Data Darwinism Part II: DataEvolve -- AI can Autonomously Evolve Pretraining Data Curation

• URL: http://arxiv.org/abs/2603.14420v1
• Date: Sun, 15 Mar 2026 15:05:00 GMT
• Title: Data Darwinism Part II: DataEvolve -- AI can Autonomously Evolve Pretraining Data Curation
• Abstract summary: We introduce DataEvolve, a framework that enables strategies to evolve through iterative optimization rather than manual design.<n>For each data category, DataEvolve operates in a closed evolutionary loop: it identifies quality issues, generates candidate strategies, executes them on sampled data, evaluates results, and refines approaches across generations.<n>Applying to 8 categories spanning 672B tokens from Nemotron-CC, DataEvolve produces Darwin-CC, a 504B-token dataset with strategies evolved through 30 iterations per category.
• Score: 21.906969988408573
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: Data Darwinism (Part I) established a ten-level hierarchy for data processing, showing that stronger processing can unlock greater data value. However, that work relied on manually designed strategies for a single category. Modern pretraining corpora comprise hundreds of heterogeneous categories spanning domains and content types, each demanding specialized treatment. At this scale, manual strategy design becomes prohibitive. This raises a key question: can strategies evolve in an automated way? We introduce DataEvolve, a framework that enables strategies to evolve through iterative optimization rather than manual design. For each data category, DataEvolve operates in a closed evolutionary loop: it identifies quality issues, generates candidate strategies, executes them on sampled data, evaluates results, and refines approaches across generations. The process accumulates knowledge through an experience pool of discovered issues and a strategy pool tracking performance across iterations. Applied to 8 categories spanning 672B tokens from Nemotron-CC, DataEvolve produces Darwin-CC, a 504B-token dataset with strategies evolved through 30 iterations per category. Training 3B models on 500B tokens, Darwin-CC outperforms raw data (+3.96 points) and achieves a 44.13 average score across 18 benchmarks, surpassing DCLM, Ultra-FineWeb, and FineWeb-Edu, with strong gains on knowledge-intensive tasks such as MMLU. Analysis shows evolved strategies converge on cleaning-focused approaches: targeted noise removal and format normalization with domain-aware preservation, echoing the L4 (Generative Refinement) principles from Part I. Ablation studies confirm iterative evolution is essential: optimized strategies outperform suboptimal ones by 2.93 points, establishing evolutionary strategy design as feasible and necessary for pretraining-scale data curation.

Related papers

• Self-Evolving Memory for Generative Recommendation [100.15207958893029]
  Generative recommendation has emerged as a promising end-to-end paradigm for personalized recommendation.<n>Existing evolving strategies, such as continual retraining and distillation-based adaptation, directly update the shared model parameters.<n>We propose a self-evolving memory paradigm for generative recommendation, aiming to enable effective evolution across heterogeneous behavioral patterns.
  arXiv  Detail & Related papers  (2026-09-14T14:00:38Z)
• Teaching LLMs to Self-Evolve: Cultivating Core Meta-Skills with Reinforcement Learning [51.17262419591891]
  Test-time scaling through iterative self-evolution with environment feedback, as demonstrated by AlphaEvolve, shows remarkable performance gains.<n>We present MetaEvolve, a framework designed to develop these meta-skills via a data synthesis pipeline, evolution-aware reinforcement learning, and inference-time evolutionary search.
  arXiv  Detail & Related papers  (2026-07-24T04:35:29Z)
• DeepBias: Adaptive In-depth Probing of Social Biases in LVLMs [71.85279836937578]
  We introduce DeepBias, an adaptive framework for the in-depth probing of social biases in LVLMs with carefully designed agents.<n>We build a benchmark named DeepBiasBench using our framework. By employing an ensemble of five diverse state-of-the-art LVLMs as anchors, the benchmark captures vulnerabilities shared across architectures.
  arXiv  Detail & Related papers  (2026-07-13T08:19:09Z)
• CurateEvo: Data-Curation Evolving for Agentic Post-Training [53.68149869349268]
  We propose CurateEvo, a failure-driven dynamic evolution framework for agentic post-training data curation.<n>CurateEvo represents the curation strategy as executable code and iteratively rewrites it using failed trajectories from a held-out development set.<n>At each epoch, the evolved strategy transforms a fixed raw corpus into supervised fine-tuning data, reinforcement learning data, and an inference-time memory bank.
  arXiv  Detail & Related papers  (2026-07-07T11:07:00Z)
• EvoDS: Self-Evolving Autonomous Data Science Agent with Skill Learning and Context Management [16.997238524083365]
  EvoDS is a self-evolving autonomous data science agent that learns to expand its skills and adaptively manage long-term context.<n>We show that EvoDS outperforms state-of-the-art open-source data science agents by an average of 28.9% across four diverse benchmarks.
  arXiv  Detail & Related papers  (2026-06-02T16:20:58Z)
• EvoSelect: Data-Efficient LLM Evolution for Targeted Task Adaptation [79.71802168256542]
  Adapting large language models (LLMs) to a targeted task efficiently and effectively remains a fundamental challenge.<n>One straightforward approach is through an iterative generation-training loop, where candidate data are synthesized through an external generator.<n>We introduce a refined paradigm, namely an iterative generation-selection-training loop, which incorporates a selection step prior to model updates.
  arXiv  Detail & Related papers  (2026-04-28T23:26:16Z)
• Controlled Self-Evolution for Algorithmic Code Optimization [33.82967000330864]
  Self-evolution methods enhance code generation through iterative "generate-verify-refine" cycles.<n>Existing approaches fail to discover solutions with superior complexity within limited budgets.<n>We propose Controlled Self-Evolution (CSE), which consists of three key components.
  arXiv  Detail & Related papers  (2026-01-12T09:23:13Z)
• OpenDataArena: A Fair and Open Arena for Benchmarking Post-Training Dataset Value [74.80873109856563]
  OpenDataArena (ODA) is a holistic and open platform designed to benchmark the intrinsic value of post-training data.<n>ODA establishes a comprehensive ecosystem comprising four key pillars: (i) a unified training-evaluation pipeline that ensures fair, open comparisons across diverse models; (ii) a multi-dimensional scoring framework that profiles data quality along tens of distinct axes; and (iii) an interactive data lineage explorer to visualize dataset genealogy and dissect component sources.
  arXiv  Detail & Related papers  (2025-12-16T03:33:24Z)
• Towards High Data Efficiency in Reinforcement Learning with Verifiable Reward [54.708851958671794]
  We propose a Data-Efficient Policy Optimization pipeline that combines optimized strategies for both offline and online data selection.<n>In offline phase, we curate a high-quality subset of training samples based on diversity, influence, and appropriate difficulty.<n>During online RLVR training, we introduce a sample-level explorability metric to dynamically filter samples with low exploration potential.
  arXiv  Detail & Related papers  (2025-09-01T10:04:20Z)
• Evolutionary Generative Optimization: Towards Fully Data-Driven Evolutionary Optimization via Generative Learning [41.44929681213813]
  We propose a fully data-driven framework empowered by generative learning.<n>EvoGO streamlines the evolutionary optimization process into three stages: data preparation, model training, and population generation.<n>Experiments on numerical benchmarks, classical control problems, and high-dimensional robotic tasks demonstrate that EvoGO consistently converges within merely 10 generations.
  arXiv  Detail & Related papers  (2025-08-01T07:17:57Z)
• DACL-RAG: Data Augmentation Strategy with Curriculum Learning for Retrieval-Augmented Generation [54.26665681604041]
  We introduce DACL-RAG, a multi-stage RAG training framework that combines a multi-level Data Augmentation strategy and a multi-stage Curriculum Learning paradigm.<n>Our framework demonstrates consistent effectiveness across four open-domain QA datasets, achieving performance gains of 2% to 4% over multiple advanced methods.
  arXiv  Detail & Related papers  (2025-05-15T16:53:04Z)
• FREE: Faster and Better Data-Free Meta-Learning [77.90126669914324]
  Data-Free Meta-Learning (DFML) aims to extract knowledge from a collection of pre-trained models without requiring the original data.<n>We introduce the Faster and Better Data-Free Meta-Learning framework, which contains: (i) a meta-generator for rapidly recovering training tasks from pre-trained models; and (ii) a meta-learner for generalizing to new unseen tasks.
  arXiv  Detail & Related papers  (2024-05-02T03:43:19Z)
• Distributionally Robust Cross Subject EEG Decoding [15.211091130230589]
  We propose a principled approach to perform dynamic evolution on the data for improvement of decoding robustness. We derived a general data evolution framework based on Wasserstein gradient flow (WGF) and provides two different forms of evolution within the framework. The proposed approach can be readily integrated with other data augmentation approaches for further improvements.
  arXiv  Detail & Related papers  (2023-08-19T11:31:33Z)
• Lottery Tickets in Evolutionary Optimization: On Sparse Backpropagation-Free Trainability [0.0]
  We study gradient descent (GD)-based sparse training and evolution strategies (ES) We find that ES explore diverse and flat local optima and do not preserve linear mode connectivity across sparsity levels and independent runs.
  arXiv  Detail & Related papers  (2023-05-31T15:58:54Z)
• BMD: A General Class-balanced Multicentric Dynamic Prototype Strategy for Source-free Domain Adaptation [74.93176783541332]
  Source-free Domain Adaptation (SFDA) aims to adapt a pre-trained source model to the unlabeled target domain without accessing the well-labeled source data. To make up for the absence of source data, most existing methods introduced feature prototype based pseudo-labeling strategies. We propose a general class-Balanced Multicentric Dynamic prototype strategy for the SFDA task.
  arXiv  Detail & Related papers  (2022-04-06T13:23:02Z)
• Distributed Evolution Strategies Using TPUs for Meta-Learning [0.0]
  We propose a distributed evolutionary meta-learning strategy using Processing Units (TPUs) that is highly parallel and scalable to arbitrarily long tasks with no increase in memory cost. Using a Prototypical Network trained with evolution strategies on the Omniglot dataset, we achieved an accuracy of 98.4% on a 5-shot classification problem. Our algorithm used as much as 40 times less memory than automatic differentiation to compute the gradient, with the resulting model achieving accuracy within 1.3% of a backpropagation-trained equivalent.
  arXiv  Detail & Related papers  (2022-01-01T02:14:02Z)

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.