Academic Intelligence · Curated Daily

Explore the Frontier of Global Academia

AcademicHub aggregates real-time literature from top journals and preprint platforms. Build your personal research radar and let large language models compile cross-disciplinary analysis briefings automatically.

01.
arXiv (CS.CL) 2026-06-16

Mapping Geopolitical Bias in 11 Large Language Models: A Bilingual, Dual-Framing Analysis of U.S.-China Tensions

Large language models are how hundreds of millions of people now encounter contested political questions, raising a subtle measurement problem: a model that simply agrees with whatever it is told can masquerade as biased, contaminating any claim that models hold political opinions. We address this by importing balanced keying from survey psychometrics, posing each proposition and its swapped reverse and signing the response so acquiescence cancels and genuine conviction accumulates. The result is a reproducible, quantitative instrument that maps geopolitical stance across 11 models and 2 languages (19,712 responses). Developer origin, query language and issue domain emerge as three near-equal, additive factors; every model, including those built in the United States, leans more Pro-China in Mandarin; and two models with identical agreement bias are told apart, one neutral, one biased. We release it as an open, interactive tool that extends to any contested-opinion domain.

02.
arXiv (CS.AI) 2026-06-17

Skill-Constrained Model Predictive Control for Resilient Manufacturing Supply Chains

arXiv:2606.17269v1 Announce Type: new Abstract: In skill-constrained production-inventory systems, the qualified human capacity available tomorrow depends on training decisions made today: production requires certified workers, certifications decay unless maintained, and training consumes the same scarce worker hours that production needs now. We study a closed-loop skill-constrained model predictive controller that, at every shift, solves a finite-horizon mixed-integer program over production, inventory, backlog, and training, with binary predicted certification, hard production eligibility, and an interpretable terminal value that prices certified-capacity gaps at the horizon boundary; only the first-period action is applied before replanning. On synthetic, seed-controlled SkillChain-Gym scenarios - announced and surprise new-skill shocks, demand shocks, absenteeism, forecast- and availability-quality modes, capacity-boundary and training-rate sweeps, and negative controls - we evaluate the controller against production-only and maintenance-only ablations, static cross-training insurance plans, and a strong reactive heuristic, under an ex-ante locked configuration and paired statistics. The result is regime dependence, not superiority: no policy class dominates. Predictive control helps when skill or labor bottlenecks are forecastable early enough for training to complete; lean static insurance remains hard to beat under surprise shocks, near the demand-capacity boundary, and wherever pre-shock slack makes insurance cheap. Attribution ablations separate certification maintenance, re-acquisition of lapsed certifications, and greenfield skill acquisition. Forecastability, not adaptivity per se, decides when predictive control pays.

03.
arXiv (CS.AI) 2026-06-18

Sparsity Curse: Understanding RLVR Model Parameter Space from Model Merging

arXiv:2606.18521v1 Announce Type: cross Abstract: Reinforcement Learning with Verifiable Reward (RLVR) has emerged as a powerful post-training paradigm that surpasses Supervised Fine-Tuning (SFT) in eliciting reasoning intelligence and resisting catastrophic forgetting. Recent studies further reveal that RLVR induces highly sparse and off-principal parameter updates compared to SFT. This naturally raises the question: does such sparsity make RLVR models more amenable to model merging? If so, model merging would offer a scalable, training-free path to aggregate diverse reasoning capabilities from independently trained RLVR models. Surprisingly, we find the opposite, uncovering a sparsity curse: the sparse RLVR updates are spread farther apart in parameter space, forming near-orthogonal shortcuts that make aggregation inherently fragile. This is likely rooted in the stochasticity of RL optimization and the diversity of emergent reasoning patterns. Unlike SFT models that converge to shared, flat basins and merge naturally, RLVR models suffer severe degradation under standard merging methods. Through systematic empirical analysis of the update geometry, we characterize the mechanisms behind this failure and propose Sensitivity-aware Resolving Merging (SAR-Merging), a merging recipe tailored for the unique structure of RLVR parameter spaces. SAR-Merging resolves conflicts in overlapping update regions via Fisher Information-based sensitivity arbitration, followed by magnitude-aware sparsification and rescaling to preserve fragile reasoning pathways. Experiments on mathematical and coding benchmarks demonstrate that SAR-Merging substantially outperforms existing merging methods on RLVR models, enabling both single-task enhancement and multi-capability fusion.

04.
arXiv (CS.LG) 2026-06-11

Trajectory Geometry of Transformer Representations Across Layers

arXiv:2606.09287v2 Announce Type: replace Abstract: Understanding how transformer representations evolve across layers, not merely what they encode, remains an open problem in mechanistic interpretability. We recast the transformer forward pass as a discrete population trajectory through a high-dimensional representation manifold, drawing on geometric tools from computational neuroscience. Rather than probing for pre-specified features, we characterize trajectory geometry using five metrics computed directly in the ambient space: trajectory length, curvature, a semantic convergence index, layerwise cosine similarity, and representational stability. Across three model families (GPT-2, TinyLlama, Qwen2.5) and five controlled prompt families, we report four findings. First, semantically related prompts converge significantly in middle-to-late layers (peak CI 0.41–0.58, p

05.
arXiv (CS.LG) 2026-06-17

Gradual Fine-Tuning for Flow Matching Models

arXiv:2601.22495v2 Announce Type: replace Abstract: Fine-tuning flow matching models is a central challenge in settings with limited data, evolving distributions, or computational constraints. While recent work has produced significant advances, particularly in the area of reward-based fine-tuning, current methods fail to demonstrate both theoretical correctness as well as strong empirical results in terms of stability, efficiency, and diversity preservation. In this work, we propose Gradual Fine-Tuning (GFT), a simple yet principled annealing-based framework for fine-tuning flow generative models when only samples from the target distribution are available. For stochastic flows, GFT defines a temperature-controlled sequence of intermediate objectives that smoothly interpolate between the pretrained and target drifts, provably approaching the true target as the temperature approaches zero. We analytically demonstrate that sample generation after GFT can be made substantially more efficient with the use of arbitrary (e.g., optimal transport) couplings, as well as by utilizing few-step inference methods. Empirically, GFT significantly improves convergence stability, while maintaining or improving generation quality, training speed, and generation diversity compared to other fine-tuning methods. Our results position GFT as a simple yet theoretically grounded and practically effective alternative for scalable adaptation of flow matching models under distribution shift.

06.
arXiv (CS.CL) 2026-06-11

LLMpedia: A Transparent Framework to Materialize an LLM's Encyclopedic Knowledge at Scale

Benchmarks like MMLU suggest flagship language models approach factuality saturation above 90\%. LLMpedia shows this picture is incomplete. We materialize ${\sim}$1.3M encyclopedia articles entirely from parametric memory across three model families, then audit every claim against Wikipedia and curated web evidence. For \texttt{gpt-5-mini}, the verifiable true rate is 68.4\% on Wikipedia-covered subjects - more than 21\,pp below MMLU - and the gap is driven by unverifiability (30.5\%), not refutation (1.2\%). Beyond Wikipedia, frontier articles audited against curated web evidence reach 57.6\%; Wikipedia covers only 56.7\% of model-surfaced subjects, and three model families overlap in just 7.3\% of subject choices. In a retrieval-trap benchmark inspired by prior analysis of Grokipedia, LLMpedia is more factual at roughly half the textual similarity to Wikipedia. Every prompt, article, and verdict is released. Data, code, interface: https://llmpedia.net.

07.
arXiv (CS.AI) 2026-06-16

Parallel Test-Time Scaling with Multi-Sequence Verifiers

arXiv:2603.03417v2 Announce Type: replace-cross Abstract: Parallel test-time scaling, which generates multiple candidate solutions for a single problem, is a powerful technique for improving large language model performance. However, it is hindered by two key bottlenecks: accurately selecting the correct solution from the candidate pool, and the high inference latency from generating many full solutions. We argue that both challenges are fundamentally linked to verifier calibration, as a well-calibrated verifier improves answer selection and enables early-stopping strategies to reduce latency. However, existing non-generative verifiers are limited as they score each candidate in isolation, overlooking rich contextual information across the set of candidates. To address this, we introduce the Multi-Sequence Verifier (MSV), a lightweight verifier that predicts each candidate's correctness conditioned on the full sampled set. MSV achieves improved calibration, which directly enhances best-of-N selection performance and empowers a novel early-stopping framework. Across challenging mathematical reasoning benchmarks, MSV improves best-of-64 accuracy by up to 6\% relative to strong baselines, and in the early-stopping setting reaches the same accuracy as baselines with less than half the latency.

08.
arXiv (math.PR) 2026-06-11

On multidimensional infinite dihedral group extensions of Gibbs Markov maps

arXiv:2601.08961v2 Announce Type: replace-cross Abstract: We obtain a local central limit theorem for cocycles associated with a class of non abelian and non compact group extensions of Gibbs Markov maps. This class consists of multidimensional infinite dihedral groups. Unlike in the set up of the random walks on groups, we cannot use the convolution of measures on the group and instead we resort to an approach based on irreducible representations. Depending on the dimension of the group, we obtain either mixing, and thus ergodicity, or dissipativity. Also, we obtain the asymptotics of the first return time of the group extension to the origin.

09.
arXiv (quant-ph) 2026-06-12

Scalar Quantum Fields: Theory Space and its Geometry

arXiv:2606.12580v1 Announce Type: cross Abstract: Scalar fields provide perhaps the simplest playground in which to develop our understanding of quantum field theory. In this lecture, we consider what it means to write down a scalar quantum field theory and how we can give geometrical interpretations to the space of such theories: the theory space.

10.
arXiv (CS.CL) 2026-06-16

MAWARITH: A Dataset and Benchmark for Legal Inheritance Reasoning with LLMs

Islamic inheritance law is challenging for large language models because solving inheritance cases requires complex, structured, multi-step reasoning and the correct application of juristic rules to compute heirs' shares. We introduce MAWARITH, a large-scale annotated dataset of 12,500 Arabic inheritance cases for training and evaluating models on the full reasoning chain: (i) identifying eligible heirs, (ii) applying blocking (\d{hajb}) and allocation rules, and (iii) computing exact inheritance shares. To the best of our knowledge, MAWARITH is the first Arabic corpus and benchmark designed for end-to-end Islamic inheritance reasoning. Unlike prior datasets that restrict inheritance case solving to multiple-choice questions, MAWARITH supports the full reasoning chain and provides step-by-step solutions with justifications grounded in classical juristic sources and established inheritance rules, as well as exact share calculations. This enables models to learn how to generate detailed, step-by-step responses to user queries that reflect real-world Islamic inheritance cases. To evaluate models beyond final-answer accuracy, we propose MIR-E (Mawarith Inheritance Reasoning Evaluation), a weighted multi-stage metric that scores key reasoning stages and captures error propagation across the pipeline. We evaluate six large language models in a zero-shot setting. A commercial model achieves about 90\%, whereas all evaluated open-source models remain below 50\%. Our error analysis identifies recurring failure patterns, including scenario misinterpretation, errors in heir identification, errors in share allocation, and missing or incorrect application of key inheritance rules such as \textquotesingle awl and radd. The MAWARITH dataset is publicly available at https://gitlab.com/nlpresearcher/mawarith.

11.
arXiv (CS.CL) 2026-06-19

Characterizing Narrative Content in Web-scale LLM Pretraining Data

The narrative composition of web-scale LLM pretraining corpora remains largely unexplored even though narrative is a fundamental mode of human communication. We present the first fine-grained study of narrative features in Dolma, a 3-trillion-token open pretraining corpus. Drawing on narrative theory, we design a framework spanning three core narrative elements (agency, setting, and events) operationalized as 11 interpretable dimensions. After sampling and annotating a diverse set of 400 passages, we finetune and validate NarraBERT, a RoBERTa-based model for fine-grained narrative prediction. We apply NarraBERT to 3M passages, resulting in a new dataset, NarraDolma. We find (i) narrative structure is measurable at scale across extremely heterogeneous data, (ii) we uncover a continuous, multidimensional narrative structure underlying web text, and (iii) narrative qualities are unequally distributed across pretraining sources and topics in ways that current curation practices neither measure nor account for. Our framework, dataset, and analyses provide a foundation for understanding how narrative qualities are distributed in LLM pretraining data and for studying how data composition affects narrative reasoning tasks. We publicly release NarraDolma and NarraBERT.

12.
arXiv (CS.AI) 2026-06-12

Deployment-Centered Evaluation: Predicting Query-Level Rejection Risk in a Clinical LLM System

arXiv:2606.12702v1 Announce Type: new Abstract: Large language models (LLMs) are increasingly integrated into clinical systems, making it essential to evaluate the real-world utility of these systems. However, static benchmarks tend to measure correctness rather than user acceptance, aggregate performance across queries, and require densely annotated datasets – leading to major blind spots for evaluating clinical systems. In this work, we perform a deployment-centered evaluation of an LLM system embedded within electronic health records at an academic medical center, where user feedback is sparse but closely reflects the deployment conditions. Specifically, we train a pre-response classifier that estimates the risk that a future interaction will result in the user rejecting the LLM response, based on query content and deployment-specific context available before generation. We conduct a prospective analysis of our model over 4.5 months of user feedback, finding that our prediction model achieves an AUROC of 0.719. Further, we estimate the benefit of such predictions in two downstream use cases (guardrail triggering and abstention). Our key conceptual insight is that making use of deployment-specific context (i.e., the provider type, department name, language model used for response), as opposed to only query content, improves the ability to predict whether the user will reject the system output. Altogether, our empirical case study demonstrates the feasibility of predicting user rejection using deployment-specific context, opening the door to targeted guardrails.

13.
arXiv (CS.CV) 2026-06-11

Towards Fully Automated Exam Grading: Fairness-Aware Recognition of Handwritten Answers with Foundation Models

Correcting handwritten exams by hand is time-consuming and error-prone, particularly for large cohorts, while fully digital exams tend to force a didactic narrowing towards closed question formats. A practical middle ground keeps paper-based, problem-oriented tasks but records the assessment-relevant answers as single capital letters in a table that a machine can read. The open question is whether this reading can be made accurate and, above all, fair enough for unsupervised grading. Earlier automated approaches reached only about 88%–91% recognition – too low – and failed on the cases that matter most: answers placed outside the cell, crossed out, or written in cursive. We show that general-purpose vision-language foundation models (VLMs), which interpret the page rather than match pixel templates, close this gap. On a benchmark of 61 anonymised exams (3141 answer positions) the best model reaches 98.4% accuracy, well above the previous baseline. Crucially, we centre the evaluation on fairness: we distinguish false negatives (a correct answer marked wrong, which disadvantages the student) from false positives, and a lightweight prompt that supplies the reference solution as context lowers the false-negative rate to 0.58%. Under an exemplary grading scheme only three of the 61 exams would be graded worse, all caught by a student self-review step. Fully automated, fairness-aware exam grading at scale is therefore defensible; we release the anonymised benchmark to support reproducibility.

14.
arXiv (math.PR) 2026-06-16

Testing for a Hidden Geometry in Random Graphs

arXiv:2606.16715v1 Announce Type: cross Abstract: We study the problem of detecting a faint geometric signal hidden in an otherwise random graph. Formally, we consider a hypothesis testing problem in which, under the null, the observed graph is an Erdős–Rényi random graph $\mathcal{G}(n,q)$, while under the alternative a random geometric graph $\mathcal{G}(k,q,d)$ is planted on $k\le n$ vertices. The planted subgraph is generated from independent random points on the unit sphere $\mathbb{S}^{d-1}$, with edges determined by latent geometric proximity and calibrated to have edge density $q$. Our goal is to characterize the statistical and computational limits of detecting this hidden geometry. We derive sharp information-theoretic lower bounds that identify regimes where detection is impossible and provide algorithms that achieve these limits whenever detection is feasible. We further investigate the computational complexity of the problem and determine when efficient polynomial-time tests exist. The model exhibits an easy–hard–impossible phase transition: some regimes allow efficient detection, others permit detection only with computationally intractable procedures, and still others render detection impossible even with unlimited computational power. As evidence for the computational barrier, we prove that all low-degree polynomial algorithms fail throughout the conjecturally hard regime, demonstrating a sharp gap between statistical and computational feasibility.

15.
arXiv (CS.AI) 2026-06-15

A Deep Reinforcement Learning (DRL)-Based Transformer Method for Solving the Open Shop Scheduling Problem

arXiv:2606.13682v1 Announce Type: new Abstract: The open shop scheduling problem (OSSP) arises in many industrial and service settings but remains computationally challenging as the number of jobs and machines increases. While exact methods quickly become intractable, classical dispatching rules and metaheuristics may require substantial tuning to maintain solution quality at large scales. This study develops a Transformer-based scheduling policy for OSSP using an encoder-decoder architecture with multi-head attention. The model is trained on Taillard benchmark instances (4x4, 5x5, 7x7, and 10x10) using only the processing-time matrix as input and produces feasible schedules with makespans typically within 15-30% of best-known values. To evaluate scalability, the trained policy is applied without retraining to randomly generated instances from 40x40 to 100x100 and compared against classical dispatching heuristics, including SPT, LPT, MWKR, and EST. Across these large instances, the Transformer achieved average gaps of 12.89-15.12% relative to a standard lower bound. Compared with EST, the Transformer remained competitive, typically within a modest margin, while substantially outperforming SPT and LPT. These results indicate that a Transformer policy trained on small OSSP instances can generalize to substantially larger problems and provide a feature-light, learning-based alternative to classical dispatching rules.

16.
arXiv (CS.CL) 2026-06-16

GRACE: Step-Level Benchmark for Faithful Reasoning over Context

Many reasoning tasks require models to reason over input context, from document-grounded question answering to rule-based deduction. Chain-of-Thought (CoT) prompting produces traces that appear transparent, yet individual steps can silently deviate from the source evidence, even when the final answer is correct. Existing methods detect hallucinations at the response level but fail to identify where in the chain a failure occurs or what type it is. We introduce GRACE, the first human-annotated step-level faithfulness benchmark with a data-driven error taxonomy for context-grounded textual reasoning. GRACE covers CoT traces from 10 models across 4 source datasets, with each step annotated for faithfulness, error category, and natural language explanation. A data-driven taxonomy, discovered bottom-up via unsupervised clustering, organizes failures into two tracks: GRACE-Inference (deductive errors) and GRACE-Grounding (factual grounding errors), with four categories each. The evaluation set is human-annotated and challenging by design. Our experiments reveal substantial headroom for current models. In addition, integrating step-level faithfulness signals into reinforcement learning pipelines improves both downstream accuracy and reasoning reliability.

17.
arXiv (quant-ph) 2026-06-17

Robust Spin Splitting and Strain-Controlled Optical Response in Monolayer CrC2N4 for Valleytronic and Optoelectronic Applications

arXiv:2606.17329v1 Announce Type: cross Abstract: Monolayer CrC2N4 recently emerged as a promising two-dimensional semiconductor, yet its spin-orbit-coupled (SOC) physics and strain-tunable optical response remained largely unexplored. Here, we investigated the electronic, valley, charge-transfer, and optical properties of pristine and biaxially strained monolayer CrC2N4 using first-principles calculations. The monolayer exhibited a direct band gap at the K/K' valleys. SOC produced valley contrasting out-of-plane spin polarization, yielding a moderate valence band spin splitting of 51.9 meV and a small conduction band spin splitting of 1.7 meV. Orbital-resolved analysis showed that the edge states were mainly governed by Cr-d and N-p hybridization, while Bader analysis indicated polar-covalent bonding through charge transfer toward N atoms. Biaxial strain in the range of -4% to +4% tuned the band gap from 1.987 to 1.421 eV and drove an indirect-to-direct gap transition near -1% strain. Tensile strain enhanced the Berry curvature and red-shifted the optical response toward the visible-near-infrared region. These results suggested monolayer CrC2N4 as a promising platform for strain-engineered valleytronic and optoelectronic device applications.

18.
arXiv (CS.CL) 2026-06-16

A Mechanistic Understanding of Pronoun Fidelity in LLMs

Faithful and robust pronoun use is important for fair and coherent generations, yet large language models largely fail when multiple referents use different pronouns. To study the interplay of reasoning, repetition, and bias in this task, prior work relies exclusively on behavioural approaches, which may not reflect a model's internal workings. Therefore, we provide a mechanistic, model-internal perspective on pronoun fidelity, testing whether three mechanisms – group entity binding (G), recency bias (R), and stereotypical bias (S) – are causally implemented across several SOTA language models. Using Boundless Distributed Alignment Search, we find all three coexist as causal subspaces distributed across network depth. No single mechanism fully explains model behaviour, but a combination of the three consistently accounts for 91-99.5%. An attention head analysis further reveals two competing copying routes; group binding and stereotype share a localized concept-level route that retrieves a bound occupation-pronoun unit, while recency uses a distributed token-level route that repeats surface forms. In sum, pronoun fidelity arises from competition between simultaneously active causal subspaces.

19.
arXiv (CS.LG) 2026-06-12

Physics-Informed Neural Networks for Chemotherapy Pharmacokinetics: Benchmarking the Clinical Estimator and Exposing Parameter Identifiability

arXiv:2606.12658v1 Announce Type: new Abstract: Physics-Informed Neural Networks (PINNs) are an attractive tool for partial-observation problems in biology, where the governing dynamics are known but some compartments cannot be measured. Chemotherapy pharmacokinetics (PK) is a clean instance: drug concentration in plasma is routinely measured, but concentration in tissue – which determines tumour kill and off-target toxicity – is not. We benchmark a PINN against the standard clinical baseline (nonlinear least-squares on the analytical biexponential plasma solution, hereafter NLS) and a physics-agnostic neural baseline (a data-only MLP) on two PK problems. On the linear two-compartment problem, NLS is near-optimal; the PINN matches it to within a small constant factor while also producing the tissue curve in a single training pass, whereas the data-only MLP fails on tissue by roughly 10x. On a Michaelis-Menten extension (saturable elimination), the biexponential closed form no longer exists, so NLS is mis-specified and silently returns meaningless rate constants. The PINN instead exposes a deeper fact: the Michaelis-Menten two-compartment model is non-identifiable from plasma alone, and the PINN reports this honestly by converging to a basin with k12 -> 0. Adding two sparse tissue observations largely resolves identifiability: across five seeds the PINN recovers k21 to within 1% of truth and Vmax, Km to within one standard-deviation bar, while k12 moves in the correct direction (0.02 -> 0.82) but remains ~2 sigma below truth – a recovery the closed-form NLS estimator cannot attempt at all, because its biexponential ansatz describes only plasma. Our claim is not that PINNs beat NLS. It is that PINNs offer a uniform recipe that ties the textbook estimator on the textbook problem, exposes structural identifiability that the textbook estimator hides, and absorbs heterogeneous measurements within a single loss.

20.
arXiv (CS.AI) 2026-06-17

IUU+DB: Tracking Illegal, Unreported, and Unregulated Fishing, Seafood Fraud, and Labor Abuse through LLM-driven Information Extraction

arXiv:2606.18181v1 Announce Type: cross Abstract: Illegal, unreported, and unregulated fishing (IUU) traditionally refers to fishing activities that violate applicable laws or occur in areas that lack applicable laws. We propose the term IUU+ to capture a broader suite of fisheries sector environmental and associated supply chain trade-related crimes and behaviors. Although IUU+ activity is widely recognized as a serious threat to marine ecosystems, markets, and livelihoods, a quantitative understanding of these incidents, e.g., their frequency, geography, species, actors, and patterns in the type of illicit activity, remains difficult to obtain. We propose IUU+DB, a large language model driven system for building a global incident database of IUU+ activity. The system ingests heterogeneous documents, classifies whether they describe relevant incidents, extracts key data elements such as actors, locations, species, vessels, violations, and enforcement outcomes, and supports deduplication and trend analysis. Case studies and validation results show that IUU+DB can help organize fragmented evidence, surface geographic and behavioral hotspots, support fisheries-domain specific research in academia and non-government organizations, assist source and species risk assessments for industry, and provide support for policy implementation and targeted enforcement efforts to government agencies.

21.
arXiv (CS.LG) 2026-06-12

Geometry of Lightning Self-Attention: Identifiability and Dimension

arXiv:2408.17221v3 Announce Type: replace Abstract: We consider function spaces defined by self-attention networks without normalization, and theoretically analyze their geometry. Since these networks are polynomial, we rely on tools from algebraic geometry. In particular, we study the identifiability of deep attention by providing a description of the generic fibers of the parametrization for an arbitrary number of layers and, as a consequence, compute the dimension of the function space. Additionally, for a single-layer model, we characterize the singular and boundary points. Finally, we formulate a conjectural extension of our results to normalized self-attention networks, prove it for a single layer, and numerically verify it in the deep case.

22.
arXiv (CS.LG) 2026-06-11

A Judge-Aware Ranking Framework for Evaluating Large Language Models without Ground Truth

arXiv:2601.21817v3 Announce Type: replace-cross Abstract: Evaluating large language models (LLMs) on open-ended tasks without ground-truth labels is increasingly done via the LLM-as-a-judge paradigm. A critical but under-modeled issue is that judge LLMs differ substantially in reliability; treating all judges equally can yield biased leaderboards and misleading uncertainty estimates. More data can make evaluation more confidently wrong under misspecified aggregation. We propose a judge-aware ranking framework that extends the Bradley-Terry-Luce model by introducing judge-specific discrimination parameters, jointly estimating latent model quality and judge reliability from pairwise comparisons without reference labels. We establish identifiability up to natural normalizations and prove consistency and asymptotic normality of the maximum likelihood estimator, enabling confidence intervals for score differences and rank comparisons. Across multiple public benchmarks and a newly collected dataset, our method improves agreement with human preferences, achieves higher data efficiency than unweighted baselines, and produces calibrated uncertainty quantification for LLM rankings.

23.
arXiv (CS.CL) 2026-06-16

DYNA : Dynamic Episodic Memory Networks for Augmenting Large Language Models with Temporal Knowledge Graphs in Continuous Learning

Large Language Models (LLMs) struggle to incorporate new knowledge without forgetting or costly retraining. We propose DYNA, a lightweight framework that augments a frozen LLM with a temporal knowledge graph where events are nodes and temporal relations are directed, timestamped edges. The graph serves as an external, updatable memory. At query time, DYNA retrieves relevant nodes via random walks and centrality measures, then augments the LLM's response. Evaluated on three temporal recall tasks, DYNA reduces catastrophic forgetting by ~7% compared to fine-tuning and improves temporal ordering by ~5% over standard RAG. Higher graph clustering coefficients correlate with better retrieval, showing that graph structure matters. Contributions: (1) episodic memory as temporal KG, (2) retraining-free LLM augmentation, (3) graph properties as predictors of retrieval performance.

24.
arXiv (CS.AI) 2026-06-18

DRIFT: Refining Instruction Data via On-Policy Data Attribution

arXiv:2606.18307v1 Announce Type: cross Abstract: Optimizing the training data distribution for Supervised Fine-Tuning (SFT) dictates the capability of Large Language Models (LLMs). While existing data curation methods excel at accelerating training under constrained budgets, they are less suited to elevating the capability upper bound. The challenge here is no longer to identify a smaller subset that preserves performance, but to refine the data distribution toward instances most capable of improving the final model. To address this problem, we explore instance-level data attribution using Influence Functions (IF). We identify that standard IF formulations struggle in this setting due to two structural limitations: a proximity gap caused by off-policy validation targets, and a severe bias towards gradient norm. We propose DRIFT (Data Refinement via On-Policy Influence Functions for Supervised Fine-Tuning). Instead of relying on external reference data, DRIFT utilizes the model's on-policy rollouts as validation targets, which empirically minimizes the parameter proximity gap and better aligns with the local neighborhood assumption of IF. It further applies signed weighting based on trajectory correctness and debiases influence scores against the gradient hacking issue, allowing a small set of validation queries to act as reliable anchors for attributing the full dataset. Experiments on 7B-parameter instruction and reasoning models show that DRIFT consistently raises the performance ceiling on both, outperforming existing data curation baselines.

25.
medRxiv (Medicine) 2026-06-16

Adverse Childhood Experiences and Growth Outcomes in Childhood: A Longitudinal EHR-Based Study

Question Are adverse childhood experiences (ACEs) associated with altered growth trajectories in childhood? Findings In this cohort study of 412,549 children and adolescents, ACEs were associated with lower height throughout childhood, earlier pubertal timing, and shorter final stature. Height differences emerged approximately 2 years before ACE documentation and were greatest among those with earlier documentation. Meaning These findings suggest that early adversity affects physical growth in children and may serve as a measurable indicator of the biological consequences of early-life stress, especially in those with documentation of ACEs prior to the onset of typical pubertal growth. Importance Adverse childhood experiences (ACEs) are among the strongest risk factors for long-term mental and physical health complications, yet their impact on physical growth in childhood remains incompletely understood. Objective To determine the association of ACEs on childhood growth trajectories and growth dynamics. Design, Setting and Participants Retrospective cohort study using longitudinal electronic health record data. Data was collected from participants between February 1999 and August 2025. A large academic medical center biobank linked to deidentified electronic health records in the southeastern United States. A total of 412,549 individuals with at least 2 recorded height measurements between the ages of 2 and 20 were included in the primary analysis. Growth curve analyses were performed in a subset of 199,844 individuals with at least 3 height measurements spanning at least 2 years. Genetic analyses were performed in a subset of 10,114 individuals of primarily European ancestry. Exposure(s) Documented exposure to adverse childhood experiences before age 18 years identified through a natural language processing algorithm. Main Outcome(s) and Measure(s) Height-for-age z-scores across childhood, final attained height, and growth curve parameters estimated using SuperImposition by Translation and Rotation (SITAR) modeling. Results Among 412,549 participants, 18,502 (4.5%) had clinically documented ACEs during childhood. ACE documentation was associated with lower height-for-age z-scores throughout childhood and adolescence. Final attained height was significantly lower among ACE-documented individuals, with mean differences of -3.0 cm among males (174.0 cm vs 177.0 cm, p < 0.001) and -1.3 cm among females (161.8 cm vs 163.1 cm, p < 0.001). Height differences emerged approximately 2 years before clinical ACE documentation. Earlier age at first ACE documentation was associated with progressively shorter final attained height, with each year decrease in age at ACE documentation associated with a decrease in final height of -0.20 cm in females and -0.35 cm in males. Those with first ACE documented prior to pubertal age also showed the most pronounced growth dynamic differences, with males demonstrating a mean reduction in size of 5.25 cm (95% CI, -6.79 cm to -3.70 cm) and 1.26-year earlier pubertal timing (95% CI, -1.50 to -1.03 years), and females demonstrating a reduction in growth curve size of 3.62 cm (95% CI, -4.83 to -2.41 cm) and 1.14-year earlier pubertal timing (95% CI, -1.29 to -0.99 years). Conclusions and Relevance In this large clinical cohort, clinically documented ACEs were associated with time-dependent reductions in stature, earlier pubertal timing, and short final attained height. These findings suggest that early childhood adversity may have lasting effects on physical development and highlight growth trajectories as a potential marker of the biological consequences of early-life stress.