Academic Intelligence · Curated Daily

探索全球前沿学术脉络

AcademicHub 汇聚顶级期刊与预印本平台的实时文献。定制您的专属科研雷达,利用大语言模型自动生成交叉领域文献分析简报。

01.
arXiv (CS.CL) 2026-06-18

ForecastBench-Sim: A Simulated-World Forecasting Benchmark

Forecasting benchmarks for general-purpose AI systems usually inherit the constraints of the real world: outcomes resolve slowly, tail events are rare, and counterfactual questions are difficult to score. We introduce ForecastBench-Sim, a simulated-world forecasting benchmark built on game rollouts from Freeciv, a turn-based strategy game modelled on the Civilization series. Forecasters receive a fixed world report (a structured snapshot of the current game state) and answer questions about hidden future states; the benchmark then continues the simulation and scores forecasts. Because the world is simulated, the same setup can generate continuous or binary forecasting questions at arbitrary time horizons, paired intervention worlds for conditional or causal questions, and resolved examples of rare or disruptive outcomes. We describe the benchmark pipeline, question families, scoring protocol, and release artifacts, and report validation slices from model evaluations and an anonymized human pilot. ForecastBench-Sim is intended to complement real-world forecasting benchmarks by providing controlled, immediately resolvable tasks for studying probabilistic reasoning under dynamic world states.

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

In-Context Environments Induce Evaluation-Awareness in Language Models

Humans often become more self-aware under threat, yet can lose self-awareness when absorbed in a task; we hypothesize that language models exhibit environment-dependent evaluation awareness. This raises concerns that models could strategically underperform, or sandbag, to avoid triggering capability-limiting interventions such as unlearning or shutdown. Prior work demonstrates sandbagging under hand-crafted prompts, but this underestimates the true vulnerability ceiling. We introduce a black-box adversarial optimization framework treating the in-context prompt as an optimizable environment, and develop two approaches to characterize sandbagging: (1) measuring whether models expressing intent to underperform can actually execute it across different task structures, and (2) causally isolating whether underperformance is driven by genuine evaluation-aware reasoning or shallow prompt-following. Evaluating Claude-3.5-Haiku, GPT-4o-mini, and Llama-3.3-70B across four benchmarks (Arithmetic, GSM8K, MMLU, and HumanEval), optimized prompts induce up to 94 percentage point (pp) degradation on arithmetic (GPT-4o-mini: 97.8\%$\rightarrow$4.0\%), far exceeding hand-crafted baselines which produce near-zero behavioral change. Code generation exhibits model-dependent resistance: Claude degrades only 0.6pp, while Llama's accuracy drops to 0\%. The intent – execution gap reveals a monotonic resistance ordering: Arithmetic $

03.
arXiv (CS.AI) 2026-06-12

Grammar of the Wave: Towards Explainable Multivariate Time Series Event Detection via Neuro-Symbolic VLM Agents

arXiv:2603.11479v3 Announce Type: replace-cross Abstract: Time Series Event Detection (TSED) aims to localize semantically meaningful events in time series data, with critical applications in high-stakes domains. Unlike statistical anomalies, events are often defined by natural-language descriptions with internal temporal-logic structures across multiple physical channels. However, in real-world settings, dense event annotations are expensive to obtain, making purely supervised learning difficult. We introduce Language-guided TSED, a setting where a model is given textual event descriptions and must ground them to intervals in multivariate signals with little or no labeled data. To address this problem, we propose Event Logic Tree (ELT), a knowledge representation framework that converts linguistic descriptions into structured temporal logic over signal primitives. Building on ELT, we present SELA, a neuro-symbolic VLM agent framework that iteratively grounds primitives from signal visualizations and composes them under ELT constraints, producing both event intervals and faithful tree-structured explanations. We further release a real-world benchmark across energy and climate domains with expert knowledge and annotations. Experiments show that SELA improves over supervised fine-tuning and existing zero/few-shot time series reasoning baselines.

04.
arXiv (CS.LG) 2026-06-17

Instrumental and Proximal Causal Inference with Gaussian Processes

arXiv:2603.02159v2 Announce Type: replace-cross Abstract: Instrumental variable (IV) and proximal causal learning (Proxy) methods are central frameworks for causal inference in the presence of unobserved confounding. Despite substantial methodological advances, existing approaches rarely provide reliable epistemic uncertainty (EU) quantification. We address this gap through a Deconditional Gaussian Process (DGP) framework for uncertainty-aware causal learning. Our formulation recovers popular kernel estimators as the posterior mean, ensuring predictive precision, while the posterior variance yields principled and well-calibrated EU. Moreover, the probabilistic structure enables systematic model selection via marginal log-likelihood optimization. Empirical results demonstrate strong predictive performance alongside informative EU quantification, evaluated via empirical coverage frequencies and decision-aware accuracy rejection curves. Together, our approach provides a unified, practical solution for causal inference under unobserved confounding with reliable uncertainty.

05.
arXiv (CS.CV) 2026-06-24

Heterogeneous Knowledge Distillation via Geometry Decoupling and Momentum-Aware Gradient Regulation

Heterogeneous Knowledge Distillation (HKD) aims to transfer knowledge across varying architectures (e.g., from Transformer to CNN) but inherently suffers from severe training instability. We reveal that this instability stems from two highly coupled challenges: massive feature norm discrepancies that cause optimization drag, and severe gradient conflicts between the primary and distillation objectives arising from distinct inductive biases. To achieve stable distillation, we propose SPOFA, a framework built upon a novel Feature and Gradient Dual Stabilization mechanism. Specifically, at the feature level, we introduce a LayerNorm-based decoupling projector that explicitly decouples feature magnitude from direction, creating a bounded and stable space for semantic alignment. At the gradient level, we propose a momentum-driven Exponential Moving Average (MEMA) dynamic scaler. By establishing a robust historical baseline of the optimization trajectory, MEMA actively evaluates instantaneous gradient conflicts and adaptively penalizes harmful distillation signals, guaranteeing stable convergence. Importantly, SPOFA achieves this dual stabilization with an extremely lightweight parameter footprint. Extensive experiments on two mainstream benchmarks demonstrate that SPOFA achieves state-of-the-art accuracy, significantly outperforming computationally expensive methods while introducing only minimal computational overhead compared to standard baselines.

06.
medRxiv (Medicine) 2026-06-16

Supplementation with Arabinoxylan Dietary Fiber at Low Doses Produces Behavioral, Metabolic, and Gut Microbial Changes in Healthy, Overweight Adults: A Randomized Placebo-Controlled Trial

Background: Dietary fiber comprises a heterogeneous group of compounds with distinct physicochemical properties and biological effects. As such, functional outcomes observed for one fiber cannot be generalized to others. Some fermentable fibers, such as arabinoxylan, may exert biologically selective effects across multiple physiological domains, highlighting the need to evaluate individual ingredients for their domain-specific activity in controlled human studies. Methods: In this randomized, double-blind, parallel, 3-arm, placebo-controlled trial, healthy, overweight adults were assigned to consume one of two low doses of an arabinoxylan dietary fiber (3.5g or 5g) or placebo over the intervention period. Self-reported appetite sensations were assessed as the primary outcome using validated visual analogue scales. Secondary and exploratory endpoints included lipid parameters, gastrointestinal outcomes, mood-related measures, and gut microbiota composition and fermentation-derived metabolites. Analyses were conducted in the full analysis set and a high-compliance population to assess responses under sustained intake conditions, as per the intended dosing regimen. Results: The primary endpoint of appetite sensations did not differ between either arabinoxylan group and placebo. In contrast, evidence of microbial fermentation and selective microbiota engagement was observed. These responses occurred alongside consistent and favorable changes in lipid parameters under conditions of sustained intake, including reductions in low-density lipoprotein cholesterol and triglycerides. Additional outcomes, including gastrointestinal symptoms and mood, demonstrated domain-specific responses. Conclusion: This study demonstrates that supplementation with low doses of arabinoxylan dietary fiber elicit biologically selective, domain-specific effects across metabolic, microbial, gastrointestinal, and behavioral outcomes, particularly under conditions of sustained intake. These responses occurred independently of changes in appetite sensation, indicating that functional effects were not mediated through appetite-related pathways. Collectively, the findings highlight the ingredient's biological versatility and contextual responsiveness across physiological systems, and suggest its prebiotic potential through alignment with ISAPP's definition of a prebiotic, supporting further investigation of specific mechanistic pathways. Clinical trial registration: https://clinicaltrials.gov/study/NCT06884449, identifier: NCT06884449

07.
arXiv (quant-ph) 2026-06-16

Morphology-resolved scrambling in a chaotic quantum billiard

arXiv:2606.16865v1 Announce Type: new Abstract: Chaotic quantum systems can retain spatial memory through scarred eigenstates, but whether these static structures control scrambling remains unclear. This work establishes a morphology-resolved connection between scarred eigenstates and eigenstate-resolved OTOCs in a peanut-shaped quantum billiard. Scalar localisation diagnostics, including differential entropy and continuum participation ratios, detect anomalous concentration but discard spatial architecture. A scale-normalised density overlap, in contrast, directly compares probability density profiles, revealing families of orthogonal eigenstates with nearly identical spatial morphology. Comparing the complete OTOC time traces of these orthogonal eigenstates reveals that morphological recurrence has dynamical content: moderate density overlap yields no universal prediction, whereas strongly recurring morphologies exhibit nearly identical OTOC growth and saturation. Thus, scarred structures act as spatial templates for operator growth, not merely static violations of ergodicity. This morphology-resolved framework turns eigenstate shape into a quantitative predictor of scrambling and provides a scale-controlled diagnostic of weak ergodicity breaking in quantum chaos.

08.
arXiv (CS.CL) 2026-06-25

Digital Twin-Driven Adaptive Sim-to-Real Alignment via Reinforcement Learning for Vibration-Based Bearing Health Monitoring Under Data Scarcity

Vibration-based health monitoring of rotating machinery requires reliable fault diagnosis under operational data constraints, yet condition assessment remains challenged by structural scarcity of fault events and heterogeneous sim-to-real gaps in digital twin-generated signals. Each fault type generates impulses with distinct periodicity, amplitude modulation, and spectral character, making feature-space discrepancies fundamentally heterogeneous across fault classes. Existing domain adaptation methods apply a class-agnostic global transformation that cannot close all fault-specific gaps without distorting inter-class separability, while uniform source-target mixing introduces distributional noise into the data-abundant Normal class. These limitations stem from treating a sequential, state-dependent alignment problem as a one-shot optimization. Each corrective transformation simultaneously reshapes all class distributions, creating state dependencies that static gradient descent cannot resolve. We formulate feature alignment as a continuous-action Markov decision process solved via Proximal Policy Optimization, where the learned policy issues fault-type-specific affine corrections responsive to the current feature-space configuration, with a dual-objective reward balancing gap minimization against separability preservation. An asymmetry-aware strategy reserves real data for the Normal class while augmenting fault classes with policy-aligned simulated samples. Validation across XJTU-SY, CWRU, and a self-built slewing bearing testbed confirms the dominant gain from reinforcement learning-driven alignment, and cross-equipment linear probing achieves 92.8% without encoder retraining, demonstrating transferable monitoring capability.

09.
arXiv (CS.CV) 2026-06-17

Not Truly Multilingual: Script Consistency as a Missing Dimension in VLM Evaluation

Current multilingual evaluations for Vision-Language Models (VLMs) assume a one-to-one mapping between language and orthography, overlooking billions of users of multi-script languages. We introduce PuMVR (Punjabi Multimodal Visual Reasoning), a benchmark of 1,000 strictly parallel image-text instances across Punjabi's three active scripts: Gurmukhi, Shahmukhi, and Roman. Evaluating 10 state-of-the-art VLMs, we expose a substantial and systematic Script Gap. Models frequently solve visual tasks in one script while failing identical tasks in another, with accuracy deltas reaching 16%. Crucially, visual input boosts absolute performance uniformly yet does not close the orthographic gap. Furthermore, cross-script in-context transfer is highly brittle, exposing script-locked knowledge representation. Supported by McNemar tests across all script pairs, our findings demonstrate that current "multilingual" VLMs are not truly multi-script. We propose the Script Consistency Rate (SCR), which falls as low as 24.8% on our benchmark, as a mandatory metric for script-agnostic evaluation to ensure equitable AI access. Data and code are available at: https://github.com/prabhjotschugh/Not-Truly-Multilingual-PuMVR.

10.
arXiv (CS.CL) 2026-06-11

Measuring language complexity from hierarchical reuse of recurring patterns

We introduce the ladderpath index as a measure of language complexity grounded in algorithmic information theory. It counts the minimum steps needed to reconstruct a sequence through hierarchical reuse of repeated substructures, capturing an exactly computable but constrained form of algorithmic compressibility related to, but distinct from, Kolmogorov complexity. We apply the ladderpath approach to 21 parallel corpora from the Parallel Universal Dependencies dataset. The ladderpath index is approximately invariant across the languages, and varies much less than the corpus length. This is more pronounced when all corpora are mapped to a unified binary representation, providing evidence for the equi-complexity hypothesis from a representation-independent perspective. We also observe trade-offs between character inventory size and corpus length, and between vocabulary-level and corpus-level reconstruction complexity, supporting the trade-off hypothesis that total complexity is conserved and redistributed across linguistic levels. The reusable substructures identified by the ladderpath approach, without any linguistic input, overlap with words and morphological components attested in the natural vocabulary. The hierarchical reuse captured by the ladderpath approach parallels the chunking mechanisms proposed in cognitive science, where the human cognitive system compresses linguistic input into nested, reusable units under shared memory and processing constraints. This connection between cognitive chunking and the ladderpath approach provides a new interpretation for the equi-complexity and trade-off hypotheses, grounding both in the shared cognitive architecture that underlies language processing across human languages.

11.
arXiv (CS.LG) 2026-06-25

Knowledge Cascade: Reverse Knowledge Distillation on Nonparametric Multivariate Functional Estimation

arXiv:2606.25927v1 Announce Type: cross Abstract: As machine learning models and datasets continue to grow, developing complex models has become increasingly computationally demanding. Knowledge distillation reduces deployment cost by compressing a large, well-trained teacher model into a compact student model, but it does not address settings where constructing the teacher itself is the bottleneck. Motivated by this challenge, we introduce Knowledge Cascade (KCas), a reverse knowledge distillation framework that uses information from a small, inexpensive student model to guide the development of a more complex teacher model. Although this direction is counterintuitive because the teacher typically has greater representational capacity, we show that student-to-teacher transfer can be principled when supported by statistical scaling relationships. We first develop KCas for nonparametric multivariate functional estimation in reproducing kernel Hilbert spaces via smoothing splines, where selecting multiple smoothing parameters is a major computational bottleneck. KCas transfers student-selected smoothing parameters to the full-sample regime through asymptotic scaling laws, substantially reducing computational cost for high-dimensional and large-scale datasets while retaining theoretical guarantees. Beyond smoothing splines, we illustrate the same principle through kernel density estimation and deep learning hyperparameter transfer. Simulations and real-data experiments show that KCas achieves substantial computational savings while maintaining strong statistical performance, and can sometimes outperform the corresponding full-sample procedure.

12.
arXiv (quant-ph) 2026-06-17

$\mathcal{PT}$-Symmetric Spin–Boson Model with a Continuous Bosonic Spectrum: Exceptional Points and Dynamics

arXiv:2512.20277v2 Announce Type: replace Abstract: This work studies a $\mathcal{PT}$-symmetric non-Hermitian spin–boson model, consisting of a non-Hermitian two-level system coupled to a continuous bosonic bath. The static properties of the system are analyzed through a projection method derived from the displacement operator. We find that only a single exceptional point (EP) emerges, in contrast to non-Hermitian spin–boson models with finite modes, which typically exhibit multiple EPs. Notably, only a single real eigenvalue is found before the EP, which differs markedly from typical non-Hermitian systems where a pair of real eigenvalues precedes the EP. The time evolution of observables is further investigated via the Dirac–Frenkel time-dependent variational principle. Compared to its Hermitian counterpart, the non-Hermitian model exhibits distinct dynamical signatures, most notably the emergence of oscillations with periodic amplified amplitude. In the $\mathcal{PT}$-unbroken phase, the system exhibits sustained oscillatory dynamics with suppressed decoherence, whereas in the $\mathcal{PT}$-broken phase, additional dissipative channels accelerate decoherence and drive rapid convergence toward a stable steady state. These results shed light on how $\mathcal{PT}$ symmetry protects coherent light–matter interactions in non-Hermitian quantum systems.

13.
arXiv (quant-ph) 2026-06-11

Non-Hermitian Delocalization Realizes Random Dirac Criticality in One Dimension

arXiv:2606.12089v1 Announce Type: cross Abstract: Non-Hermitian systems can evade Anderson localization and exhibit delocalized states even in one dimension. Here, we show that such non-Hermitian delocalized states under periodic boundary conditions (PBC) are intrinsically critical, realizing the universality class of one-dimensional random Dirac fermions. By linking spectral winding to topological Anderson transitions via Hermitization, we demonstrate that the delocalized PBC states exhibit a Dirac-type criticality with universal algebraic correlations. In contrast to Hermitian systems, where this criticality occurs only at fine-tuned transition points, it emerges generically in non-Hermitian systems as a consequence of spectral topology. These results identify a universal mechanism by which non-Hermiticity promotes criticality, providing a unified description of non-Hermitian delocalization in one dimension.

14.
arXiv (CS.LG) 2026-06-25

Symmetric Linear Dynamical Systems are Learnable from Few Observations

arXiv:2512.05337v2 Announce Type: replace-cross Abstract: We consider the problem of learning the parameters of a $N$-dimensional stochastic linear dynamics under both full and partial observations from a single trajectory of time $T$. We introduce and analyze a new estimator that achieves a small maximum element-wise error on the recovery of symmetric dynamic matrices using only $T=\mathcal{O}(\log N)$ observations, irrespective of whether the matrix is sparse or dense. This estimator is based on the method of moments and does not rely on problem-specific regularization. This is especially important for applications such as structure discovery.

15.
arXiv (CS.AI) 2026-06-24

ReM-MoA: Reasoning Memory Sustains Mixture-of-Agents Scaling

arXiv:2606.24437v1 Announce Type: new Abstract: Mixture-of-Agents (MoA) architectures improve inference-time scaling by organizing multiple LLM agents into layered reasoning pipelines. However, existing MoA variants fail to sustain gains as depth increases, exhibiting degradation, early plateauing, or saturation. We propose ReM-MoA, a memory-augmented MoA framework that sustains scaling through two mechanisms: (1) a Ranked Reasoning Memory that persistently stores and ranks reasoning traces from all layers using a comparative Reviewer Agent, and (2) a Curated Diversified Memory Routing scheme that exposes different agents to distinct combinations of successful and failed traces, preserving exploration diversity while propagating high-quality reasoning. We further introduce an optional multi-domain Reviewer distillation pipeline that improves ranking quality through frontier-model supervision. Across five reasoning benchmarks spanning math, formal logic, code, knowledge, and commonsense, ReM-MoA consistently outperforms prior MoA variants across both depth and width scaling, and its advantage widens with depth, establishing structured cross-layer reasoning memory as a key missing mechanism for scalable multi-agent inference.

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

Collapsibility in Multiparametric Models of Random Simplicial Complexes

作者:

arXiv:2606.15276v1 Announce Type: cross Abstract: We study collapsibility in the multiparametric models of random simplicial complexes, namely the lower and upper models. In the upper model, we improve upon a result of Farber and Nowik, and assert that the homology is a.a.s concentrated in a single dimension by proving that the complex collapses to that \di. In the lower model, we prove that the complex a.a.s collapses to the \di\ with maximal non-trivial cohomology. We then compare this threshold to the ones derived previously for the special cases of the clique complex (by Kahle) and the Linial-Meshulam model.

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

Probes of chaos over the Clifford group and approach to Haar values

arXiv:2603.29695v3 Announce Type: replace Abstract: Chaotic behavior of quantum systems can be characterized by the adherence of the expectation values of given probes to moments of the Haar distribution. In this work, we analyze the behavior of several probes of chaos using a technique known as Isospectral Twirling [1]. This consists in fixing the spectrum of the Hamiltonian and picking its eigenvectors at random. Here, we study the transition from stabilizer bases to random bases according to the Haar measure by T-doped random quantum circuits. We then compute the average value of the probes over ensembles of random spectra from Random Matrix Theory, the Gaussian Diagonal Ensemble and the Gaussian Unitary Ensemble, associated with non-chaotic and chaotic behavior respectively. We also study the behavior of such probes over the Toric Code Hamiltonian.

18.
arXiv (CS.CV) 2026-06-18

Revealing Hidden Vulnerabilities in Autoencoders through Gradient Signal Restoration

Adversarial robustness of deep autoencoders (AEs) has received less attention than that of discriminative models, although their compressed latent representations induce ill-conditioned mappings that can amplify small input perturbations and destabilize reconstructions. Existing white-box attacks for AEs, which optimize norm-bounded adversarial perturbations to maximize reconstruction damage, often converge to suboptimal perturbations, thereby potentially overstating AE robustness. We show that this limitation is linked to vanishing adversarial loss gradients during backpropagation through ill-conditioned layers, associated with near-zero singular values in their intermediate weight matrices. To address this, we propose GRILL (Gradient Signal Restoration in Ill-Conditioned Layers), a framework designed to mitigate gradient degradation and improve the reliability of adversarial robustness evaluation in encoder-decoder architectures. GRILL is designed to mitigate adversarial gradient degradation during optimization, enabling attacks to better approximate high-distortion perturbations under fixed norm constraints. Through extensive experiments across multiple AE architectures, under both sample-specific and universal attacks, as well as standard and adaptive attack settings, we show that GRILL significantly increases attack effectiveness, thereby exposing vulnerabilities hidden by existing attack limitations. Beyond AEs, we provide preliminary evidence that modern multimodal encoder-decoder architectures exhibit similar vulnerabilities.

19.
arXiv (CS.CV) 2026-06-17

Structured Adversarial Camouflage via Voronoi Diagrams

Pixel-wise adversarial patches are computationally heavy and often visually detectable, limiting utility in security-critical systems. We present adversarial Voronoi camouflage that optimizes only seed-point locations under fixed, printable palettes using a soft assignment, producing structured, splinter camouflage-like patterns without additional regularization. Evaluated on person detection with COCO-style AP@[.5:.95], naive placement (Inria -> COCO) performs comparably bad, while garment-level application via segmentation mask (3DPeople) results in a significant AP drop. The attack transfers to out-of-domain backgrounds and across detector families (YOLOv9/10/11/12), indicating robustness in black-box settings. Repainting with different palettes largely nullifies the effect, and single-color tweaks show limited tolerance (

20.
arXiv (CS.CL) 2026-06-18

SenFlow: Inter-Sentence Flow Modeling for AI-Generated Text Detection in Hybrid Documents

Sentence-level AI-generated text detection (S-AGTD) for hybrid documents, where humans and LLMs co-author one text, faces two gaps: existing methods classify each sentence in isolation, discarding inter-sentence dependencies, and existing benchmarks omit the newest generation of generators. We construct MOSAIC, a benchmark of 16,000 hybrid documents over PubMed and XSum, generated by DeepSeek-V3.2 and Kimi K2 under stringent quality controls including a perplexity-consistency filter absent from prior benchmarks. We recast S-AGTD as structured prediction over the document sentence sequence and instantiate it as SenFlow, integrating graph-based inter-sentence propagation with linear-chain CRF decoding in a single document-level pass over a sentence graph. SenFlow reaches state-of-the-art performance on MOSAIC, with a +4.15 pp average Macro-F1 margin on cross-domain transfer, the hardest of three protocols of increasing difficulty. We further find that even after the perplexity filter equalizes overt cues, AI insertions retain a generator-dependent sentence-length gap that sentence-level detectors still exploit. Code and data: https://github.com/luojingkun22/SenFlow

21.
Nature (Science) 2026-06-25

How long-term dietary cholesterol can slow down its own clearance by liver cells

作者: 未知作者

Cholesterol carried by low-density lipoprotein (LDL) drives heart disease and is cleared by liver cells expressing the receptor LDLR. A cell-signalling mechanism has been discovered through which high cholesterol promotes the enzyme-mediated degradation of LDLR. Blocking this enzyme restores LDLR levels in liver cells, suggesting a new strategy for treating high cholesterol. Chronic exposure to cholesterol can move the receptor protein LDLR into cells for degradation.

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

Architecture-Aware Reinforcement Learning Makes Sliding-Window Attention Competitive in Math Reasoning

arXiv:2606.11634v1 Announce Type: new Abstract: The rapid progress of reasoning and agentic large language models (LLMs) has increased the demand for long-context inference, but self-attention (SA) scales quadratically with context length. To address this, we study SWARR (Sliding-Window Attention with Reinforced Adaptation for Math Reasoning), a practical recipe for adapting SWA models to mathematical reasoning. SWARR has two stages: (1) efficient conversion from a pretrained SA model to SWA with supervised fine-tuning (SFT), which avoids pretraining a new base model, and (2) policy adaptation with reinforcement learning (RL). We find that SWA still underperforms SA after SFT, and we hypothesize that this gap is caused in part by a data-architecture mismatch: most SFT data are prepared for SA models and may contain long-range dependencies that are difficult for SWA to model. Because on-policy RL optimizes self-generated trajectories under the SWA constraint, it can adapt trajectories to better match SWA. Experiments on mathematical reasoning benchmarks show that this recipe substantially narrows the gap between SWA and SA, recovering much of the accuracy lost during SWA conversion while preserving the efficiency benefits of linear-complexity attention. Our central contribution is the empirical finding that RL changes the conclusion one would draw from conversion and SFT alone about SWA's viability for math reasoning.

23.
arXiv (CS.AI) 2026-06-19

LoRDO: Distributed Low-Rank Optimization with Infrequent Communication

arXiv:2602.04396v2 Announce Type: replace-cross Abstract: Distributed training of foundation models via $\texttt{DDP}$ is limited by interconnect bandwidth. While infrequent communication strategies reduce synchronization frequency, they remain bottlenecked by the memory and communication requirements of optimizer states. Low-rank optimizers can alleviate these constraints; however, in the local-update regime, workers lack access to the full-batch gradients required to compute low-rank projections, which degrades performance. We propose $\texttt{LoRDO}$, a principled framework unifying low-rank optimization with infrequent synchronization. We first demonstrate that, while global projections based on pseudo-gradients are theoretically superior, they permanently restrict the optimization trajectory to a low-rank subspace. To restore subspace exploration, we introduce a full-rank quasi-hyperbolic update. $\texttt{LoRDO}$ achieves near-parity with low-rank $\texttt{DDP}$ in language modeling and downstream tasks at model scales of $125$M–$720$M, while reducing communication by $\approx 10 \times$. Finally, we show that $\texttt{LoRDO}$ improves performance even more in very low-memory settings with small rank/batch size.

24.
arXiv (CS.LG) 2026-06-19

Online Dynamic Batching with Formal Guarantees for LLM Training

arXiv:2606.19989v1 Announce Type: cross Abstract: Modern LLM training breaks a core assumption behind offline batch samplers: the true training cost of a sample is only observable after preprocessing, augmentation, templating, tokenization, and multimodal visual-token expansion. Unless one pays for a preprocessing- and augmentation-dependent length cache, batch construction is therefore blind to the quantity that determines padding, memory use, and GPU saturation. We introduce Online Dynamic Batching (ODB), a DataLoader-side drop-in system that moves batch formation to this point of accurate observability while preserving DDP step alignment. We formalize this synchronization requirement as the Distributed Group Alignment Problem and prove deadlock-free bounded termination with default join-mode identity coverage and opt-in non-join sample-quota closure. ODB requires no model, optimizer, or attention-kernel changes and is released as online-dynamic-batching with lightweight trainer adapters. Across public 2B/8B Qwen3-VL runs on UltraChat/LLaVA/ShareGPT4o, ODB improves literal emitted-sample throughput vs. fixed-batch Standard by 1.58-2.51x on single-node Full FT/LoRA and 1.71-3.78x on two-node Full FT, with Standard-comparable quality; production MM-Mix reaches 4.43x. Against GMT/BMT offline token-budget oracles, ODB is within 15% on UltraChat/LLaVA and faster on high-CV ShareGPT4o: 2.24-2.39x single-node Full FT/LoRA and 3.06-3.69x two-node Full FT. Together, ODB occupies the online/drop-in regime for high-heterogeneity LLM fine-tuning: large throughput gains at Standard-comparable quality, formal DGAP guarantees, and no length-cache precompute or kernel rewrites.

25.
arXiv (CS.CL) 2026-06-11

Beyond Compaction: Structured Context Eviction for Long-Horizon Agents

We present Context Window Lifecycle (CWL), a context-management scheme that gives long-horizon LLM agents an effectively unbounded working horizon. As a session accumulates history, CWL keeps the context within budget through graduated, semantically-aware eviction: the agent annotates its trajectory as typed, dependency-linked episodes as work proceeds, and a deterministic, LLM-free policy evicts content in priority order within that structure when a token budget is exceeded. CWL preserves user turns and the exploratory context the agent is actively reasoning over, while aggressively shedding action episodes whose effects are already persisted in the environment, keeping active context near a stable ceiling that also avoids the performance degradation associated with very large prompts. Compared to summarization-based compaction, CWL avoids four well-known limitations: unpredictable lossiness, destruction of causal structure, blocking model cost, and compression-induced hallucination. Compared to recency truncation, CWL is semantically aware: it drops the oldest-and-most-recoverable content according to the dependency graph rather than oldest-in-time regardless of relevance. We describe the annotation protocol, the episode graph, the eviction policy, and the token-accounting loop, and evaluate CWL on long-horizon agentic benchmarks: a single agent session completing 89 sequential tasks across 80 million tokens with no measurable degradation in task accuracy relative to per-task isolated sessions