Paper 2601.09274
$A^3$-Bench: Benchmarking Memory-Driven Scientific Reasoning via Anchor and Attractor Activation
- Published
- Jan 2026
- Research lab
- Independent
- Citations
- 1
- GitHub
- Not linked
01 In brief
Summary
The paper introduces A3-Bench, a benchmark for evaluating memory-driven scientific reasoning in large language models (LLMs).
It is grounded in the concepts of anchors (foundational knowledge units) and attractors (experience-based templates), which are activated during reasoning.
The authors annotate 2,198 science problems across math, physics, and chemistry using a four-step SAPM process (subject, anchor & attractor, problem, and memory developing).
They propose a dual-scale memory evaluation framework and the AAUI (Anchor–Attractor Utilization Index) metric to quantify memory activation rates.
Experiments with ten LLMs and three memory paradigms (vanilla, anchor & attractor activation, and annotated anchor & attractor activation) show that memory activation consistently improves accuracy, especially on hard problems, and reduces inference time.
The AAUI metric correlates with accuracy, and the approach generalizes to other benchmarks like OlympiadBench.
The work provides a memory-centric, interpretable evaluation paradigm for scientific reasoning.
02 From the paper
Abstract
Scientific reasoning relies not only on logical inference but also on activating prior knowledge and experiential structures. Memory can efficiently reuse knowledge and enhance reasoning consistency and stability. However, existing benchmarks mainly evaluate final answers or step-by-step coherence, overlooking the \textit{memory-driven} mechanisms that underlie human reasoning, which involves activating anchors and attractors, then integrating them into multi-step inference. To address this gap, we propose $A^3$-Bench~ https://a3-bench.github.io, a benchmark designed to evaluate scientific reasoning through dual-scale memory-driven activation, grounded in Anchor and Attractor Activation. First, we annotate 2,198 science reasoning problems across domains using the SAPM process(subject, anchor & attractor, problem, and memory developing). Second, we introduce a dual-scale memory evaluation framework utilizing anchors and attractors, along with the AAUI(Anchor--Attractor Utilization Index) metric to measure memory activation rates. Finally, through experiments with various base models and paradigms, we validate $A^3$-Bench and analyze how memory activation impacts reasoning performance, providing insights into memory-driven scientific reasoning.