026

Frontis-MA1: Training an AI4AI Model towards Recursive Self-Improvement in Machine Learning Engineering

Junlin Yang, Che Jiang, Yu Fu +21

The paper presents Frontis-MA1, a 35‑billion‑parameter model trained as a meta‑evolution agent for machine learning engineering, using a new OpenMLE stack that combines operator le…

#recursive self-improvement#ai4ai#machine learning engineering#meta‑learning

cs.RO2026

PAC-MAN: Perception-Aware CBF-RL for Whole-Body Safety in Humanoid Dodgeball

Lizhi Yang, Junheng Li, Aaron D. Ames

The paper introduces PAC-MAN, a perception-aware control-barrier-function reinforcement learning framework that enables a humanoid robot to safely dodge balls using depth segmentat…

#humanoid robots#control barrier functions#reinforcement learning#perception

cs.CL2026

Harness-G: A Graph-Structured Harness for Search Agents

Yanning Hou, Haoyuan Chen, Sihang Zhou +7

The paper introduces Harness-G, a graph-structured retrieval framework that turns free-form query generation into finite action selection for reinforcement learning search agents a…

#reinforcement learning#search agents#graph-structured retrieval#question answering

#image-to-svg generation#visual feedback loop#reinforcement learning#multimodal large language models

RefineSVG: Visual Feedback-Driven Reinforcement Learning for Image-to-SVG Generation

Shaobo Liu, Feiqiao Mao, Shuaishuai Zhou +4

RefineSVG introduces a closed-loop visual feedback system that lets large multimodal language models iteratively correct SVG code by rendering the output, comparing it to the targe…

#counterfactual explanations#reinforcement learning#model interpretability#class-aware learning

Class-Aware Reinforcement Learning for Counterfactual Explanation Generation

Muhammad Adil Saleem, Syed Ali Raza, Mary-Anne Williams

The paper investigates adding the predicted class of an instance to the reinforcement‑learning state representation for generating counterfactual explanations, showing that this cl…

cs.IR2026

VIG-RL: Learning to Search and Insert for Verified Image Grounding

Qinhan Yu, Jun Guang, Chong Chen +1

The paper introduces VIG-RL, a reinforcement‑learning based agent that dynamically decides when to retrieve, select, and insert authentic images into text responses, improving veri…

#verified image grounding#reinforcement learning#multimodal retrieval#tool use

#computer-use agents#synthetic environments#stateful applications#co-evolution training

Echoverse: Deep, Evolving Environments for Training Computer-Use Agents at Scale

Yash Pandya, Sahil Gupta, Sarthak Harne +10

Echoverse introduces a pipeline that compiles specifications into deep, stateful synthetic applications for training computer-use agents, using a co‑evolution loop that repairs env…

#agentic visual reasoning#tool use adaptiveness#multimodal large language models#reinforcement learning

Beacon: Knowing When and How to Perform Agentic Visual Reasoning

Qixun Wang, Yang Shi, Letian Cheng +11

The paper introduces Beacon, an agentic visual reasoning system that learns when to invoke external tools and how to use them effectively, improving multimodal large language model…

#image forgery detection#vision-language models#reinforcement learning#reasoning traces

Can Vision-Language Models Reason about AI Edits in Images?

Darsha Udayanga, Pin-Yu Chen, Payel Das +1

The paper explores training vision-language models with reinforcement learning to detect and localize AI-generated image edits, using reasoning traces and a lightweight segmentatio…

cs.CL2026

LEEPS: Latent-Guided Explore-Exploit Prompt Sampling for Efficient RLVR in Large Language Models

Shuang Liang, Haoyang Zhou, Yifan Gong +2

The paper introduces LEEPS, a latent-guided explore‑exploit prompt sampler that selects prompts before rollout to reduce wasted generation budget and improve reinforcement learning…

#prompt sampling#reinforcement learning#large language models#explore-exploit

#gui agents#foundation models#mobile automation#reinforcement learning

Qwen-UI-Agent Technical Report: Toward Next-Generation Real-World Centric Foundation GUI Agents

Hanzhang Zhou, Panrong Tong, Xu Zhang +13

The paper introduces Qwen-UI-Agent, a foundation model for GUI agents that can operate across mobile, desktop, web, and search environments, combining GUI actions with CLI commands…

cs.RO2026

X-NavDP: Generalizing Navigation Diffusion Policy to Novel Behavior and Embodiments with Group Q-score Reweighted Matching

Tianyu Yang, Yiming Zeng, Wenzhe Cai +5

The paper introduces X-NavDP, a diffusion-based visual navigation policy that is fine‑tuned with a novel Group Q-score Reweighted Matching (GQRM) reinforcement learning framework t…

#visual navigation#diffusion policies#reinforcement learning#cross‑embodiment