Papers

Firms making inventory decisions have access to operational data, optimization tools, and large language models (LLMs). Typically, data characterize the operating environment, optimization selects parameters within a prespecified inventory policy class, and LLMs support coding and decision analysis. We develop an integrated framework that combines these resources to automate inventory policy design. Given demand data, the framework iteratively uses an LLM to generate parameterized policy classes and an external solver to optimize its parameters within each class. Across 30 lost-sales inventory instances, the mean cost reduction relative to optimized base-stock benchmarks increases from 17.5% after one generation to 30.0% after ten generations. Parameter optimization is central to this performance: an LLM-only variant performs substantially worse, whereas optimization-guided feedback improves policy quality, accelerates search, and directs the LLM toward better policy classes rather than merely better parameter values within a fixed class. The strongest discovered policies are also interpretable: they combine recognizable inventory-control motifs, including capped orders, discounted or weighted pipeline inventory, and threshold-based replenishment logic. The search thereby produces new policy-class functional forms that, to our knowledge, have not previously been studied in the lost-sales inventory literature. These functional forms are not specified ex ante but emerge from the search process. Moreover, after their parameters are re-optimized, three discovered policy classes achieve average cost reductions of 21.75% to 22.60% across 10,064 new inventory instances. Overall, the results show that data-driven parameter optimization can guide LLM-based search over a broad space of inventory policy classes and identify high-performing, interpretable, and transferable decision rules.

The brain uses discrete spikes for dynamic computation, yet, how neural microcircuits (NMCs) solve temporal credit assignment using local spike timing remains a fundamental open question. Dominant spiking neural network (SNN) approaches circumvent this by approximating backpropagation through surrogate gradients, decoupling learning from biological spike timing. Here, we reformulate temporal credit assignment as a state separation problem: extracting task-required components induced by historical perturbations directly from the current neural state. This enables an online feedback learning framework for NMCs through a gradient tunneling (GT) algorithm and the lead-lag expansion technique that derives credit assignment from local synaptic spike timing, while remaining compatible with ANN-SNN hybrid architectures. Experimentally, GT-trained NMCs excel at long-timescale evidence integration and noise-robust memory retention, and perform comparably to leading SNN online learning methods on real-world benchmarks with far fewer parameters. The proposed framework addresses the two-decade-old NMC feedback learning problem and suggests a computationally plausible explanation for the brain's learning mechanisms.

The learning-rate schedule is a consequential choice in training deep networks, yet the policies in common use are heuristic, and published comparisons are hard to read, because architecture, dataset, and budget tend to vary alongside the schedule. We study BrachistoneLR, a schedule built by mapping the vertical coordinate of the brachistochrone, the curve of fastest descent under gravity, onto the range between a peak and a floor rate. Expanding the definition shows it to be cosine annealing with the half-period set to E - 1 instead of E, the configuration a standard implementation gives when its period argument is one less than the number of epochs. The rate therefore reaches its floor at the last epoch trained rather than one epoch later, and we show this difference decays as E^-2, making it a short-horizon effect. We then benchmark six schedules over 72 runs on three image classification datasets (MNIST, Fashion-MNIST, CIFAR-10) and four architecture families (fully connected, convolutional, recurrent, residual), fixing the optimizer, data pipeline, and evaluation protocol so that only the schedule varies. Schedules that fall smoothly from peak to floor beat the constant rate and calendar-based decay by margins that grow with task difficulty, reaching 2.5 points of dataset mean on CIFAR-10. Within that leading group, BrachistoneLR, cosine annealing, and warmup-cosine lie within 0.06 accuracy points and 0.17 of a mean rank, which one seed per configuration cannot separate. BrachistoneLR is best on both residual networks and has the highest CIFAR-10 mean, and it sets no milestones, decay factor, warmup length, or restart period. We conclude that the shape of a schedule matters more than its parameterization, that the choice of whether to use a smooth schedule matters more than the choice among them, and that the terminal-rate distinction is worth attention only over short horizons.

Updating a language model's knowledge through fine-tuning is essential for keeping its outputs current, yet can also induce factual forgetting and new hallucinations. Prior work shows that long-tail knowledge is harder to acquire and newly memorized long-tail facts are difficult to retain during later fine-tuning. We study a complementary question: among facts that a model has encoded correctly, which are most vulnerable to collateral corruption during other updates? To investigate this question under a realistic factual distribution, we construct a large-scale graph FACTPROP of verified Wikipedia facts by linking triples that share head or tail entities, thereby preserving connections among factual knowledge. We fine-tune models on factual statements and measure correct-to-incorrect facts after each update. Our results reveal a pattern distinct from prior findings on long-tail vulnerability during acquisition and retention: among facts that models already answer correctly, those associated with highly connected entities are more likely to be corrupted by neighboring updates, and updates to such facts propagate errors more broadly. Structural popularity therefore predicts both vulnerability and downstream damage. Inspired by this finding, we propose Popularity-based Anchoring (PopAnchor), a lightweight rehearsal strategy that preserves a small set of popular facts and reduces forgetting.

Short-recurrence Krylov methods that are equivalent in exact arithmetic often diverge in floating-point arithmetic. To illustrate this, we provide an exactly representable two-cycle for steepest descent with a recursively updated residual. A complementary convergence theorem gives a sufficient condition under which the stored residual decreases geometrically. We then show that a second positive definite family yields different outcomes for a Hestenes–Stiefel Conjugate Gradient (CG) implementation and a direct Lanczos–Galerkin approach. CG finds the exact solution after four updates, whereas the two-step projected system becomes inconsistent. Under stated perturbation and transfer hypotheses, a common spectral enclosure gives comparable convergence bounds. For a dense, left-to-right evaluation order, we prove a sufficient precision bound for a prescribed backward error within $n$ updates for an $n\times n$ matrix, together with a computable stopping test and specified exponent-range assumptions. We use polynomial estimates and numerical experiments to examine the trade-off between working precision and the number of CG iterations needed to achieve a prescribed backward error. We also compare results on inexact matrix-vector products and preconditioning with the frameworks of Paige and Greenbaum. For block Lanczos, we analyze Householder orthogonalization and singular-value truncation as the block size changes. Under componentwise and normwise error bounds, the computed coefficients satisfy a controlled local recurrence and an exact block Lanczos relation for a nearby symmetric problem in a larger space. A block form of Paige's identity bounds the overlap with a Ritz vector along its residual coordinate direction. An inter-block recurrence describes the evolution of overlap, and a Gram-matrix argument gives a count of additional nearby Ritz values.

Large Language Models (LLMs) are increasingly deployed in settings where rare but severe harmful generations can have significant consequences. Existing Risk-Averse RLHF addresses this issue by optimizing Conditional Value-at-Risk (CVaR), but it trains policies for fixed risk levels and therefore cannot adjust the desired degree of risk aversion at inference time. In this paper, we propose risk-conditioned RLHF, a framework that trains a single policy that provides a continuous risk-control interface, enabling users to select different degrees of risk aversion without retraining or deploying multiple risk-specific models. Experiments across multiple benchmarks demonstrate that a single risk-conditioned policy can adapt to different risk levels at inference time, enabling more flexible and risk-aware LLM deployment.

With revocable delegation, two histories can yield identical current permissions yet require opposite decisions for the same query after the same revocation. We introduce ResidualAuth, a theory-grounded framework characterizing the authorization state agent systems must preserve and evaluating its maintenance and use. Its formal core, residual authorization state, equates histories exactly when every future sequence of grants, revocations, and uses is valid after both or neither. We prove that exponentially many distinct residual states can nevertheless agree on who can reach whom through delegation paths. The analysis also yields exact or tight memory bounds as delegation redundancy varies and an average decision-error lower bound under an explicit bound on retained information. ResidualAuth evaluates information access, online state maintenance, and information use through a restricted executable benchmark and separate diagnostics. The benchmark pairs episodes differing in authorization-relevant history and requiring opposite decisions; pair accuracy requires both answers to be correct. To test use of supplied decisions, four open-weight models received trusted current-query Allow/Deny decisions alongside deterministic 256-token event extracts, achieving 15-16/16 correct pairs versus 0-2/16 with extracts alone. Memory diagnostics identified invalid reconstructions; models also answered incorrectly from valid memories that passed fixed future authorization tests. Together, these results distinguish what future authorization requires a system to retain from whether agents can access relevant information, maintain state across updates, and use available information to make correct decisions.

The remarkable generalization of modern over-parameterized neural networks defies classical statistical learning bounds. Here we establish that optimization trajectories in billion-parameter models do not sample the ambient dimension $\mathbb{R}^N$ homogeneously; instead, they are dynamically constrained to a low-dimensional execution submanifold $\mathcal{M}_{\text{exec}}$ (with empirically observed effective dimension $d_{\text{eff}} \le 12$). By equipping the parameter space with a regularized Hessian metric $g_{ij} = \nabla_i \nabla_j L + \epsilon \delta_{ij}$ and analyzing stochastic Langevin diffusion on loss sublevel sets, we derive a rigorous Topological Generalization Bound grounded in projected subspace persistent homology. Furthermore, by linking Morse index transitions to Betti number annihilation ($\beta_1 \to 0$), we formulate an exact mathematical mechanism for exponential escape from strict saddles under anisotropic noise. We validate our framework across 80+ architectures (ResNets, Vision Transformers, and 7B LLMs). To bridge theoretical topology with computational feasibility, we resolve the curse of dimensionality by computing projected persistent homology proxies via randomized Lanczos iterations restricted strictly within active Hessian subspaces. Building upon this mathematical foundation, we introduce Topological-Adaptive Optimization (Top-SGD and Top-AdamW)—a family of wrapper algorithms that dynamically modulate learning rates using active curvature signatures. Extensive benchmarks demonstrate a 14.1% reduction in total wall-clock training time on ImageNet with all computational and memory overheads fully accounted for. Code is publicly available at https://github.com/imslab-research/Topological-SGD.

Automated research pipelines increasingly rely on fixed LLM judges to rank, filter, and revise proposals, yet these judges are routinely replaced by newer versions. We study whether such replacements preserve the underlying measurement standard. Using a preregistered, frozen-weight design, we evaluate three adjacent same-family transitions across Qwen, Gemma, and Mistral on frozen ML research proposals under randomized information conditions. The common assumption that newer judges are simply harsher fails: two transitions shift toward greater leniency, while the third shifts in the opposite direction on anchors but attenuates off-anchor. More importantly, version changes can alter not only score location but also criterion geometry and resolution; one transition collapses clean-flawed discrimination, placing 93% of pool scores on a single value. Supplying identical facts does not close the version gap, placebo-controlled interventions show that observed packet effects are largely format-driven rather than content-driven, and global recalibration repairs the mean while explaining little object-level dispersion. We organize these failures with a measurement model in which a judge replacement can change its origin, criterion geometry, or resolution. A small stratified anchor audit of about 200 calls identifies the failure class before deployment, and frozen thresholds correctly diagnose two post-registration API judge replacements. We release six preregistration locks and the full audit trail of 14882 judge calls.

We introduce GardenBench, a benchmark with 11 test suites that evaluate the capabilities of Large Language Models (LLMs) across a range of tasks. By systematically prompting LLMs with various lightweight queries, we document day-to-day changes in model performance, including run costs and times, that other benchmarks miss. This allows us to compare the accuracy and features of responses across model providers and versions. We account for daily variations in model accuracy, capturing both model progress and degradation, for instance, when a model is changed post-deployment or when a new model is released. We base our evaluation on general-use tasks and make our results publicly accessible via a live dashboard, diversifying existing approaches to LLM testing. Our main contribution is an innovative, lightweight, and inexpensive benchmark, designed to be consumer-facing and for everyday tasks.

The advancement of natural language processing (NLP) has expanded AI-based text classification in the legal domain. However, accurately classifying legal documents remains challenging due to the complexity of legal texts and subtle differences between legal categories. This study evaluates legal text classification models ranging from traditional machine learning techniques to large language models (LLMs) using ten categories of Korean sexual offense precedents. The results show that fine-tuning small-scale models such as KLUE-BERT on legal data outperforms general-purpose models such as GPT-3.5 and GPT-4.0, as well as traditional machine learning models. KLUE-BERT achieved the highest accuracy of 99.3%, indicating that domain adaptation and fine-tuning can be more important than model size for legal document classification. We further employ explainable AI (XAI) techniques to analyze model predictions and misclassification cases. XAI analysis identifies linguistic features influencing model decisions and limitations in capturing subtle textual cues. Using KICS data, which closely resembles real-world legal case records, we further evaluate the model's generalization capabilities and find that it struggles to interpret implicit contextual cues. These findings highlight the importance of both performance and interpretability in legal AI and demonstrate how XAI can improve transparency in legal text classification. AI-assisted tools can support legal professionals in tasks including document classification, legal information retrieval, and case assessment.

Critique in design education depends on interpreting work in progress, articulating intentions and translating feedback into revisions. This technical report presents Critsly, an artefact-aware AI critique workspace, and StudioCrit, its architecture-studio research mode. Critsly combines a visual board, design-intention fields, guided reflection, perspective-based critique and action planning. StudioCrit adds studio/class organisation, role-based access, cognitive and architectural classification, educator analytics and exportable evidence. The report consolidates implementation and simulation evidence recorded in a research project submitted in July 2026. Three simulated studio scenarios yielded 109 classified evidence rows, including 85 assigned to higher-order Bloom categories. A separate rehearsal using 50 disposable learner accounts yielded 56 evidence rows, including 46 assigned to higher-order categories. A subsequent hardening rehearsal recorded 50 completed sessions, 50 successful board pulls and 50 denials of student access to analytics. These are software and synthetic-trace observations, not measurements of learning gains or human cognitive performance. Automated classifications remain provisional, and the source report does not establish classifier accuracy or inter-rater reliability. The contribution is an implemented critique-to-evidence workflow and a bounded account of its readiness for further controlled evaluation.

Detailed profession-specific system prompts raise token use and estimated cost per response without a consistent accuracy gain. We evaluate Scientific Agents, an open-source corpus of 503 profession-specific AGENTS.md profiles, with Gemini 3.8 Flash via OpenRouter in the Pi agent harness. We compare matched profiles with four controls: a minimal baseline ("You are a helpful assistant"), the profile's opening role sentence, a generic scientific rigor guide, and a profile from an unrelated domain. Across nine text-based science benchmarks (4,531 sampled questions, 100 matched profiles), 4,488 items completed all five conditions after API-error retries, scored with automated, rule-based grading. The average profile-baseline accuracy difference is -0.6 percentage points (95% bootstrap interval [-1.5, +0.2] across fixed tasks), and no benchmark shows a statistically clear improvement. Matched profiles produced 1.5-2.3 times as many output tokens and cost 2.2-4.5 times more per successful call. On 60 tool-using BioMysteryBench bioinformatics problems (three runs each for baseline and profile), mean solve rates were 46.7% with the profile and 56.7% at baseline, a difference of -10.0 percentage points (95% interval [-16.7, -3.3]) driven by more frequent token- and time-limit stops under the profile. Longer prompts had one unexpected operational advantage: on SuperGPQA, frequent provider API drops left the short baseline with a correct first-pass answer on only 54.0% of items, against 71.6% with the profile. Generic and mismatched prompts were about as reliable, so this gain comes from prompt length or formatting rather than domain expertise. For the tested model and tasks, loading full profession profiles by default does not improve accuracy and costs considerably more; whether selective retrieval of profile sections or open-ended scientific tasks would change this remains to be tested.

For a long-horizon LLM agent, the memory question is not what was once recorded but what currently holds. Most designs answer it only indirectly: every interaction is stored, and the present is reconstructed at query time by retrieving and reconciling records, so stale values re-enter and the same conflicts are re-litigated. Committing the current value at write time avoids this, but existing write-time (CRUD) memories overwrite, so a wrong update is unrecoverable and prior state is lost. We take the missing combination—commit on arrival while retaining what is displaced—and formalize it as Memory of Memory (MoM): memory tracks not only content but the provenance, status, and history of its own entries. We instantiate MoM as Provenant Memory (P-Mem), a typed provenance graph whose active frontier exposes one current value per resolved key while displaced values are retained as provenance; typed operations decide whether a new observation supports, supersedes, contests, rejects, revokes, or resolves an existing value. P-Mem's decisive gain is validity rather than accuracy: its turn-level read matches the strongest retrieval memory in accuracy at $\sim$4$\times$ fewer read tokens—a retrieval-granularity effect—while graph-guided turn pruning cuts the knowledge-update stale-answer rate (19.4%$\rightarrow$10.9%); on revision chains it stays at 100% where query-time reading collapses to 25%, and, because displaced values are retained rather than overwritten, it recovers committed errors a CRUD memory cannot (100% vs.\ 0%).

Can one language model hand its live memory to another without the receiver rereading the context? We demonstrate useful persistent hybrid-state transfer across one architecture-matched Qwen3.5 4B-to-9B sibling pair. To our knowledge, this is the first demonstrated cross-model handoff of persistent recurrent inference state between differently sized hybrid language models without target prefix replay. Translated attention KV alone leaves a large gap; adding the Gated DeltaNet (GDN) persistent-state package lowers teacher-forced negative log-likelihood (NLL), the average next-token log-loss, by 0.747 nats/token (95% paired document bootstrap CI [0.6921, 0.8047]), improving all 64 PG19 documents. Direct recurrent and convolution reuse outperforms the tested learned GDN maps, consistent with partial functional compatibility of persistent-state coordinates. A fresh component factorial selects translated KV with direct recurrent and convolution state. An additional 434,176-parameter correction improves that base on 64 fresh web documents: continuation loss is 0.076 nats/token above native 9B (excess NLL), Jensen-Shannon (JS) divergence is 0.022, and native context recovery (NCR) is 0.918. Corrected 9B significantly beats continued 4B inference while processing zero historical prefix tokens. Evidence covers one direction, one geometry-matched Base-model pair, and 4K teacher-forced continuation; the near-native gate failed, the 16K branch was not run, and free-generation equivalence and a general state interface remain unproven.

State-of-the-art models for audio-driven digital human generation have achieved photo-realistic results in talking-head synthesis. However, extending these models to singing-head synthesis remains challenging due to a significant Domain Gap: singing requires more exaggerated expressions, vivid jaw openings, and precise rhythmic synchronization. Current models, primarily trained on speech datasets, often struggle with "rhythmic drift" and constrained dynamics. To address this, we introduce Hi-Singers, the first large-scale, high-quality, in-the-wild video dataset specifically tailored for singing head synthesis. Hi-Singers undergoes a rigorous automated and manual filtering pipeline, ensuring strict thematic adherence, high-resolution rendering, and stable motion, resulting in 29,608 video segments totaling approximately 170 hours. We further establish a dedicated evaluation benchmark balanced across linguistic and musical styles. Extensive experiments across diverse architectures, including 3D-coefficient and diffusion-based models, demonstrate that Hi-Singers consistently and significantly improves performance across all dimensions. Specifically, it enables models to achieve superior visual realism, enhanced lip-sync consistency, and more precise rhythmic dynamics, effectively bridging the domain gap and setting a new performance standard for the singing synthesis task. The dataset is available at https://huggingface.co/datasets/CharlesZhang-USTC/Hi-Singers

Medical time series (MedTS), including electrocardiograms (ECG), electroencephalograms (EEG), photoplethysmography (PPG), and vital-sign recordings, are central to clinical diagnosis and health monitoring. As large language models (LLMs) have advanced, a growing body of work has examined how their reasoning, generation, and knowledge-integration capabilities can support MedTS analysis. Yet existing studies remain scattered, and the field still lacks a clear view of how these models should be designed, integrated into clinical workflows, and evaluated. This review synthesizes recent work on large language models for medical time series analysis (MedTSLLMs), covering both methodological progress and issues related to real-world deployment. We review model architectures, data resources, and processing pipelines, and prompt design strategies adapted for diverse clinical scenarios. We further organize existing MedTS applications, ranging from diagnostic interpretation and report generation to longitudinal health monitoring and physiological signal synthesis, highlighting task-specific design choices, common evaluation protocols, and empirical findings reported across studies. By bringing together current practices and open challenges, this review aims to provide a clearer foundation for developing, evaluating, and deploying MedTSLLMs responsibly in healthcare. We also maintain a regularly updated list of MedTSLLM studies and resources at: https://github.com/hy727/MedTSLLM-Review.

In post-disaster environments, the failure of terrestrial communication infrastructure necessitates the rapid deployment of unmanned aerial vehicles (UAVs) as aerial base stations to restore wireless connectivity. This paper addresses the joint UAV activation-and-placement problem in continuous space, with the objective of minimizing the number of deployed UAVs while satisfying coverage and minimum-separation constraints. To solve this problem, we propose a Hybrid K-means Quantum-Inspired Evolutionary Algorithm (HKQEA) that combines K-means-guided initialization, a calibrated penalty-based feasibility objective, non-elitist evolutionary search, and a quantum-inspired learning update. Experimental results over 50 independent runs show that HKQEA attains a best fully feasible solution with 8 UAVs, while achieving average values of 98.94% for coverage, 99.94% for non-overlap, and 99.68% for minimum-distance satisfaction. Comparative evaluation against standard Non-dominated Sorting Genetic Algorithm II (NSGA-II), Particle Swarm Optimization algorithm (PSO) and an elitist variant of HKQEA further shows that the proposed method provides a more favorable balance among exploration, convergence behavior, and reliable feasibility preservation in constrained deployment problems. An illustrative procurement-level cost analysis also indicates that reducing the fleet from 10 UAVs to 8 can yield a 20% reduction in hardware count, corresponding to a simplified savings ratio of 25% for the studied deployment setting. These results demonstrate the potential of the proposed framework for resource-efficient post-disaster communication restoration.

Contract-clause extraction benchmarks measure reference recovery, but interpreting a system for review assistance also requires measuring unnecessary output and the evidence available within categories. This paper presents a retrospective, reproducible evaluation of a fixed public RoBERTa checkpoint on CUAD's published 102-contract test split. Candidate generation is preserved from an earlier frozen run; matching errors are corrected and category thresholds are reselected on 62 training-split contracts using the original 90% recall target. The corrected operating point recovers 82.2% of reference spans at 8.2% CUAD precision. Of 30,464 returned candidate strings, 19,562 occur on questions with no annotated answer. The system answers 1,335 of 2,938 such questions, a 45.4% false-positive rate (95% contract-bootstrap interval: 43.3-47.9%). Matching candidates constitute 20.3% of returned strings, illustrating why reference-based precision and candidate-level review burden require different denominators. Only nine of 41 categories have at least 30 positive and 30 negative contracts; this count changes to 18 and two under support minima of 20 and 40. Two categories have no negative contracts, making their no-answer false-positive rates undefined. These findings describe one checkpoint, decoder, and threshold policy. Earlier test access and uncertain checkpoint training overlap limit confirmatory interpretation. The paper supplies the original and corrected analyses, predictions, and software checks as ancillary artifacts; it establishes neither a new release criterion nor production readiness.

Gravitational-wave denoising must handle the full diversity of spinning, precessing binaries, since the recovered waveform underpins parameter estimation, tests of general relativity, and population studies. Matched filtering achieves this at a cost that becomes prohibitive as next-generation detectors push event rates higher; deep learning offers real-time reconstruction, but current methods are developed on narrow parameter spaces, precluding principled comparison and reliable deployment. We present the first controlled comparison of five neural-network architectures for gravitational-wave denoising, trained identically across the full astrophysically-motivated spinning binary-black-hole parameter space. A unifying principle emerges: matching network structure to the spectral anatomy of a coalescence – inspiral, merger, ringdown – outperforms brute-force model scaling. Our Multi-Scale Frequency-Aware architecture embodies this via dedicated parallel branches per frequency regime, achieving the best fidelity while using fewer parameters than larger models. It generalizes from simulated training to real LIGO-Virgo-KAGRA data without retraining, recovering merger morphology across confirmed events spanning three observing runs and both detectors, despite training on a single detector's simulated noise. We construct population-level uncertainty bands from denoising residuals, validate their calibration, and stress-test the framework on extended mass ratios and a spin population resembling hierarchical-merger remnants. Applied to real noise with no known signal, the network suppresses its output almost everywhere, with rare exceptions traced to noise transients rather than a general weakness – indicating the statistic discriminates signal from noise and motivating a future detection study. Released weights give a deployable, reproducible benchmark for future extensions.

Medical image segmentation needs diverse training data, but hospitals hold complementary scans they cannot share for privacy and regulatory reasons. Knowledge distillation can bridge this gap by exporting learned feature representations instead of images, but those representations still encode patient-specific anatomy and remain vulnerable to membership-inference and feature-inversion attacks. Adding calibrated Gaussian noise restores a differential-privacy guarantee, yet three issues have been overlooked. First, prior DP feature-distillation pipelines re-sample noise at every student iteration, so each patient image is released many times and the privacy cost composes over those releases, growing by orders of magnitude. We present a sample-once-per-image release, realized by a single precomputation pass, under which each patient contributes one release. Second, uniform noise is wasteful because channels differ in task importance. Using task-gradient energy as the importance measure, we derive CANAL, a closed-form water-filling allocation that gives important channels proportionally less noise, and prove it strictly minimizes importance-weighted distortion at a fixed budget. Third, the clipping caps and importance scores that drive the allocation are themselves data-dependent, so releasing them in the clear silently breaks the guarantee. We give a DP-honest budget split that charges each to the privacy budget, so the reported epsilon is the true epsilon. Across three medical segmentation benchmarks spanning dermoscopy, colonoscopy, and ultrasound, CANAL retains more task-relevant signal than uniform noise at the same privacy budget.

In distributed model predictive control for multi-drone collision avoidance, a fixed prediction horizon forces a compromise: a short horizon is inexpensive but reacts late to approaching neighbors, whereas a long one anticipates conflicts at a per-step cost that grows superlinearly with its length. We propose a conflict-predictive variable horizon that each drone sets locally, leaving the distributed model predictive control itself unchanged. From a short history of observed positions, a drone extrapolates the flight lines of its neighbors, tests each against its own using confidence funnels that narrow with prediction range, and obtains each time to conflict in closed form. The horizon is then the smallest admissible value whose planning window covers the farthest predicted conflict. It collapses to its minimum in clear airspace and grows only when a conflict lies ahead. Provided this minimum meets a single computable feasibility bound, we prove that recursive feasibility and asymptotic stability are preserved for every horizon the policy can select. These guarantees hold for a linear model, and a cascaded inner loop reduces each quadrotor's translational dynamics to a perturbed double integrator, so they carry over to the linearized quadrotor model and, as practical stability, to the full nonlinear one. In simulation on dense antipodal-swap benchmarks, the variable horizon reduces both per-step solver cost and total computation well below those of a long fixed horizon, and it maintains separation in every run, which a short fixed horizon of comparable per-step cost does not.

Gesture recognition on video is normally posed as classification: label each frame, then act on the label. That is adequate for control, where a command may be obeyed several frames late without a user noticing, and inadequate for synchronisation, where an output must be aligned to the frame on which the gesture physically occurred. We take the synchronisation problem for one common movement, the rotation of an open hand about its long axis, and show that it admits an exact solution needing no classifier, no training data and no calibration. Let s be the normalised two-dimensional cross product of the two palm edges at the wrist and the outer knuckles, under the projection the camera already performs. We prove that s factorises as k(theta)cos(theta) with |k| > 0 everywhere, so s vanishes exactly when the palm is edge-on and its sign tracks the face presented to the camera. Detection is therefore a zero crossing of one scalar, which yields an instant rather than an interval, and we prove the criterion invariant to image mirroring, hand scale and handedness, all from its algebraic form rather than from the landmark estimator. A two-handed interaction uses four fingertips as a window onto a restyled version of the same scene. We give a coverage predicate that stays correct when the hands cross, where the usual triangulation does not, and derive each free parameter of the three stylisation operators from a noise statistic rather than by inspection. Against a corpus with exact ground truth the criterion detects 95% of flips with no false positive in 240 near-miss sequences, and places each within 6.7 ms on average of the instant it occurred, a sixth of the interval at which the hand is observed, against 23.0 ms for reporting the bracketing sample. Reporting an event more finely than one samples follows from treating a gesture as the zero of a continuous quantity rather than a label.

Annual reports, diligence packs, and infographic dashboards bury numbers in page images: axes, cell grids, and footnotes that OCR pipelines flatten and that page-level visual retrievers still treat as interchangeable in-context examples. We keep a frozen Qwen2.5-VL-7B-Instruct generator and a ColPali / VisRAG-Ret page index, and insert three modules. A capability-aware visual router (CAVR) tags each retrieved page as text, table, chart, layout, or mixed and mixes specialist experts before generation. Weak-to-strong page selection (WSPS) distils a frozen 7B answerability teacher into a 3B selection head so ranking is no longer a single InfoNCE score. Visual evidence threading (VET) builds layout-anchored paths of length at most three and lets the generator read the thread rather than a flat top-$k$ list. On gold-page DocVQA / ChartQA / InfographicVQA the 7B system reaches 96.3 / 90.1 / 85.4. Under the VisRAG top-3 protocol the mean generation accuracy is 62.74 versus 59.39 for the same backbone with concatenation. On MMLongBench-Doc retrieve-then-read, F1 moves from 19.2 to 22.6 and multi-page accuracy from 16.4 to 21.2. ViDoRe nDCG@5 after WSPS reranking is 83.6, with TAT-DQA financial reports at 70.4.

Scientific charts encode quantities in axes, legends, and geometric marks, yet large vision-language models still treat them as natural photographs. Visual in-context examples do not expose the coordinate frame; unconstrained chain-of-thought can name a plausible number that was never read from a bar. We present STEER (Structure-Token Evidence-anchored Reasoning), which freezes a Llama-3.2-Vision encoder and inserts three modules: a chart structure graph encoder (CSGE) that binds ticks, legend items, and marks; evidence-anchored step reasoning (EASR) that forces every arithmetic step to cite a graph node; and weak-parser strong-reasoner alignment (WPSR) that uses a specialized table extractor only as a teacher of node attributes. On ChartQA, STEER reaches 82.70 average relaxed accuracy versus 80.16 for ChartGemma and 76.40 for a LLaVA-CoT backbone trained on the same mix. Gains widen on CharXiv reasoning (33.60 vs. 29.20 InternVL Chat V1.5) and ChartQAPro CoT (40.70 vs. 37.17 Qwen2-VL-7B), where OCR shortcuts disappear. Ablations show that dropping node serialization or numeric candidate constraints undoes most of the reasoning lift.

Near-infrared (NIR) imaging does not consistently outperform standard color cameras for daytime agricultural traversability once spatial data leakage is eliminated. Prior benchmarks suggesting an NIR advantage used sequence-level splits that permitted spatially autocorrelated imagery into test sets, artificially inflating NIR performance, especially on difficult paddy-boundary segmentation. When evaluated across strictly held-out recording sites using the AI Hub autonomous driving corpus, none of the four tested configurations (color, NIR, a luminance control, or their fusion) reliably surpasses standard color across any of the five traversability classes. Furthermore, apparent performance gains observed at a single held-out site completely fail to replicate when the held-out site is rotated, demonstrating that the observed benefits were site-specific artifacts rather than generalizable improvements. Correcting for spatial leakage erases the apparent lead of NIR entirely rather than uniformly degrading performance across all sensor types. Consequently, adding a forward-facing NIR camera to a daylight agricultural vehicle sensor suite remains unproven, and evaluations compromised by location leakage risk distorting sensor rankings and misleading procurement decisions.

Generating human-centric videos that preserve both visual identity and person-specific expressive behavior remains a fundamental challenge. In addition to reproducing appearance, a model must replicate the facial behaviors that characterize how a subject expresses emotion over time. However, most state-of-the-art methods condition generation on a single reference image, which contains no information about these temporal dynamics. As a result, they tend to preserve the subject's visual identity but often produce expressions with limited variation and weak subject specificity. To mitigate this issue, we introduce BEACON, a lightweight framework for person-specific video generation that produces more expressive videos by disentangling visual identity from expressive behavior. BEACON conditions generation on two complementary signals: a reference image encoding the identity and a reference video capturing subject-specific facial dynamics. By conditioning on these complementary signals, BEACON generates videos that better preserve both the subject's appearance and characteristic facial dynamics, while also supporting identity-expression transfer. Our experiments on the MEAD and RAVDESS datasets show that by fine-tuning on approximately 2,000 pairs and updating about 1% of the pretrained Wan video diffusion model, BEACON improves facial expressivity over state-of-the-art video generation methods while maintaining competitive identity preservation.

Projective shape analysis provides a geometric framework for studying landmark configurations in digital images acquired by pinhole cameras. In the classical projective shape (PS) model, three-dimensional configurations ($k$-ads) are represented as points in $(\mathrm{RP}^3)^q$, $q = k - 5$. A nonparametric test is developed in Patrangenaru et al. [12], for this framework, to determine whether an object matches a design blueprint, with each configuration reconstructed from a single uncalibrated stereo pair. Such two-view reconstructions are identified only up to a 3D projective transformation, which may reverse orientation, so that the sign-blind PS summary was the only available, before oriented projective shape (OPS) was recently considered. Multi-view Structure-from-Motion (SfM) technology removes this obstruction: its bundle adjustment is identified up to an orientation-preserving projective transformation. In this paper we revisit a well-cited three-cube object, from Patrangenaru et al. [12], with $n = 8$ SfM reconstructions built in Agisoft Metashape Professional 2.3.0, and validate our implementation by reproducing the published stereo construction from the original data. This allows us to perform what is to the best of our knowledge the first three-dimensional OPS analysis, compute its extrinsic total-variance index and perform statistical inference in this novel setting. Due to the high concentration of SfM data, the OPS index is asymptotically one-half the PS index, a structural consequence of concentration rather than a property of the object. Here our blueprint hypothesis is not rejected for any of the $q = 14$ non-frame landmarks, while the SfM reconstructions are about 26 times more concentrated than the stereo ones, a substantial gain in reconstruction precision. Sample-size, photograph-count, and frame-ordering analyses support the robustness of these conclusions.

An OpenReview bug in November 2025 broke anonymity at several conferences and prompted calls for open review, which motivate us to ask what shifting from blind to open would mean for authors. Analyzing over 18,000 reviewed submissions to ICLR 2026, split into de facto open and blind groups by arXiv preprint timing, we find that ratings rise with author reputation under both mechanisms, with a steeper slope under open review that is statistically significant, and that the open-blind difference is concentrated at the borderline ratings. The pattern holds across five reputation proxies (including institution, h-index, and citation count), three author-aggregation rules, and five definitions of the open window. A controlled simulation with five AI models as reviewers, holding the manuscript fixed and varying the author reputation, reproduces the effect. With claude-opus-5 as the reviewer, for example, rating rises by 0.5 points as the author moves from low to high reputation.

We give a deterministic algorithm that computes the parity of the number of Hamiltonian cycles in an $n$-vertex directed graph in $O(n^4(3/2)^n)$ time and $O(n^2)$ bits of working space, improving the $O^*(\varphi^n)$ bound of Björklund and Husfeldt. Their local-degree formula reduces the problem to a weighted sum over solutions of structured quadratic equations. We cover the corresponding ternary state space by binary subcubes, each inducing an affine system. The Kuang–Wang cover can be regenerated within the target bound; canonical ownership resolves its overlaps, while self-loop conditional expectations bound every affine solution visit. Rollback elimination shares the work across cover prefixes. The same cover gives a Las Vegas algorithm listing all $L$ solutions of $m$ affine product constraints in $N$ Boolean variables in expected time $\operatorname{poly}(N,m)((3/2)^m+L)$ and polynomial space. Finally, we show that complete enumeration can require $\Omega((3/2)^n)$ visits even on strongly connected digraphs after an optimal self-loop choice. This is a limitation of the enumeration method, not a general lower bound for Hamiltonian-cycle parity.