mlsanityData to 7 Oct 2026

AtManRL: Towards Faithful Reasoning via Differentiable Attention Saliency

arXiv 2604.16158 ↗First posted 17 Apr 2026cs.CL cs.AI cs.LGcited by 0

On the map: LLM reasoning → Multimodal reasoning

Abstract (arXiv)

Large language models (LLMs) increasingly rely on chain-of-thought (CoT) reasoning to solve complex tasks. Yet ensuring that the reasoning trace both contributes to and faithfully reflects the processes underlying the model's final answer, rather than merely accompanying it, remains challenging. We introduce AtManRL, a method that leverages differentiable attention manipulation to learn more faithful reasoning through reinforcement learning. By training an additive attention mask that identifies tokens in the CoT crucial for producing correct answers, we derive a saliency reward signal that encourages the model to generate reasoning traces that genuinely influence its final predictions. We integrate this saliency reward with outcome-based rewards within the GRPO framework to jointly optimize for correctness and interpretability. Experiments on GSM8K and MMLU with Llama-3.2-3B-Instruct demonstrate that our approach can identify influential reasoning tokens and enable training more transparent reasoning models.

Key facts. No citing papers in the map yet (data through 7 Oct 2026). Placed in Multimodal reasoning, one of 17 directions in LLM reasoning. Builds on 14 of its 23 references that are in the map. First posted on arXiv 17 Apr 2026.

Builds on

14 of 23 references.

References · 14

  1. 2024The Llama 3 Herd of Modelscited by 14,192
  2. 2021Training Verifiers to Solve Math Word Problemscited by 9,743
  3. 2024DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Modelscited by 8,851
  4. 2025Qwen3 Technical Reportcited by 8,195
  5. 2023Let's Verify Step by Stepcited by 3,598
  6. 2017Decoupled Weight Decay Regularizationcited by 1,722
  7. 2024OpenAI o1 System Cardcited by 1,220
  8. 2023Measuring Faithfulness in Chain-of-Thought Reasoningcited by 668
  9. 2020Towards Faithfully Interpretable NLP Systems: How should we define and evaluate faithfulness?cited by 590
  10. 2026Reinforcement Learning via Self-Distillationcited by 285
  11. 2024Faithfulness vs. Plausibility: On the (Un)Reliability of Explanations from Large Language Modelscited by 104
  12. 2026Why Does Self-Distillation (Sometimes) Degrade the Reasoning Capability of LLMs?cited by 87
2 more
  1. 2024Critical Tokens Matter: Token-Level Contrastive Estimation Enhances LLM's Reasoning Capabilitycited by 80
  2. 2025Ignore the KL Penalty! Boosting Exploration on Critical Tokens to Enhance RL Fine-Tuningcited by 33

Built on by

Not cited yet.

Similar, not cited

Close in wording, no citation either way

  1. 2026Faithful-MR1: Faithful Multimodal Reasoning via Anchoring and Reinforcing Visual Attention
  2. 2025Visual Attention Reasoning via Hierarchical Search and Self-Verification
  3. 2025Attention Illuminates LLM Reasoning: The Preplan-and-Anchor Rhythm Enables Fine-Grained Policy Optimization
  4. 2025Reasoning Under 1 Billion: Memory-Augmented Reinforcement Learning for Large Language Models
  5. 2026Saliency-R1: Enforcing Interpretable and Faithful Vision-language Reasoning via Saliency-map Alignment Reward
  6. 2026Credit Where It is Due: Cross-Modality Connectivity Drives Precise Reinforcement Learning for MLLM Reasoning
  7. 2026Do MLLMs Really See It: Reinforcing Visual Attention in Multimodal LLMs
  8. 2026Balancing Efficiency and Efficacy: Training-Free Attention-Guided Switching Between Explicit and Latent Thoughts for MLLMs