SAGE: Mitigating Long-Horizon Reasoning Biases via Topological Guidance
SAGE claims structural guidance can reduce two failure modes that derail long multi-step reasoning in LLMs.
The paper frames long-horizon brittleness as exploration bias toward unstable branches and compounding bias from small errors accumulating before sparse rewards appear. Its SAGE framework combines algebraic sparsification with hyperbolic guidance to constrain candidate reasoning paths and add depth-wise signals. The authors report gains across 12 benchmarks and 7 model families, including up to an 8-fold improvement on the Andrews-Curtis problem. HF Daily Papers' note
The paper frames long-horizon brittleness as exploration bias toward unstable branches and compounding bias from small errors accumulating before sparse rewards appear. Its SAGE framework combines algebraic sparsification with hyperbolic guidance to constrain candidate reasoning paths and add depth-wise signals. The authors report gains across 12 benchmarks and 7 model families, including up to an 8-fold improvement on the Andrews-Curtis problem. HF Daily Papers' note
score 5