AtManRL: Towards Faithful Reasoning via Differentiable Attention Saliency
Addresses the need for faithful and interpretable reasoning in LLMs, which is important for trust and safety in AI systems.
AtManRL uses differentiable attention manipulation and reinforcement learning to improve faithfulness of chain-of-thought reasoning in LLMs, achieving more transparent reasoning on GSM8K and MMLU with Llama-3.2-3B-Instruct.
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.