Learning to Pose Problems: Reasoning-Driven and Solver-Adaptive Data Synthesis

TL;DR

This paper introduces a reasoning-driven, solver-adaptive data synthesis method for training reasoning models, improving problem quality and difficulty calibration across multiple benchmarks.

Contribution

It presents a novel problem generator that explicitly reasons about problem directions and adapts difficulty based on solver feedback, enhancing data quality for reasoning models.

Findings

Achieves a 3.4% average improvement on 10 reasoning benchmarks.

Effectively calibrates problem difficulty to the solver's ability.

Demonstrates robust generalization across language and vision-language models.

Abstract

Data synthesis for training large reasoning models offers a scalable alternative to limited, human-curated datasets, enabling the creation of high-quality data. However, existing approaches face several challenges: (i) indiscriminate generation that ignores the solver's ability and yields low-value problems, or reliance on complex data pipelines to balance problem difficulty; and (ii) a lack of reasoning in problem generation, leading to shallow problem variants. In this paper, we develop a problem generator that reasons explicitly to plan problem directions before synthesis and adapts difficulty to the solver's ability. Specifically, we construct related problem pairs and augment them with intermediate problem-design CoT produced by a reasoning model. These data are used to bootstrap problem-design strategies in the generator. Then, we treat the solver's feedback on synthetic problems as a reward signal, enabling the generator to calibrate difficulty and produce complementary problems near the edge of the solver's competence. Extensive experiments on 10 mathematical and general reasoning benchmarks show that our proposed framework achieves a cumulative average improvement of 3.4%, demonstrating robust generalization across both language and vision-language models.