![[2506.07751] AbstRaL: Augmenting LLMs' Reasoning by Reinforcing Abstract Thinking](https://arxiv.org/static/browse/0.3.4/images/arxiv-logo-fb.png){kind=logo}
[2506.07751] AbstRaL: Augmenting LLMs' Reasoning by Reinforcing Abstract Thinking
Summary
The paper presents AbstRaL, a method to enhance large language models' reasoning capabilities by reinforcing abstract thinking, particularly addressing robustness in grade school math reasoning.
Why It Matters
As large language models (LLMs) become increasingly integrated into various applications, ensuring their robustness in reasoning tasks is crucial. This research highlights a novel approach to improve LLM performance in the face of distribution shifts, which is essential for their reliability in real-world scenarios.
Key Takeaways
- AbstRaL enhances LLMs' reasoning by promoting abstract thinking.
- The method significantly reduces performance degradation in grade school math reasoning.
- Reinforcement learning is more effective than supervised fine-tuning for this task.
- Improving robustness in math reasoning also benefits general reasoning tasks.
- The research suggests that abstract reasoning can enhance LLM generalizability.
Computer Science > Computation and Language arXiv:2506.07751 (cs) [Submitted on 9 Jun 2025 (v1), last revised 23 Feb 2026 (this version, v4)] Title:AbstRaL: Augmenting LLMs' Reasoning by Reinforcing Abstract Thinking Authors:Silin Gao, Antoine Bosselut, Samy Bengio, Emmanuel Abbe View a PDF of the paper titled AbstRaL: Augmenting LLMs' Reasoning by Reinforcing Abstract Thinking, by Silin Gao and 3 other authors View PDF HTML (experimental) Abstract:Recent studies have shown that large language models (LLMs), especially smaller ones, often lack robustness in grade school math (GSM) reasoning. In particular, they tend to experience performance drops when faced with distribution shifts, such as changes to numerical or nominal variables, or insertions of distracting clauses. A possible strategy to address this involves generating synthetic data to further "instantiate" reasoning problems on potential variations. In this work, we instead focus on the strategy of "abstracting" reasoning problems. This not only helps counteract distribution shifts but also facilitates the connection to symbolic tools for deriving solutions. Focusing on GSM, we find that this abstraction process is better acquired through reinforcement learning (RL) than just supervised fine-tuning, which often fails to produce faithful abstractions. Our method, AbstRaL -- which promotes abstract reasoning in LLMs using RL on granular abstraction data -- significantly mitigates performance degradation on recent GS...
