Challenge: Current critic-free RL methods for large reasoning models suffer from severe inefficiency when training on positive homogeneous prompts.
Approach: They propose a method that repurposes the policy’s intrinsic uncertainty as a self-supervised reward signal, with no external supervision, auxiliary models, or additional inference cost.
Outcome: Evaluated across six reasoning benchmarks on Qwen3-4B and Qwend3-8B base models, the proposed method achieves state-of-the-art performance among the other four methods.

Similar Papers

Unlocking Exploration in RLVR: Uncertainty-aware Advantage Shaping for Deeper Reasoning (2026.findings-acl)

Copied to clipboard

Challenge: Reinforcement Learning with Verifiable Rewards (RLVR) has shown significant promise for enhancing the reasoning capabilities of large language models (LLMs).
Approach: They propose a model-free method that refines credit assignment by leveraging the model's internal uncertainty signals.
Outcome: Extensive experiments on five mathematical reasoning benchmarks show that the proposed method outperforms strong RLVR baselines on multiple model scales, including 1.5B and 7B.
Verifier-Free RL for LLMs via Intrinsic Gradient-Norm Reward (2026.findings-acl)

Copied to clipboard

Challenge: Large Language Models (LLMs) have demonstrated remarkable capabilities in reasoning tasks and general tasks.
Approach: They propose a "Verifier-free Intrinsic Gradient-Norm Reward" that uses only the policy model itself.
Outcome: The proposed reward outperforms the state-of-the-art RLIF baseline INTUITOR on math benchmarks and shows cross-domain transfer to code benchmarks when trained only on math data.
Reinforced Efficient Reasoning via Semantically Diverse Exploration (2026.acl-long)

Copied to clipboard

Challenge: Existing methods for reinforcement learning with verifiable rewards suffer from limited exploration diversity and inefficient reasoning.
Approach: They propose a method that rewards concise and correct reasoning while penalizing unnecessarily long reasoning chains.
Outcome: Extensive experiments on Qwen and Llama models validate the effectiveness and efficiency of ROSE.
Exploration-Exploitation Reshaping towards Efficient Reasoning for Large Language Models (2026.findings-acl)

Copied to clipboard

Challenge: Large Reasoning Models (LRMs) are constrained by the overthinking issue.
Approach: They propose a policy optimization framework that reshapes the exploration and exploitation through two core components: self-imitation and self-guidance exploration.
Outcome: The proposed model achieves superior reasoning efficiency without compromising overall accuracy.
A Comprehensive Survey on Learning from Rewards for Large Language Models: Reward Models and Learning Strategies (2025.findings-emnlp)

Copied to clipboard

Challenge: Recent developments in Large Language Models have shifted from pre-training to post-training and test-time scaling.
Approach: They present a comprehensive overview of learning from rewards from the perspective of reward models and learning strategies across training, inference, and post-inference stages.
Outcome: The proposed paradigm enables the transition from passive learning from static data to active learning from dynamic feedback.
Low-probability Tokens Sustain Exploration in Reinforcement Learning with Verifiable Reward (2026.findings-acl)

Copied to clipboard

Challenge: Recent studies show that RLVR training is slow and results plateau as policy entropy collapses . low-probability regularization (Lp-Reg) reduces the number of low-quality exploratory tokens induced by RL training .
Approach: They propose a method to reduce RLVR over-penalization by eliminating low-probability exploratory tokens . they propose 'Low-provability Regularization' to reduce the gradual elimination of low-quality exploratory entropy tokens.
Outcome: The proposed method eliminates low-probability exploratory tokens and prevents suppression of potentially valuable low-property candidates.
Beyond Reasoning Gains: Mitigating General-Capability Forgetting in Large Reasoning Models (2026.findings-acl)

Copied to clipboard

Challenge: Reinforcement learning with verifiable rewards (RLVR) has delivered impressive gains in mathematical and multimodal reasoning . however, the recipe introduces a significant risk of capability regression, where models forget foundational skills after prolonged training without employing regularization strategies.
Approach: They propose a replay strategy with dynamic objective reweighting for general knowledge preservation using short-horizon signals of convergence and instability.
Outcome: The proposed method preserves general capabilities and improves reasoning . it can be applied to existing RLVR pipelines without training additional models or tuning .
EMTIR-GRPO: Efficient Multi-Tool Augmented Large Language Models via Reinforcement Learning (2026.findings-acl)

Copied to clipboard

Challenge: Tool-integrated reasoning (TIR) enables large language models to invoke external tools for tasks beyond their internal capacity but often suffers from tool overuse.
Approach: They propose an algorithm that uses a composite reward to model tool costs and tool efficiency.
Outcome: The proposed algorithm models heterogeneous tool costs and encourages more cost-effective tool-use strategies.
MASPO: Unifying Gradient Utilization, Probability Mass, and Signal Reliability for Robust and Sample-Efficient LLM Reasoning (2026.acl-long)

Copied to clipboard

Challenge: Existing RLVR algorithms rely on rigid, uniform, and symmetric trust region mechanisms . current algorithms lack robustness, asymmetric signal reliability and inefficient gradient utilization .
Approach: They propose a framework to harmonize three dimensions of RLVR algorithms, a paper argues . a binary cutoff is used to discard valuable reinforcement signals, they argue .
Outcome: The proposed framework outperforms baselines in evaluating a robust RLVR solution.
Chaining the Evidence: Robust Reinforcement Learning for Deep Search Agents with Citation-Aware Rubric Rewards (2026.acl-long)

Copied to clipboard

Challenge: Existing methods for reinforcement learning (RL) rely on binary outcome rewards that fail to capture the comprehensiveness and factuality of agents’ reasoning process.
Approach: They propose a reward framework that emphasizes reasoning comprehensiveness, factual grounding, and evidence connectivity.
Outcome: The proposed framework outperforms standard outcome-based RL baselines across multiple deep search benchmarks and shows that it discourages shortcut exploitation and promotes comprehensive, evidence-grounded reasoning.

What is GenGO?

GenGO is an NLP powered publication search system. It currenctly indexes 30k+ papers from ACL Anthology, and implements multi-aspect summarization, semantic search, and more!

Information

About
Limitations