Data-Curation

Because of its collaborative nature, Wikidata suffers from errors, in- consistencies, and excessive complexity, such as redundant classes, ambiguity between instances and classes, wrong taxonomic paths, and type constraint violations. The manual curation of these issues is infeasible at scale. To address these challenges, we introduce WiCleanData, a refined version of Wikidata with a consistent tax- onomy and free from type constraint violations. Specifically, we have designed an automated pipeline that first cleans the taxonomy with language model assistance, then simplifies type constraints by hierarchical aggregation, and finally filters facts accordingly. The resulting knowledge graph, free from any type violation, is made publicly available via a Web interface, enabling easy exploration and downstream applications.

LLM agents have recently shown promise in automating machine learning engineering by editing model and training code under execution feedback. Data, however, remains largely outside this agentic optimisation loop. We frame pre-training data selection as heuristic engineering over per-document features, i.e., lexical statistics, categorical labels, and perplexity. We introduce AutoData, an agent that searches directly over executable selection algorithms. Unlike prior data mixture methods that optimise weights over a fixed set of domains, AutoData searches a richer program space of scoring, stratification, and stochastic selection rules, discovering feature interactions automatically by iteratively refining algorithms with validation feedback from a proxy model. Within an overnight search, AutoData discovers a selection algorithm that outperforms existing human-designed curation pipelines. Despite being searched only on this small proxy, the discovered recipe transfers to larger scales and improves the downstream metric CORE. These results suggest that data engineering can be treated as an agentic machine learning problem, extending autonomous research from model and training-code optimization to the data.

Training embodied foundation models typically requires massive-scale datasets and extensive computational resources, yet often suffers from three critical limitations: (1) inefficient sample utilization due to low-informative samples; (2) imbalanced gradient contributions across heterogeneous tasks; and (3) severe credit assignment problem in long-horizon planning, where trajectory-level rewards indiscriminately penalize all tokens. To address these issues, we propose an efficient training paradigm that achieves state-of-the-art average performance through strategic data selection and hierarchical policy optimization. Our approach consists of three synergistic stages. First, Rejection Sampling-based Fine-Tuning (RSFT) filters out low-informative samples to establish robust behavioral priors while preventing distributional collapse. Second, Iterative Rejection GRPO (IR-GRPO) employs task-specific queues stratified by difficulty to keep datasets balanced across reinforcement learning iterations, coupled with a hybrid reward mechanism for precise cross-task feedback. Third, to enhance long-horizon task planning, we introduce Trie-GRPO, a novel reinforcement learning algorithm based on action prefix trees, which enables step-level advantage estimation. This resolves the credit assignment problem by isolating intermediate correct decisions from downstream errors, while effectively balancing exploration efficiency and depth compared to conventional search trees. As a result, EmbodiedMind achieves a state-of-the-art average performance of 70.02% across 18 benchmarks, and significantly outperforms other embodied foundation models in long-horizon task planning accuracy. Our project will be released for reproducibility.

Persistent memory is entering production-oriented agent platforms to help long-horizon agents accumulate experience across sessions. Yet a post-task curator agent restricted to completed trajectories can preserve errors, overgeneralize partial evidence, or retain stale knowledge. We introduce environment-probing curation, a deployment-compatible extension that gives an existing asynchronous curator agent least-privilege, read-only world tools to check, scope, and refresh candidate memories. It requires no model retraining and leaves the task agent, retriever, memory representation, and production write authority unchanged. In a production-like GitHub Copilot (GHCP) harness built on its SDK, we compare stateless execution, full in-context learning, GHCP + Mem, and GHCP + Mem (w/ Env Probing) on CLBench database exploration and 90 adapted APEX management-consulting tasks. On CLBench, probing raises pass rate from 39% to 73% and pass-discounted reward from 8.60 to 22.60 while reducing queries from 8.8 to 4.7 per question and task-agent cost from $3.38 to $1.68. Across six APEX worlds, all 18 memory-versus-baseline mean reward comparisons are positive and task-agent tool calls fall by 16--75%; probing gives the best task-agent reward gain per dollar in five worlds. Probing also attains higher mean reward than GHCP + Mem on both Sonnet 4.6 and Opus 4.7 without schema drift. Environment probing therefore turns existing agent-memory curation into an environment-informed, auditable process while preserving a compact task-time interface.

Nonverbal vocalizations (NVVs), such as laughter, sighs, breaths, and coughs, convey affective and interactional information that conventional automatic speech recognition (ASR) systems often discard. We present a bilingual Mandarin-English system for Track 1 of the NVVSpeech Challenge at ISCSLP 2026, which requires joint transcription of lexical content and 16 NVV categories at their transcript-relative positions. Our NVV-Aware Whisper adapts Whisper-medium through checkpoint-compatible vocabulary remapping, enabling lexical tokens and inline NVV tags to be decoded within a unified autoregressive sequence without expanding the vocabulary. To provide reliable and diverse supervision, we further introduce a source-adaptive data curation strategy that refines public NVV corpora through acoustic augmentation and multimodal LLM filtering, while mining spontaneous NVVs from in-the-wild media through automated preprocessing and annotation. Under the official bilingual evaluation protocol, the proposed system improves final score from 33.32 to 53.61, with ablations confirming the complementary benefits of the proposed data-curation components.

Extracting structured fields from hundreds of millions of documents annually remains costly in regulated industries: bespoke OCR cascades cover only a fraction of workflows, privacy rules preclude external models, and existing open-source VLMs that clear quality thresholds cost more to serve than human annotation. We present a deployed document-understanding system built on a Mixture-of-Experts VLM (35B total, 3B active), fine-tuned on in-house production data mixed with open-domain documents curated by a Difficulty-Aware pipeline for layout diversity, fact-extractability, and cross-model consistency. Fitting on a single H100 and serving heterogeneous workflows via prompting, the model leads all deployable (non-reasoning) baselines up to an order of magnitude larger. A quality-adjusted cost analysis, with confirmation and correction costs calibrated from production telemetry, shows it reduces expected costs by over 80% against the human baseline and by more than 50% against the best competing open-source model, while larger baselines remain economically unviable.

Agentic software engineering benchmarks are typically summarized by nominal category labels such as "bug fix" or "feature implementation," yet benchmarks carrying the same label are built through very different curation pipelines. A label thus reveals little about the engineering work a benchmark demands. We introduce the Spread--Novelty--Centrality (SNC) profile, a three-axis characterization of the demands of repository-level coding tasks, grounded in empirical software engineering research. We apply the profile to five widely used benchmarks and 14,922 trajectories of two model families at three scales, and report three findings. (1) A label is an unreliable proxy for task demands, as every pair of benchmarks is statistically separated on at least two SNC axes, and the separations trace back to specific curation decisions. (2) Agent behaviour reveals demands that the human-written gold solution cannot. Agents produce larger solutions than the gold where problem statements withhold hints and smaller ones where curation inflates the gold. How a task is phrased shapes what an agent produces. (3) Task demands correlate with success uniformly, with resolved runs concentrating in the low-SNC region for every family and scale, whereas the behavioural signatures of success are family-specific. Claude succeeds by matching the scope of the gold solution, and its parity share on files rises from $0.17$ at the smallest scale to $0.54$ at the largest. Qwen succeeds by exceeding the gold scope at every scale, and editing too little marks failure for both families.

Motivation: LinkML is a suitable language for the representation of the structural and content constraints of different kinds of biomedical data. Even if it is a quite recent proposal, it has been applied in several biomedical contexts. Developing and maintaining LinkML schemas presents several challenges, particularly for novice curators. Non-expert bio-curators may struggle with LinkML syntax and best practices, requiring significant time and effort to develop well-structured schemas. Results: In this paper we propose SchemaLink, a web-based environment for the graphical construction and enhancement of LinkML schemas that address the following requirements: $(i)$ introduce a graphical language for the specification of LinkML schemas, $(ii)$ make uniform the specification of schemas in similar contexts, $(iii)$ simplify the design and curation processes by exploiting a RAG-based approach to assist curators in creating new schemas from scratch and editing already developed ones. Several experimental analyses show the quality of the produced LinkML schemas through the AI-based editing facilities. Availability and Implementation: SchemaLink is available online at: https://SchemaLink.biodata.di.unimi.it. SchemaLink code and testing data are available as open-source on GitHub at: https://github.com/AnacletoLAB/{schemalink-webapp,schemalink-api}.

Extraction produces candidate entities and relationships; writing them into a graph is where identity is decided, and identity decisions are destructive in a way extraction errors are not. A wrong type can be corrected later, but two records merged under one identity cannot be separated once their properties have been combined, and the merge leaves no error behind. This paper describes the ingestion and ontology-tagging layer that turns a validated extraction stream into a knowledge graph of 537,157 entities and 2,198,567 relationships drawn from 98,795 government documents. We describe a record-identity ladder that decides sameness from identifier columns, name columns, display names and type-scoped position rather than from name similarity. The ladder governs de-duplication within parsed tables, while the graph write applies a coarser canonical-name key, so records sharing a canonical name merge automatically on exact equality. We argue rather than demonstrate that this is where the automation line belongs: no identity benchmark is reported, and the over-merges the key permits are undetectable by construction. That policy, under which entity resolution only ever flags candidates, followed an incident in which two surface forms of one name were merged, corrupting a correct record and deleting eight entities from an unrelated document. We then describe multi-class ontology tagging and an evidence asymmetry we did not anticipate: an entity name is an instance label rather than a type assertion, so matching name fragments against a class index invents classifications. Requiring anchored evidence cut role assignments on an enriched sample from 36 to 4, all confirmed correct. We quantify the graph's conformance debt, show secondary classifications compensating for a mis-parented primary class, and describe a curation queue grown to 48,403 pending proposals against 775 human decisions.

Although existing instruction data selection methods have introduced various metrics, the inherent complexity of real-world datasets makes it impractical for any single metric to generalize across all scenarios. Developers are thus often forced to manually inspect data and craft heuristic rules for each new application---a tedious and error-prone process. In this paper, we propose a paradigm shift from manual configuration to automated orchestration via the Instruction Data Selection Agent (DataMaster), which interprets user intent and autonomously composes optimal selection strategies. By allowing users to specify data needs through natural language descriptions, DataMaster simplifies data curation and removes the burden of manual strategy design. Extensive experiments across the math, medical, and code domains show that DataMaster outperforms static baselines in most settings and surpasses full-pool training in a substantial number of cases. The implementation of DataMaster and the scripts needed to reproduce the reported pipeline are publicly available at https://github.com/nju-websoft/DataMaster.

Agentic code generation has shown promise in automating and accelerating software development by utilizing Large Language Models (LLMs) to generate, test, and deploy code. For engineers and scientists, such systems have the potential to accelerate the development of applied and scientific workflows while reducing barriers to entry in domains that have yet to fully realize their benefits. However, a key gap remains: existing coding agents primarily focus on code generation and do not address the entire workflow lifecycle, including deployment and sharing. As a result, users develop and stitch modules independently while managing deployment on their own. To address this gap, we propose CURATE - Composition, User-in-the-loop, Reuse, and Automated Task Execution - a novel human-in-the-loop multi-agent system that uses LLM agents to manage and develop composable workflows across their entire lifecycle. A key feature of the system is a catalog that allows for the storage and reuse of modules across workflows. Module catalogs provide a foundation that can be expanded to support FAIR principles by facilitating the sharing and reuse of curated modules and subgraphs. We demonstrate the feasibility of our system with an initial prototype using Claude Opus 4.8, comprising 6 experiments: reproducing and adapting 4 workflows derived from the SeBS-Flow benchmark suite, and automating the development and scaling of a workflow that leverages a complex mechanistic model in environmental engineering used to simulate anaerobic digestion.

Long-video question answering requires identifying sparse yet critical evidence from videos containing thousands of frames under a constrained visual-token budget. Existing methods either select query-aware frames in a single pass or rely on timestamped text solely as retrieval guidance, leading to two key limitations. First, selected frames tend to cluster around local relevance peaks, and once the budget is exhausted, omitted evidence cannot be recovered. Second, textual and visual evidence remain weakly aligned. We propose GCR, a training-free framework that casts fixed-budget frame selection as a joint evidence curation problem. Ground converts timestamped text into temporal events, selects query-relevant real frame anchors, and renders each event text onto its temporally aligned frame. Cover supplements grounded events with direct visual anchors for complementary visual evidence and applies global maximal marginal relevance to preserve diverse context. Refine revisits omitted temporal regions and replaces the weakest revisable context frame with a real-frame medoid---but only when the medoid offers greater evidence value. GCR maintains a fixed number of chronologically ordered frames and requires no VLM training or architectural modification. Experiments on LongVideoBench and Video-MME, across three 7B backbones and frame budgets of 8, 32, and 64, demonstrate consistent improvements in long-video QA. With the 7B LLaVA-OV backbone and 32 frames, GCR achieves 64.25% and 62.15% on the two benchmarks, outperforming the strongest reproduced baselines by 2.54 and 1.93 percentage points, respectively.

Platform teams hosting agent-extensibility surfaces face a regression-economics paradox: every onboarding customer ships an evaluation set tuned to their domain, but the platform's regression set must live under a hard query-count ceiling bounded by release cadence. To our knowledge, no published industrial pipeline addresses this platform-side curation problem: existing evaluation frameworks are customer-side, and benchmark-compression work treats benchmarks as fixed pools rather than streams of incoming sets. We describe a capability-taxonomy-driven curation pipeline applied to declarative agents with custom actions in Microsoft 365 Copilot. It takes an agent specification and a customer's eval set as input, projects each query into a platform-owned capability taxonomy, and outputs per-query decisions (admit, drop, swap, or human review), under the philosophy that a healthy regression set is the minimal set of queries capturing the maximal spread of capability signatures -- distinct combinations of capabilities a query exercises together. Three components instantiate this: a classifier producing per-(query, capability) verdicts via a hybrid of deterministic specification-based extraction and large-language-model (LLM) semantic inference; an Invocation Quality (IQ) rater scoring how thoroughly a query exercises each capability, so a new query sharing a signature with an existing entry can still be recognized as a better test and displace it; and a consolidator comparing incoming queries against the regression set on coverage and quality through a rule-based decision cascade, backed by a conservative curator that only suggests evictions. The mechanism is taxonomy-agnostic and applies to any regression eval-set curation problem with a typed capability taxonomy, including taxonomies that evolve in response to the very evidence the pipeline surfaces.

Training large language models (LLMs) to write register-transfer level (RTL) requires large corpora of paired specifications and code, and such data is scarce enough that most public corpora are now synthesized. Synthesis provides scale but not correctness, and in two widely used RTL datasets only 24.4% and 53.5% of pairs pass generated functional tests. This raises the question of how much of such a corpus to keep and which part of it. Correctness alone is a poor answer. A pair that misbehaves in one corner case still shows valid syntax and interface conventions, and complex sequential designs are both harder to generate and harder to validate, so filtering by correctness leaves a corpus of short and simple modules. Correctness is also hard to obtain, since behavior leaves little trace on the surface in RTL, and validating an entire corpus only sorts pairs into passed and failed. We present RTLCurator, which learns a behavior-aware compatibility prior by contrasting each specification with implementations that fail simulation, and calibrates it to a new corpus using a small number of validated pairs. It then constructs the retained subset by balancing alignment, representation coverage, and RTL structural richness. On CodeV and RTLCoder, keeping 80% of the corpus this way improves on training with the full corpus across all reported metrics while validating only 10% of the pool, whereas ranking by the score alone falls below random selection and filtering the whole pool by simulation does no better.

Pretraining data processing is critical to the downstream performance of Large Language Models (LLMs). However, many existing approaches define a fixed processing strategy at the corpus or domain level and apply it uniformly to many examples, without adapting to the needs of each example. We propose DataOrchestra, a framework that unifies different processing operations and orchestrates an example-specific pipeline for each example. Given a chunk of pretraining data, an orchestrator decides whether to drop, untouch, or clean it. For a chunk to be cleaned, it selects one or more downstream operations, ranging from programmatic editing to different forms of LLM-based rewriting. For each rewriting step, it further generates a concrete instruction, which is executed by the corresponding downstream tool model. We pretrain models from 0.5B to 7B from scratch on web data processed by DataOrchestra and observe stable average gains over individual data-processing methods across 11 benchmarks. DataOrchestra is also effective for math continued pretraining and outperforms stronger processing baselines, while reducing processing compute by skipping unnecessary downstream operations.

Advanced Persistent Threats (APTs) remain difficult to detect because only a small fraction of events in large-scale logs are attack-related, and investigation is expensive and hard to scale. Prior machine-learning approaches can reduce analyst workload, but they often rely on heavily curated training data and sophisticated preprocessing pipelines. Building and maintaining such pipelines require substantial domain expertise and engineering cost. Motivated by insights from a study of a strong APT detection baseline, we propose CAPTAIN (Context-Augmented Perplexity-based Threat Activity log detectIoN), a perplexity-based detector that leverages general, pre-trained language models with minimal, domain-agnostic preprocessing, enabling robust scoring of long, minimally processed log entries. CAPTAIN encodes recent history with an encoder model and a Q-Former-style bridge, then injects the compact context tokens into the decoder input so that perplexity reflects temporal context. To improve stability, CAPTAIN additionally applies smoothing filters to the perplexity time series. Across APT-oriented benchmarks, CAPTAIN competes with strong existing baselines and remains robust under substantially less curated inputs, that reduces the development and operational cost of advanced log preprocessing.

Aims: Cardiovascular magnetic resonance (CMR) imaging enables non-invasive assessment of myocardial structure, function, and pathology, but requires substantial experience in interpretation of CMR images that could be supported by artificial intelligence (AI)-based models. However, use of AI models for enhanced CMR reading is limited by labor-intensive data curation, suboptimal model performance, and unclear implementation pathways. Methods and results: We developed an automated data curation pipeline for CMR-based cardiovascular disease (CVD) diagnosis, integrating open-source locally-run large language models (LLMs) to extract diagnostic labels from narrative CMR reports and preprocessing multimodal imaging data, including cine and late-gadolinium-enhancement (LGE) CMR sequences. Three vision foundation models (DINO, VST, UMedPT) were fine-tuned across these modalities in a two-stage approach. The dataset comprised hypertrophic cardiomyopathy (HCM), dilated cardiomyopathy (DCM), ischemic cardiomyopathy (ICM), cardiac amyloidosis (CA), and normal controls (NOR). A total of 988 curated cases were randomly divided into 742 for training and 246 for validation. Fine-tuned AI-models achieved high discriminative diagnostic performance on an independent test set comprising 1067 patients , with individual AUC-ROC values of up to 0.937 for the correct diagnosis of HCM and 0.945 for cardiac amyloidosis. Ensemble strategies combining multiple models and modalities further improved AI-based diagnostic accuracy and robustness, achieving the highest overall diagnostic performance for HCM (AUC=0.959, CI [0.936-0.978]), CA (AUC=0.966, CI [0.939-0.986]), NOR (AUC=0.872, CI [0.852-0.894]), DCM (AUC=0.848, CI [0.808-0.885]) and ICM (AUC=0.840, CI [0.809-0.868]). All training and inference code, along with the trained model weights, are publicly available on https://github.com/sinaamirrajab/CMR_CVD.

Deployed AI systems are often trained from broad candidate data pools, necessitating data curation towards the deployment test distribution. However, standard data curation methods score training-side criteria rather than directly optimizing deployment match. We introduce TTCov (Test-Time Coverage), a data-level test-conditioned curation method that uses test-side information before training instead of updating model weights at inference. TTCov decomposes deployment-conditioned curation into coverage and distribution. To represent coverage, it builds a task Atlas, a collection of LLM-based atomic propositions (APs) describing deployment-relevant concepts, seeded from open task knowledge and expanded with unmatched APs extracted from unlabeled deployment samples. To represent distribution, it instantiates the matched deployment APs with their frequencies, yielding a Knowledge Atlas (K-Atlas) that operationalizes the deployment distribution as a curation target. TTCov then selects a budgeted training set whose deployment APs distribution approximates this target. We apply TTCov towards autonomous driving (AD), keeping adaptation off the inference path while selecting data with greater deployment-relevant coverage, closer K-Atlas matching, and stronger downstream end-to-end driving performance than data-curation baselines, including seamless adaptability to novel domains via city-to-city expansion.

Multimodal Large Language Models (MLLMs) are increasingly deployed for nuanced content safety and moderation tasks, yet they remain vulnerable to adversarial attacks and out-of-distribution edge cases. Traditional active learning and manual annotation fail to scale against the complexity and volume of novel multimodal threats. In this paper, we propose an automated, agentic red-teaming framework that systematically synthesizes difficult examples using an iterative strategy that proposes novel hypotheses as well as mutating on past attempts. Leveraging a multi-agent architecture that consists of a high-reasoning Architect agent, an advanced image generator, and a multi-level verification committee of LLM raters, our system autonomously uncovers boundary-pushing violations and ambiguous policy edge cases without any human intervention. By employing these carefully synthesized adversarial examples as in-context demonstrations via test-time Retrieval, we substantially improve the target model's robustness, reducing the False Negative Rate (FNR) from 41.2% to 24.5% in a public image safety benchmark without relying on any human labeling.

Large language model (LLM) agents require post-training methods that can improve long-horizon decision making from environment feedback. However, existing agentic post-training pipelines often treat data curation as a fixed preprocessing step, focusing mainly on data augmentation while neglecting filtering, refinement, and adaptation to downstream failures. We propose CurateEvo, a failure-driven dynamic evolution framework for agentic post-training data curation. CurateEvo represents the curation strategy as executable code and iteratively rewrites it using failed trajectories from a held-out development set. 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. The evolution process first improves effectiveness by diagnosing recurring failure modes and augmenting, filtering, or refining data accordingly, and then improves efficiency by pruning redundant or low-utility training turns under a cost-aware objective. Experiments on ACEBench-Agent, BFCL-V4, and τ^2-Bench under both labeled and wild-data settings show that CurateEvo consistently outperforms prior curation methods, improving average scores by 3.2 and 2.7 points, respectively. Further analyses demonstrate that CurateEvo is compatible with different post-training recipes and substantially reduces curation overhead.

Public battery aging datasets are a critical asset for advanced health management, but their practical use is often limited by inconsistent formats, unclear schemas, and metadata scattered across repositories and publications. Current curation remains largely manual and hard to reproduce, while general-purpose data integration tools miss the domain-specific semantics of electrochemical time-series data. We present BatteryLake, a governed data lakehouse that turns raw public battery data into benchmark-ready assets through an agentic, physics-grounded curation framework, with three contributions. First, LLM agents extract metadata and synthesize dataset-specific converters, grounding every output in verbatim evidence and abstaining when none supports a value. Second, a human-in-the-loop mechanism frames verification as selective prediction and gates admitted data through 26 schema, statistical, and physical-plausibility rules. Third, we release an open benchmark of 41 datasets from over 25 institutions, with standardized SOH and RUL tasks, three split protocols, and eight baseline model families. The platform, benchmark, and curation protocol are publicly available at https://tianwen1209.github.io/batterylake/.

Recent generative models can produce high-quality synthetic images, offering scalable training training data for data-hungry models. Existing approaches to exploiting this potential typically involve 1) training or fine-tuning generators, or 2) using lightweight post-hoc adaptation like prompt engineering or inference-time guidance, making them generator-specific and expertise-intensive. We study a complementary question: given a fixed pool of generated images, can downstream utility be improved purely by selecting an informative subset? The answer is yes. We show that effective selection must counter a structural bias of modern generators: they tend to over-produce canonical modes of each class while underrepresenting intra-class variation. Building on this insight, we split each real class into a canonical Homogeneous (HO) subset and a non-redundant Heterogeneous (HE) subset, then score synthetic images by a fidelity-diversity criterion that rewards semantic alignment while penalizing canonical redundancy. The method is generator-agnostic and requires no retraining. Across multiple benchmarks, it consistently outperforms state-of-the-art data selection baselines and matches the real-data performance with up to 40% fewer synthetic samples. The same criterion remains effective when applied on top of stronger task-tuned generators, with gains on both classification and segmentation tasks. Post-generation selection is therefore not a substitute for better generators, but a complementary mechanism for improving the utility of synthetic data.

Scaling imitation learning requires large datasets, yet human teleoperation inevitably produces mixed-quality demonstrations containing hesitations and recoveries. Prior frame-level progress reward models supervise on absolute temporal progress proxies that suffer from label noise, or require costly human annotations to define subtask boundaries. We present WARP (Warp-Augmented Relative Progress), a novel fully self-supervised algorithm for learning dense, signed relative progress magnitudes directly from successful demonstrations. WARP generates per-frame progress targets via time-warp augmentations of demonstrations (variable playback speeds and reversals) and we train WARP-RM to predict the normalized elapsed time between input frames. Aggregating these predictions across overlapping windows yields a dense frame-level progress signal. We then introduce WARP-BC, which leverages these scalar reward estimates to upweight high-advantage action chunks during behavior cloning, where chunk-level advantage is obtained by aggregating per-frame rewards. We evaluate our approach on a physical bimanual robot system performing a long-horizon deformable object manipulation task: folding T-shirts from a random crumpled start. To evaluate policy robustness against suboptimal data, we construct training datasets of varying quality using episode length as a proxy for teleoperation sub-optimality. As the dataset is widened to admit more inefficiencies, WARP-BC maintains a 19/20 success rate compared to vanilla BC's collapse to 2/20, improving throughput by up to 18x. Furthermore, we evaluate a bottle-in-bin placement task in the real-world, as well as in a reproducible simulation of the task, where gains in success, speed, and throughput hold under paired significance tests, and we release all simulation code and evaluation artifacts. Project page: https://uynitsuj.github.io/warp-rm/

Supervised fine-tuning (SFT) on a small, high-quality set of long reasoning traces is an effective approach for eliciting strong reasoning capabilities in Large Language Models (LLMs). However, existing methods for curating high-quality SFT data rely heavily on strong reasoning models to filter examples based on diversity and difficulty, making the curation process costly while often yielding suboptimal data quality. In this work, we show that diverse and challenging reasoning examples can be identified using only the initial reasoning tokens. Specifically, we demonstrate that difficult problems can be reliably detected based on the loss of the first 100 reasoning tokens evaluated at a randomly perturbed checkpoint of the pretrained model. We further show that examples exhibiting similar loss patterns over their first 1k reasoning tokens across a small number of perturbed checkpoints extrapolating along the fine-tuning trajectory provably induce similar gradients. We validate our approach through extensive experiments on fine-tuning Qwen2.5-7B and Llama3.1-8B models on the M23K medical reasoning and OpenThoughts-Math datasets. Our method outperforms existing baselines by up to 1.7% while being 91% more token efficient.

Inference efficiency is typically pursued by shrinking the model: distillation, pruning, quantization, and sparse routing each lower per-token cost while treating token count as fixed. But output length has been inflating, and it is precisely the component the standard toolkit leaves untouched. Here, we argue that brevity is the missing inference-efficiency lever, and that pretraining data curation is a practical way to pull it: a model trained on concise, correct data learns to answer in fewer tokens; i.e. it has a lower Cost-of-Pass. We apply our VLM curation pipeline to the MAmmoTH-VL single-image subset, and compare models trained on our curated data, the standard MAmmoTH-VL data, and external open-weight frontier VLMs. On a controlled 20-evaluation set and 14 VLMs at 1B-4B activated parameters, we hold output length fixed with a per-model regression, separating brevity from quality, and price models in FLOPs per correct answer. Curation buys a 35x Cost-of-Pass advantage over the most verbose 4B comparator (Qwen3.5-4B) within $\sim$1 pp of accuracy (0.41 vs 14.58 TFLOPs per correct answer; 0.691 vs 0.704 mean accuracy). Curation also buys a +17.55-percentage-point matched-length accuracy gain over the uncurated baseline that grows with model scale (from +16.7 pp at 1B to +21.2 pp at 4B). This brevity improvement concedes no quality: generic verbosity buys no accuracy at any capability or scale, and the window where reasoning-structured verbosity still earns its tokens shrinks from 4 of 8 capability groups at 2B to 1 of 8 at 4B. Per example, the concise model even reaches correct answers the verbose reasoning model misses, marking reasoning as a distinct curation target rather than something brevity gives up. Inference efficiency in this regime is a tokens-per-correct problem, and brevity is the lever that targets it directly.

Agentic language models dramatically expand the applications of AI yet little is publicly known about how to curate training data for broadly capable agents. Existing open efforts such as SWE-Smith, SERA, and Nemotron-Terminal typically target a single benchmark, leaving open the question of how to train models that generalize across diverse agentic tasks. The OpenThoughts-Agent (OT-Agent) project addresses this gap with a fully open data curation pipeline for training agentic models. We conduct more than 100 controlled ablation experiments to systematically investigate each stage of the pipeline, yielding insights on the importance of task sources and diversity. We then assemble a training set of 100K examples from our pipeline and fine-tune Qwen3-32B on this dataset, which yields an average accuracy of 44.8% across seven agentic benchmarks and a 3.9 percentage point improvement over the strongest existing open data agentic model (Nemotron-Terminal-32B, 40.9%). Moreover, our training data exhibits strong scaling properties, outperforming alternative open datasets at every training set size in compute-controlled comparisons. We publicly release our training sets, data pipeline, experimental data, and models at openthoughts.ai to support future open research on agentic model training.

Automatic review generation is a promising direction for accelerating scientific progress. While most work adopts an end-to-end setup, its fully automated nature makes it less suitable for settings that demand accountability. To better balance automation and accountability, we formalize judgment-grounded expansion, a human-AI collaboration mode where a reviewer provides an evaluative claim and the system expands it into review comment candidate(s). We model it as a structured generate-check-refine process and conduct a user study to collect human-model interaction data. We study two practical challenges for judgment-grounded expansion: scalable evaluation and candidate set curation. We develop methods to simulate the process for large-scale evaluation, and show that conformal prediction is well suited to balancing candidate set size and target coverage. Our work establishes judgment-grounded expansion as a concrete task and provides empirical and methodological foundations for the design of future collaborative review generation systems.

We frame the system as diagnostic data curation for a large audio-language model: before fine-tuning, we probe Qwen3-Omni-30B-A3B-Instruct under normal, empty-audio, and shuffled-audio conditions to identify how the model's answers change when audio evidence is removed or mismatched. These model confusion patterns are used to bucket training samples into text-prior, shuffle-leak, strong audio-dependent, and hard or misleading cases. Our strongest train-only system fine-tunes only on strong-audio items, where the normal audio-question pair is correct but both counterfactual variants fail, plus a small number of empty-audio negatives and a text-only response normalizer for parse-failed generations. On the official development set, the best train-only system reaches 67.27% accuracy after response normalization, compared with 65.90% for our local Qwen3-Omni baseline. Final submissions additionally include models trained using train+development splits and a three-model ensemble.

Web data curation has been widely studied for decoder Large Language Model (LLM) pretraining. Encoders for dense-terminology domains such as medicine, by contrast, are pretrained on small, manually-curated corpora that limit scalability and writing style diversity, a bottleneck even more severe in non-English clinical settings. Whether web-scale data curation also benefits encoder Masked Language Modeling (MLM) in a dense-terminology domain remains an open question. To address this, we introduce two complementary levers. Medical-term density filtering selects documents rich in medical terms. Signal-amplifying rephrasing uses an LLM to rewrite documents into denser variants with broader entity contexts. We instantiate the recipe on French medical NLP. The medical-term density filter outperforms the widely-used educational quality filter on downstream medical tasks, and the two complement each other. Signal-amplifying rephrasing alone improves on raw web data, and mixing it with filtered web data produces the largest gain. The recipe yields FineMed, a French medical pretraining corpus, and DoctoBERT, a state-of-the-art French medical encoder family evaluated on both the public benchmark DrBenchmark and a proprietary clinical Named Entity Recognition (NER) task.

Data curation is a critical part of post-training pipelines for large language models, yet existing tools often treat ingestion, deduplication, synthetic generation, and quality filtering as separate stages. This fragmentation makes it difficult to audit pipeline decisions or understand why individual samples are rejected. CuratorKIT is an open-source Python library that covers this full lifecycle in a single configurable pipeline. The framework is composed of six source format readers and automatic schema detection, a pre-generation data hygiene layer for credentials, PII, and toxic content, eight LLM-powered generation tasks, three complementary quality gates with provenance-exact hallucination verification, structured adaptive recovery, and five training-ready export formats compatible with TRL, Unsloth, and AlignTune. Every pipeline decision is recorded in an append-only per-sample provenance chain, and rejected samples carry structured failure reasons rather than being silently discarded. CuratorKIT supports 100+ LLM providers through LiteLLM, exposes both a Python API and a YAML-driven CLI, and is designed for practitioners who need reproducible, auditable data pipelines at scale .

Scientific discovery workflows often depend on structured curation from the literature. This is difficult for current agents because the key evidence is scattered across long text, dense tables, and figures, and the final records often require reasoning across multiple evidence fragments rather than copying a single span. We study scientific curation from multimodal sources and introduce Beaver, an agent harness that extracts structured information from scientific papers while preserving provenance to the supporting evidence. Beaver combines a frontier agent with multimodal evidence tooling, task scaffolding, and artifact-grounded autoresearch. These components turn curation into a staged, auditable workflow and enable an iterative evaluate--diagnose--revise loop, where persistent run artifacts expose stage-localized failures and guide harness updates. Experiments show that Beaver reaches 81.0 on Gold-Referenced Attribute Score (GRAS), an attribute-level measure of agreement with gold curated records, outperforming frontier agents by over 23 absolute points. Ablations show that task scaffolding, multimodal evidence tooling, and provenance traces each contribute meaningfully to performance, while attribute-level analysis shows the largest gains on high-value attributes that require cross-modal reasoning and normalization. These results show that, for scientific curation from papers with multimodal evidence, harness design is a central determinant of agent performance.

Preference-based post-training has become a central paradigm for aligning language models. A common data-collection strategy is to generate a small set of completions for each prompt and label the resulting comparison pairs. However, human preference labels are often much more expensive than generating additional completions, suggesting a different use of the same labeling budget: generate a larger pool of completions, but label only the most informative comparison pairs. This paper studies which pairs should be compared in preference-based post-training. We formulate comparison curation as a sampling-design problem and evaluate designs by the quality of the final policy under the preference-based post-training objective. We instantiate this framework for Direct Preference Optimization (DPO), analyzing how the choice of labeled pairs propagates through DPO training to downstream policy performance. Our main results provide matching upper and lower bounds on the post-training optimality gap of the DPO-trained policy. The bounds show that comparison selection affects downstream performance through a single design-dependent information matrix, which links label allocation to parameter estimation error and policy suboptimality. This yields an explicit optimization criterion for budgeted comparison curation and motivates practical sampling designs for selecting informative pairs from large generated completion pools. Experiments on synthetic settings and language-model post-training benchmarks show that the proposed designs consistently improve sample efficiency over common comparison-selection heuristics.

In robot imitation learning, influence functions provide a principled approach to quantify each demonstration's effect on robot task outcomes, yet scaling them to billion-parameter Vision-Language-Action (VLA) models is limited by computational and multitask bottlenecks. To this end, we propose ATHENA, an influence function framework tailored for multitask VLA data curation at a billion-parameter scale. Concretely, it leverages the Kronecker structure of linear-layer gradients to reduce projection cost, and approximates dense Hessian inversion with a rank-r Random Truncated Approximation, achieving about a 313.4x speedup in influence computation. Furthermore, ATHENA formulates global and local interactive influence to balance data curation across 50 jointly trained tasks. Extensive evaluations on RoboTwin 2.0 and real-robot deployment, covering 9.34 and 6.90 hours of demonstrations, respectively, show that ATHENA matches or exceeds full-data joint fine-tuning using only 50% of demonstrations in simulation and 66.7% of data across six real-robot tasks. Overall, ATHENA demonstrates its effectiveness for data curation in billion-parameter multitask VLA fine-tuning.

We introduce an in-domain supervised pipeline designed to counter the out-of-distribution performance drop that hampers supervised biomedical NLP models, a problem observed when models trained on pathology reports are moved across cancer registries. Our contribution is a reproducible recipe for training a supervised classifier from routinely collected cancer registry data. It describes how to build the in-domain training set and a production-matched holdout, and to choose operating points that keep the false-negative rate (FNR) very low while keeping reviewer workload manageable. The pipeline standardizes data curation with facility-stratified sampling and separate handling of reports linked to registry cases, and includes a blinded manual audit to estimate positive-case prevalence and label noise. On a 418k-report holdout set, the Kentucky model achieved FNR 0.003 and false-positive rate (FPR) 0.097, improving over the Seattle-trained MOSSAIC OncoID baseline (FNR 0.010, FPR 0.183) and raising F1 from 0.860 to 0.922. In a blinded manual review of 600 reports, estimated positive prevalence declined from 0.500 to 0.398, indicating substantial label noise with errors concentrated in rare primary sites.

LLM agents can improve without weight updates by accumulating natural-language skills from experience, but current systems entrust every decision about which skills to keep and how to apply them to LLM judgment alone. We argue that this conflates two distinct roles: generating a skill from experience is a creative act that judgment handles well, while deciding whether that skill actually helps requires empirical evidence across many tasks. Measuring per-skill causal contributions via randomized masking, we find that skill libraries exhibit pervasive causal heterogeneity: individual skills routinely help on some task types while hurting on others, yet their opposing effects cancel in aggregate, making them invisible to global curation methods. We propose ASSAY, a framework that separates generation from curation: it computes a per-skill causal attribution on a small development set, restructures the library offline, and suppresses skills with negative predicted effect for each test task. Across seven base models spanning four providers and two benchmarks (AppWorld and tau-bench), ASSAY consistently improves over prior skill-curation approaches. On AppWorld's hardest split, DeepSeek-V3 achieves 69.3% task-goal completion (47.4% relative improvement), a new state of the art among all published methods including weight-tuned approaches. On tau-bench retail, GPT-4.1 improves by 8.7% relative, advancing past o4-mini, o1, and GPT-4.5 on the public leaderboard without any weight modification. Ablation traces the dominant gain to per-task masking, confirming that the bottleneck is matching skills to tasks at inference time, not removing bad skills globally. Code is available at https://github.com/aiming-lab/assay.

Large language models (LLMs) are increasingly capable of mathematical problem solving and can even assist with research-level proofs, yet we still lack a scalable and reproducible way to measure step-level reasoning in long proofs across diverse sources. This evaluation gap limits trustworthy AI assistance in proof-certified scientific progress. Existing evaluations often emphasize final answers or rely on costly expert grading, while end-to-end proof generation remains open-ended and hard to verify automatically. We introduce Mask-Proof, a pipeline that turns real proofs into automatically checkable masked-step tasks. It masks key formula steps, provides the necessary surrounding context, and evaluates model reconstructions with an LLM-based equivalence judge using repeated votes for stability. The resulting Mask-ProofBench contains 292 curated problems across diverse research areas. Experiments with 17 models show that reasoning-enhanced models outperform standard models by 12% to 27%. Our evaluator achieves 96.8% agreement with expert annotators, enabling faithful, reproducible, and comparable measurement of step-level mathematical reasoning. Benchmark, annotations, and code are available at https://github.com/weating/Mask-Proof.

Manipulation demonstrations have temporal phase structure, and a natural hypothesis is that demonstration-curation metrics should be applied within phases rather than globally. The idea is to segment each trajectory into phases, score each phase with the metric that is locally most informative, and then aggregate. This follows directly from prior work showing that a single global metric can be the best detector of a defect and yet the worst curator of the resulting policy. We test the per-phase hypothesis on three contact-rich LIBERO pick-and-place tasks with a controlled early-release structural defect, comparing phase-gated curation against the same metrics applied uniformly and against a strong single global metric. Across all three tasks and five random seeds per condition, phase-gated curation is never the best curation strategy, and it is the worst of the three on two of the three tasks (Task 1: 86.0 vs. 92.0 for global; Task 3: 22.7 vs. 48.0 for uniform). We trace the failure to a concrete mechanism. When the defect signal is concentrated in a single phase, rank-aggregating across phases dilutes that signal with uninformative scores from defect-free phases, selecting a worse demonstration subset than simply applying the defect-informative metric everywhere. We further show that the per-phase metric selection does not transfer across tasks, since no phase shares a winning metric between any two tasks, so the selection cannot be reused and must be re-derived per task from a noisy sweep. These results bound a plausible and previously untested method, and they argue that practitioners should prefer identifying a single defect-informative metric over decomposing curation by phase. We release the full pipeline, all metric implementations, and per-seed results.

Long-form video generation requires recurring subjects to remain consistent across various shots, viewpoints, motions, and scene transitions. Existing temporal decomposition methods improve scalability by generating videos shot by shot. However, they mainly focus on optimizing plausible next-shot continuations without verifying whether the historical memory preserves identity-critical subject evidence. Consequently, as generation proceeds, recurring subjects may be diluted, overwritten, or forgotten. In this paper, we propose Memento, a subject-reconstruction-guided framework that treats subject preservation as an explicit identity grounding problem, based on the premise that a memory bank faithfully preserving a subject should support reconstructing that subject from memory alone. Specifically, Memento jointly trains autoregressive next-shot generation with memory-based subject reconstruction, recovering target appearances using historical memory and global story captions. To disentangle long-range subject evidence from short-range cues, Memento introduces a dual-query memory mechanism, where one query retrieves identity-relevant memory and the other selects short-context keyframes for coherent continuation. Additionally, a subject-aware cinematic data pipeline provides precise reconstruction supervision via consistent, pronoun-free subject descriptions. Experiments demonstrate that Memento achieves state-of-the-art performance in long-term subject consistency, cross-shot coherence, and visual quality.

Automated testbench generation has become a critical bottleneck in large language model (LLM)-driven Register Transfer Level (RTL) workflows, where large numbers of candidate designs must be verified rapidly and reliably. Existing prompt-based approaches treat testbench generation as unconstrained code synthesis, yielding stochastic outputs with high token cost, low reproducibility, and insufficient coverage. To address this gap, we present STG, a Structured Testbench Generation framework that exploits the inherent structure of hardware designs to generate deterministic testbenches. As a direct verification tool, STG runs 720x faster than an iterative LLM-based testbench generation flow and higher rate of successful compilation, achieves higher coverage, and reduces false-pass verdicts on incorrect DUTs. STG also helps identify errors in RTL generation benchmarks by exposing faulty benchmark testbenches. As a data curation engine, it is 11x faster than LLM-based filtering on a single CPU core with 127x less energy, and the resulting distilled models provide state-of-the-art performance in our multi-benchmark evaluation. As a test-time scaling oracle, it reduces node count by 14-47%. Our models are available at https://huggingface.co/collections/AS-SiliconMind/siliconmind-v12.

Synthetic post-training pipelines commonly filter generated samples with reward models or holistic LLM judges, yet two practices remain rarely examined together: whether the filtering signal is grounded in the source evidence that induced each generation, and whether rejected samples can be systematically recovered rather than permanently discarded. We present a controlled study of both questions across gate configurations, recovery strategies, and generator scales, using adversarially injected corpora to provide ground-truth failure labels. We find that exact source provenance improves faithfulness gating for stronger judges, that hallucination and reward gates reject largely disjoint sample populations making both necessary, and that an adaptive recovery pipeline combining failure diagnosis with targeted regeneration achieves higher yield, recovery rate, and injection recall than naive resampling. Downstream fine-tuning quality is driven primarily by generator scale, with filtration and recovery conditions contributing meaningfully but secondarily.

Although some very common test beds exist for assessing the performance of clustering methods, large scale benchmarking is typically limited to relatively simplistic simulation set-ups. Here we describe the production and curation of close to 3000 synthetic data sets, derived from more than 200 publicly available data sets; the majority of which arose from real-world applications. By fitting a flexible non-parametric distribution to each base data set we are able to retain much of the nuance in real-world data which is difficult to reproduce in standard simulations, while also producing data sets whose sizes are sometimes substantially greater than the data sets from which they are derived. The synthetic data sets, plus an accompanying R package, are available for download from https://github.com/DavidHofmeyr/ClusBench.

We study whether demonstration-curation metrics that detect defective training episodes also improve the downstream behavior-cloning policy that trains on the curated data. On a contact-rich LIBERO pick-and-place benchmark with a controlled structural defect (early gripper release during the carry phase), we find that the two quantities are sharply decoupled. The metric with the highest defect-detection AUROC (0.804) produces the worst curated policy (13.3% task success), while a metric with a substantially lower AUROC (0.638) produces a policy that nearly matches the oracle trained on ground-truth clean data (90.0% vs. 93.3%). We further show that five of the seven metrics we evaluate exploit episode length as a trivial proxy for the defect label, a confound that inflates reported AUROCs to near-perfect values and disappears once episode length is controlled. Across all conditions, the contaminated baseline succeeds on only 3.3% of rollouts, and the two best curation methods close this to within 3 percentage points of the 93.3% oracle ceiling. Our results argue that curation methods should be evaluated by the policy they produce, not the defects they flag, and that any curation benchmark must control for episode length before reporting detection accuracy. We release the testbed, all metric implementations, and the evaluation pipeline.

Self-supervised data curation provides a pathway to scaling and improving the generalization capabilities of machine learning models. By leveraging self-supervised learning (SSL) for data curation, the demand for massive training datasets required by foundation models can be effectively met. SSL greatly alleviates the costs associated with annotation and manual dataset curation while minimizing the need for human oversight. However, the integrity of SSL-curated datasets must be rigorously checked, as reliance on anonymous and unvetted external sources can substantially increase the risk of data poisoning. In this paper, we propose a Poisoned Data Detector (PDD), an active defense mechanism designed to ensure the integrity of SSL-curated datasets prior to foundation model training. PDDs are designed using a combination of the pretrained ImageBind model and traditional classifiers, including Random Forest (RF), k-Nearest Neighbors (KNN), Naive Bayes (NB), and Support Vector Machines (SVM). We rigorously evaluated PDDs using 176,200 images from three diverse datasets and three different adversarial attacks encompassing both in-distribution and out-of-distribution scenarios. Notably, SVM-PDD achieves superior performance for both in-distribution (Set3-Set5) and out-of-distribution (TrueFace and 140K RealFace) datasets. Our design demonstrates strong scalability and enables the rapid integration of new adversarial attack detectors through an ensemble approach.

Electronic health record (EHR) notes are dense medical documents containing large amounts of information, often filled with complex medical jargon. Highlighting all details in EHRs helps reduce the likelihood of missing crucial information by drawing attention to key content. This study proposes the design of a Cardiology Interface Terminology (CIT) to accurately highlight all details in EHR notes of cardiology patients. We introduce an innovative Machine Learning (ML) technique for the design of CIT. The ML technique requires training data. Manual preparation of such training data is time-consuming and expensive. The process of the CIT design includes three phases. In the first two phases, we innovatively derive a training data CIT to be used by the third phase, ML technique. We start by designing an initial CIT, composed of several components: the cardiology-related sub-hierarchies of SNOMED, other SNOMED concepts mined from EHRs of build set, and necessary components of terms e.g., medical abbreviations and medications. Utilizing an iterative process, fine-grained phrases containing initial CIT concepts are extracted from build set as CIT concept candidates. The candidate concepts are semi-automatically reviewed before being added to CIT, yielding the training data CIT, TCIT. In the third phase, a ML model is trained with TCIT to identify candidates fitting to be concepts in the CIT. This model is used to extract further concepts from build set, yielding the final CIT. The final CIT is then used to highlight the test set and evaluate the extent to which it captures details in an unseen EHR dataset. For this purpose, four evaluation metrics, coverage, breadth, completeness, and conciseness are used. The highlighted test set has a coverage of 74.21%, with a breadth of 1.68. For 20 random notes in test set, the average completeness is 98.2% and average conciseness is 84.2%.

AgriGov is a curated, trilingual (English-Hindi-Marathi) dataset designed to address the scarcity of domain-grounded multilingual resources for agricultural policies and farmer welfare schemes. Initially, we collected and structured data from 50 government schemes sourced from trusted portals using automated scraping techniques, organizing it into predefined semantic fields (e.g., title, eligibility, application process, documents, exclusions). Translations were performed using a pipeline combining Google Translate API, MarianMT, and human post-editing, resulting in a domain-specific Hindi-Marathi dataset comprising approximately 2100 source segments. To enhance coverage, we augmented this dataset with sentences from the Samanantar corpus, leading to approximately 8,000 sentence-aligned Hindi-Marathi parallel pairs. The dataset now offers robust resources for fine-tuning machine translation models in this domain. AgriGov is designed for applications in domain-adaptive machine translation, question answering, information retrieval, and summarization systems. Its key contribution is a schema-driven, human-corrected multilingual alignment pipeline that ensures domain fidelity, provides provenance, and supports reproducible experiments, enabling retrieval-augmented applications for farmer-facing tools.

Dominant web data curation pipelines for pretraining collapse document quality into a single composite score, systematically missing high-value content along dimensions the scorer underweights. We present a taxonomy-driven framework that recovers this value by filtering along semantically meaningful dimensions that composite scores fail to capture. First, building on the ESSENTIAL-WEB taxonomy, we introduce two novel dimensions: timeliness and cultural specificity, both of which show low pairwise NMI with existing ones. We annotate 14M documents using Qwen2.5 32B and distill into a lightweight 0.5B model. To enable rapid corpus-wide annotation, we additionally train a 73M multi-task MLP on E5 embeddings, achieving 50x inference throughput. Second, to navigate the combinatorial explosion of filter configurations, we introduce a compute-efficient two-pass framework: Pass 1 identifies the strongest dimension signals at small scale; Pass 2 constructs and evaluates conjunctive and disjunctive compound filters from the top performers - identifying high-performing configurations at a fraction of full scaling-law cost. Applying the selected filters to deprioritized web data, taxonomy-filtered subsets outperform their unfiltered baselines and even surpass the highest-quality tier. On mid-tier data, our best filter improves over its unfiltered baseline by 12.1% on reasoning, 9.5% on coding, and 2.0% on knowledge benchmarks, exceeding unfiltered top-tier data by 6.7% on reasoning and 13.7% on coding. Furthermore, filtered data from two tiers below the typical production threshold improves by 22.3% on reasoning and 19.5% on coding over its unfiltered baseline, surpassing top-tier data on coding benchmarks. These results establish that vast latent value remains locked in deprioritized web data, and that multi-dimensional taxonomy filtering is a principled, compute-efficient key to unlocking it.

Curating training data is among the most consequential yet labor-intensive parts of modern AI development: practitioners iteratively propose, implement, evaluate, and revise data policies against noisy benchmark feedback. We ask whether generalist coding agents can automate this data-curation loop. We introduce Curation-Bench, an agent-centric benchmark that fixes the model, training recipe, and evaluation suite while giving agents command-line access to inspect data, implement policies, submit them to a fixed training/evaluation pipeline, and revise. In a vision-language instruction-tuning instantiation, out-of-the-box agents reach strong published data-selection baselines within ten iterations. However, trajectory analysis reveals a persistent execution-research gap: agents mainly tune local policy variants rather than explore new policy families, even when given strategy guides and paper references. Scaffolds requiring each iteration to cite, instantiate, and adapt a prior method shift agents toward method-guided exploration. The scaffolded agent autonomously composes -- without human design input -- a data-selection policy that outperforms strong published baselines at one-tenth their data budget. Overall, current agents can run the curation loop, but reliable data research requires scaffolded method adaptation, not open-ended prompting alone. Code and benchmark are open-sourced.

The supply of high-quality training tasks is a central bottleneck for reinforcement learning from verifiable rewards (RLVR) on agentic language models. Each task requires a sandboxed setup, a prompt, and a hand-authored reward function, and only tasks that pass a quality bar produce useful training signal. Hand-curation at this quality bar does not scale economically to the task counts effective RL training requires, and the substitution rate between automatically generated task variants and human-authored ones is not yet established. We investigate using pre-specified, gate-filtered augmentations of a small hand-authored base as a substitute for additional human curation during RLVR. We formalize the cost-adjusted trade rate $ρ_{\text{cost}}$ between augmented and human-authored tasks, measure it through a controlled ablation across training corpora with varying augmentation share, and characterize the end-to-end economics of the augmentation pipeline. Substituting augmented content for additional human-authored tasks retains aggregate held-out generalization on a ten-benchmark suite spanning code, instruction following, reasoning, and multi-turn agentic function-calling. The cost-adjusted trade rate $ρ_{\text{cost}}$ between gated synthetic and human-authored RLVR tasks stays in $[1.4\times, 11.6\times]$ across the plausible $c_{\text{human}}/c_{\text{aug}}$ range.

Recent unified audio generation models can support diverse tasks across speech, sound effects, and music, but most of them still focus on isolated task-level synthesis. However, real video production often requires multiple components of a complete audio track to be generated jointly and consistently for the same video. We present Foley-Omni, a unified multimodal audio generation model that extends isolated task-level synthesis to complete video soundtrack generation by jointly modeling speech, sound effects, and music within a shared latent generation process. To support training and reproducible evaluation, we develop an audiovisual data curation pipeline and introduce V2ST-Bench, a benchmark for holistic video soundtrack generation evaluation. Experiments show that Foley-Omni achieves competitive performance with expert systems on individual synthesis tasks, while improving speech intelligibility, audiovisual consistency and perceptual quality for mixed soundtrack generation.

Large Language Models (LLMs) have revolutionized various fields, yet their training efficiency is heavily reliant on effective data curation. While data selection has been widely studied, the strategic data organization for enhanced training remains an underexplored area, particularly since current LLMs are often trained for only one or a few epochs. This paper systematically explores the influence of data organization on LLM training by reusing pre-computed sample-level scores originally generated for data efficiency, thereby incurring minimal additional computational overhead. We identify and formalize four key guidelines for optimizing data organization: Boundary Sharpening, Cyclic Scheduling, Curriculum Continuity, and Local Diversity. Guided by them, we introduce two novel data ordering methods termed STR and SAW. Extensive experiments across different model scales and data sizes, encompassing both pre-training and SFT stages, validate the effectiveness of our summarized guidelines. They also demonstrate the robustness of our proposed data ordering methods in enhancing the stability and performance of LLM training. Github Link: https://github.com/microsoft/data-efficacy/

Retrieval-augmented LLM agents increasingly rely on curated skill banks: collections of reusable textual principles that guide decision making on complex tasks. Existing approaches typically expand these banks in an append-only fashion, continuously adding new skills without removing redundant, outdated, or harmful ones, resulting in inefficient and poorly curated repositories. In this paper, we formulate the skill bank curation as a constrained multi-objective problem: a desirable bank must be useful for the agent, diverse in its content, and provide good coverage of the query distribution. To this end, we introduce SkillBrew, a multi-objective curation framework that formalizes skill bank curation as Pareto-aware optimization under a utility constraint, and solves it via a bi-level propose-then-verify loop. We evaluate our approach on two public benchmarks. Our findings suggest that treating skill banks as objects of principled curation, rather than ever-growing append-only logs, is an important step toward building self-improving LLM agents.

Foundation models are increasingly trained on synthetic data generated by prior model iterations rather than exclusively on real data. This self-consuming training paradigm can lead to model collapse, divergence, or bias amplification. Recent work (Ferbach et al., 2024) shows that incorporating human curation into the loop can steer a self-consuming model toward human-aligned behavior, but these analyses focus on a single, isolated model that solely consumes its own outputs. In practice, however, models often interact and train on input-output pairs produced by other models. This paper studies self-consuming training in the multi-model regime. We first formalize a framework for interacting self-consuming models and characterize when the resulting dynamical system converges to a stable point. We then examine how human curation of one model affects its own alignment (self-influence) and how such effects propagate to other models (cross-influence). Unlike isolated settings where human curation always enhances model alignment, we show that cross-model interactions can dampen or even invert this effect, ultimately degrading long-term alignment.

Linking free-text phenotype descriptions to ontology terms, typically referred to as phenotype annotation, is essential for the cross-study integration of comparative morphological data. This labor intensive process has heavily relied on highly trained human experts, which makes it challenging to scale and thus a key bottleneck. Dahdul et al. (2018) established a Gold Standard (GS) of Entity-Quality (EQ) annotations across seven phylogenetic studies and used it to evaluate three human curators and the Semantic CharaParser NLP tool with ontology-based semantic similarity metrics; they reported that machine-human consistency was significantly lower than inter-curator (human-human) consistency. Here we revisit that benchmark with five frontier hosted LLMs from Anthropic and OpenAI, each operating as an "agentic curator" within a self-contained workspace that supplies the source publication PDF, the same annotation guide used by the original human curators, the four project ontologies (UBERON, PATO, BSPO, GO), and a validation script. Evaluated against the same Gold Standard, every agent fell within the range of inter-curator variability of the three trained human biocurators of the original study; the best performing agents approached but did not reach the best performing human curator. Agents substantially outperformed Semantic CharaParser on all four metrics.

Post-training via Supervised Fine-Tuning (SFT) and Reinforcement Learning (RL) is crucial for enhancing reasoning in Multimodal Large Language Models (MLLMs), yet existing paradigms often reach a performance bottleneck due to the limitations of static data. While current methods leverage self-reflection or self-evolution to push these boundaries, they still suffer from cognitive drift and hallucinated reasoning paths caused by low-quality synthetic data. To address these challenges, we propose Anchor Evolution (AnE), a new paradigm that integrates truth-anchored data curation and model evolution, achieving faithful and steady performance gains at the reasoning frontier. Specifically, we propose Truth Anchor Expansion, which pinpoints the model failing frontier via trajectory rollouts and leverages ground-truth databases to retrieve high-fidelity anchors for faithful data curation. Subsequently, we introduce the Scaffold-Stripping Mechanism to internalize reasoning capabilities. This mechanism first anchors reasoning paths via scaffold-augmented supervision to mitigate the learning complexity and distribution drift of direct SFT on raw data, then leverages RL to strip the scaffold template, thereby effectively transitioning the reasoning paths into intrinsic model capabilities. Experimental results on multimodal reasoning benchmarks show that our method substantially advances the model performance frontier, improving the base model by 10.3% across eight multimodal benchmarks and achieving state-of-the-art results. The code will be made publicly available.

High-quality training datasets are essential for the performance of neural networks. However, the audio domain still lacks a large-scale, strongly-labeled, and single-source sound event dataset. The FSD50K dataset, despite being relatively large and open, contains a considerable fraction of multi-source samples where background interference or overlapping events could limit the usefulness of the data. To address this challenge, we introduce a data curation framework designed for large-scale open audio corpora. Our approach leverages a generative diffusion model to synthesize clean single-class events to construct controlled noisy mixtures for supervision. We subsequently employ a pre-trained audio encoder coupled with a discriminative classifier to automatically identify and filter out multi-source samples. Experiments show that our framework achieves strong performance on a human expert-curated test set. Finally, we release FSD50K-Solo, a model-curated subset of FSD50K containing single-source audio samples identified by our method. Beyond FSD50K, our method establishes a scalable paradigm for curating open source audio corpora.

Existing reasoning data curation pipelines score whole samples, treating every intermediate step as equally valuable. In reality, steps within a trace contribute very unevenly, and selecting reasoning data well requires assessing them individually. We present GRACE, a gradient-aligned curation method that views each reasoning trace as a sequence of optimization events and scores every step by two complementary signals: its alignment with the answer-oriented gradient direction, and its consistency with the preceding reasoning trajectory. Step-level scores are aggregated into a sample-level value for subset selection, using only the model's internal optimization signals and no external reward models or step annotations. To make this scalable, GRACE introduces a representation-level gradient proxy that estimates step-level alignment from token-level upstream signals in a single forward pass. Post-training Qwen3-VL-2B-Instruct on MMathCoT-1M, GRACE reaches 108.8% of the full-data performance with 20% of the data and retains 100.2% with only 5%, with subsets that transfer effectively across model backbones.

Data curation has shifted the quality-compute frontier for language-model and contrastive image-text pretraining, but its role for vision-language models (VLMs) is far less established. We ask how far data curation alone can take VLM performance, holding architecture, training recipe, and compute fixed and varying only the training data. Our pipeline, applied to the MAmmoTH-VL single-image subset, lifts performance by +11.7pp on average across 20 public VLM benchmarks (spanning grounding, VQA, OCR/documents, captioning, spatial/3D, counting, charts, math, brand-ID, and multi-image reasoning) and by +11.3pp on average across all nine capability axes of DatBench, our high-fidelity VLM eval suite. At 2B, our curated model surpasses InternVL3.5-2B by 9.9pp at ~17x less training compute and closes the gap to Qwen3-VL-2B to within 1.8pp at ~87x less compute, from pretraining alone. Beyond accuracy, curation delivers four further properties: (1) Reliability: per-capability std across training seeds drops by ~67% and the lift survives a 4k-to-16k context-length sweep; (2) OOD generalization: the 9-eval OOD average rises by +7.2pp, and multi-image BLINK rises by +3.09pp despite single-image-only training, with Visual Correspondence gaining +11.8pp; (3) Behavioral gains beyond benchmarks: across ~1,100 open-ended queries the curated 2B is more honest and more specific than the matched-compute baseline, and more concise and less refusal-prone than a frontier 2B reference; (4) Pareto-dominance on inference cost: at every scale (1B, 2B, 4B) the curated model raises accuracy while lowering response FLOPs vs. the matched-compute baseline, and the curated 4B matches near-frontier accuracy at 3.3x lower response FLOPs than Qwen3-VL-4B. Data curation is a high-leverage tool for building better VLMs, reaching near-frontier accuracy at up to ~150x less training compute.

Scientific data processing often requires task-specific algorithms or AI models, creating a barrier for domain scientists who need to analyze their data but may not have extensive computing or image-processing expertise. This barrier is especially pronounced when data are noisy, have a high dynamic range, are sparsely labeled, or are only loosely specified. We introduce CVEvolve, an autonomous agentic harness with a zero-code interface for scientific data-processing algorithm discovery. CVEvolve combines a multi-round search strategy with tools for code execution, evaluation implementation, history management, holdout testing, and optional inspection of scientific data and visual outputs. The search alternates between discovery and improvement actions, and uses lineage-aware stochastic candidate sampling to balance exploration and exploitation. We demonstrate CVEvolve on X-ray fluorescence microscopy image registration, Bragg peak detection, high-energy diffraction microscopy image segmentation, and hybrid analytical-learning-based affine registration. Across these tasks, CVEvolve discovers algorithms that improve over baseline methods, while holdout test tracking helps identify candidates that generalize better than later over-optimized alternatives. These results show that zero-code, autonomous LLM-powered algorithm development can help domain scientists turn unstructured scientific image data into practical algorithms and downstream scientific discoveries.

Predictive models in biomedicine depend on structured assay data locked in the text, tables, and supplements of primary publications. This bottleneck is especially acute in targeted protein degradation (TPD), where each assay record must combine compound identity, degradation target, recruiter, assay context, and endpoint values reported across sections, tables, and supplementary files. Inconsistent compound identifiers and incomplete or implicit assay context further demand domain-specific logic that generic LLM pipelines do not provide. Existing molecular glue and PROTAC databases are manually curated and often lack the experimental context required for downstream modeling. We formulate TPD database extraction as a domain-specific curation task and present an expert-in-the-loop LLM workflow, evaluated through a triangular comparison among LLM predictions, standardized baseline records, and expert-annotated ground truth. A lightweight cross-validated prompt-refinement module adapts extraction instructions from scarce expert annotations. With only seven annotated molecular glue publications, the workflow achieved record-level $F_1 = 0.98$ and transferred to PROTACs by terminology substitution alone, maintaining record-level $F_1 > 0.93$. Applied at scale, it expanded molecular glue and PROTAC databases by 81% and 92% records, respectively, with 92% and 82.5% of newly recovered records validated as correct upon expert review. The workflow also recovered kinetic and assay-context information essential for cross-study potency comparison and condition-aware degradation modeling. We release the workflow, prompts, evaluation code, and extracted datasets as resources for TPD data curation and AI-assisted scientific curation more broadly.

Manually curated biomedical repositories -- spanning bioactivity, genomics, and chemistry -- are expensive to maintain, lag behind primary literature, and discard experimental context, obscuring nuances needed to assess data correctness and coverage. We show that PubMed itself can be autonomously and cost-effectively turned into structured datasets that are larger, more nuanced, and more accurate than the curated databases they replace. We present three coupled contributions: (1) an LLM-based entity-tagging pipeline, grounded in nine biomedical ontologies, that tags 4.5B entities across 19 categories in a 22.5M-paper, 2.5T-token PubMed corpus; (2) hybrid sparse-dense retrieval supporting entity-filtered semantic queries over the tagged corpus; and (3) Starling, a multi-agent deep research system that, given only a natural-language task description, designs precision- and recall-targeted retrieval filters, induces an extraction schema, and emits structured records with nuance-rich fields and supporting passages. Across six tasks -- blood-brain barrier permeability, oral bioavailability, acute toxicity (LD50), gene-disease associations, protein subcellular localization, and chemical reactions -- Starling produces ~6.3M records (91K-3M per task); several are, to our knowledge, the largest public datasets for their property. Frontier-model rejection of our extractions is 0.6-7.7% across tasks, far below error rates we measure on widely used curated counterparts (e.g., 16.5% on BBB_Martins, 7.3% on Bioavailability_Ma). Beyond scale and accuracy, the supporting passages carry nuance tabular databases discard -- e.g., oral bioavailability may depend on fed vs. fasted state. Together, the corpus, retrieval, and agent establish a foundation for AI-driven therapeutic design. Code and datasets: https://github.com/starling-labs/starling.