Beyond Memorization: Reasoning-Driven Synthesis as a Mitigation Strategy Against Benchmark Contamination

• URL: http://arxiv.org/abs/2509.00072v2
• Date: Mon, 06 Oct 2025 14:10:14 GMT
• Title: Beyond Memorization: Reasoning-Driven Synthesis as a Mitigation Strategy Against Benchmark Contamination
• Authors: Terry Jingchen Zhang, Gopal Dev, Ning Wang, Nicole Ni, Wenyuan Jiang, Yinya Huang, Bernhard Schölkopf, Mrinmaya Sachan, Zhijing Jin,
• Abstract summary: We present an empirical study using an infinitely scalable framework to synthesize research-level QA directly from arXiv papers.<n>We evaluate a lack of significant performance decay near knowledge cutoff dates for models of various sizes, developers, and release dates.<n>We hypothesize that the multi-step reasoning required by our synthesis pipeline offered additional complexity that goes deeper than shallow memorization.
• Score: 77.69093448529455
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: Capability evaluation of large language models (LLMs) is increasingly shadowed by rising concerns of data contamination that cast doubts on whether static benchmarks measure genuine reasoning or mere memorization. We present an empirical study using an infinitely scalable framework to synthesize research-level QA directly from arXiv papers, harnessing the natural temporal structure of research publications where performance decay after knowledge cutoffs may indicate potential contamination. We evaluated 4 frontier model represented by 2 models of different knowledge cutoff dates per family on 1,643 multi-step reasoning questions synthesized from 20,277 arXiv papers stratified over 26 months, covering at least 6 months before and after all cutoff dates. Our results consistently showed a lack of significant performance decay near knowledge cutoff dates for models of various sizes, developers, and release dates. We further performed a comparative analysis with previous longitudinal studies that reported significant post-cutoff performance decay using directly retrieved questions based on public data. we hypothesize that the multi-step reasoning required by our synthesis pipeline offered additional complexity that goes deeper than shallow memorization, which effectively serves a mitigation strategy against benchmark contamination. We fully open source our code and dataset to aid reproducibility and advocate for a paradigm shift that prioritize reasoning-driven synthesis to construct benchmarks over simply collecting newly released questions periodically.

Related papers

• Observationally Informed Adaptive Causal Experimental Design [55.998153710215654]
  We propose Active Residual Learning, a new paradigm that leverages the observational model as a foundational prior.<n>This approach shifts the experimental focus from learning target causal quantities from scratch to efficiently estimating the residuals required to correct observational bias.<n> Experiments on synthetic and semi-synthetic benchmarks demonstrate that R-Design significantly outperforms baselines.
  arXiv  Detail & Related papers  (2026-03-04T06:52:37Z)
• Sparse Additive Model Pruning for Order-Based Causal Structure Learning [4.55744151522751]
  Causal structure learning aims to estimate causal relationships between variables as a form of a causal directed acyclic graph (DAG) from observational data.<n>One of the major frameworks is the order-based approach that first estimates a topological order of the underlying DAG and then prunes spurious edges from the fully-connected DAG induced by the estimated topological order.<n>We introduce a new pruning method based on sparse additive models, which enables direct pruning of redundant edges without relying on hypothesis testing.
  arXiv  Detail & Related papers  (2026-02-17T02:06:42Z)
• Can synthetic data reproduce real-world findings in epidemiology? A replication study using tree-based generative AI [0.6268282038459305]
  We propose the use of adversarial random forests (ARF) as an efficient and convenient method for synthesizing epidemiological data.<n>We replicated statistical analyses from six epidemiological publications and compared original with synthetic results.<n>Results from multiple synthetic data replications consistently aligned with original findings.
  arXiv  Detail & Related papers  (2025-08-19T22:51:40Z)
• Using Imperfect Synthetic Data in Downstream Inference Tasks [50.40949503799331]
  We introduce a new estimator based on generalized method of moments.<n>We find that interactions between the moment residuals of synthetic data and those of real data can improve estimates of the target parameter.
  arXiv  Detail & Related papers  (2025-08-08T18:32:52Z)
• Reasoning or Memorization? Unreliable Results of Reinforcement Learning Due to Data Contamination [67.67725938962798]
  Pre-training on massive web-scale corpora leaves Qwen2.5 susceptible to data contamination in widely used benchmarks.<n>We introduce a generator that creates fully clean arithmetic problems of arbitrary length and difficulty, dubbed RandomCalculation.<n>We show that only accurate reward signals yield steady improvements that surpass the base model's performance boundary.
  arXiv  Detail & Related papers  (2025-07-14T17:55:15Z)
• Retrieving Classes of Causal Orders with Inconsistent Knowledge Bases [0.8192907805418583]
  Large Language Models (LLMs) have emerged as a promising alternative for extracting causal knowledge from text-based metadata.<n>LLMs tend to be unreliable and prone to hallucinations, necessitating strategies that account for their limitations.<n>We present a new method to derive a class of acyclic tournaments, which represent plausible causal orders.
  arXiv  Detail & Related papers  (2024-12-18T16:37:51Z)
• Understanding Synthetic Context Extension via Retrieval Heads [51.8869530817334]
  We investigate fine-tuning on synthetic data for three long-context tasks that require retrieval and reasoning.<n>We find that models trained on synthetic data fall short of the real data, but surprisingly, the mismatch can be interpreted.<n>Our results shed light on how to interpret synthetic data fine-tuning performance and how to approach creating better data for learning real-world capabilities over long contexts.
  arXiv  Detail & Related papers  (2024-10-29T17:55:00Z)
• Demystifying amortized causal discovery with transformers [21.058343547918053]
  Supervised learning approaches for causal discovery from observational data often achieve competitive performance.<n>In this work, we investigate CSIvA, a transformer-based model promising to train on synthetic data and transfer to real data.<n>We bridge the gap with existing identifiability theory and show that constraints on the training data distribution implicitly define a prior on the test observations.
  arXiv  Detail & Related papers  (2024-05-27T08:17:49Z)
• The Real Deal Behind the Artificial Appeal: Inferential Utility of Tabular Synthetic Data [40.165159490379146]
  We show that the rate of false-positive findings (type 1 error) will be unacceptably high, even when the estimates are unbiased. Despite the use of a previously proposed correction factor, this problem persists for deep generative models.
  arXiv  Detail & Related papers  (2023-12-13T02:04:41Z)
• A Discrepancy Aware Framework for Robust Anomaly Detection [51.710249807397695]
  We present a Discrepancy Aware Framework (DAF), which demonstrates robust performance consistently with simple and cheap strategies. Our method leverages an appearance-agnostic cue to guide the decoder in identifying defects, thereby alleviating its reliance on synthetic appearance. Under the simple synthesis strategies, it outperforms existing methods by a large margin. Furthermore, it also achieves the state-of-the-art localization performance.
  arXiv  Detail & Related papers  (2023-10-11T15:21:40Z)
• Rethinking Complex Queries on Knowledge Graphs with Neural Link Predictors [58.340159346749964]
  We propose a new neural-symbolic method to support end-to-end learning using complex queries with provable reasoning capability. We develop a new dataset containing ten new types of queries with features that have never been considered. Our method outperforms previous methods significantly in the new dataset and also surpasses previous methods in the existing dataset at the same time.
  arXiv  Detail & Related papers  (2023-04-14T11:35:35Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.