Papers

Weak-to-strong generalization is the phenomenon where a strong student model trained with labels produced by a weak teacher model is able to generalize better than the teacher. In this paper, we study this phenomenon in two-layer random feature networks where the model strength is determined by its width. Using tools from random matrix theory, we derive deterministic equivalents for the population errors of an optimally trained teacher and a student trained with gradient flow. For ReLU activation and a pure spherical harmonic target, we obtain sharp asymptotics under a Gaussian universality assumption, showing a quadratic improvement: the student error scales as the square of the teacher error. These results attain the general lower bound of Medvedev at al (2025). We also analyze how the student behaves under more general stopping times and targets supported on multiple harmonic degrees, characterizing the regimes in which weak-to-strong generalization occurs and identifying the transition between quadratic, non-quadratic, and no improvement.

In human-AI collaborative decision making, human review can prevent unsafe AI decisions, but each human judgment is costly. Treating human intervention after AI abstention as a one-off fallback misses the opportunity to improve future AI decisions for greater automation, yet AI adaptively learning from selectively queried human feedback breaks safety guardrails calibrated for old models. We approach this challenge with CARE—calibrated adaptive rectification and escalation—an end-to-end pipeline that combines AI models and human reviewers to guarantee safe, human-aligned decisions, while continuously learning from human feedback to achieve greater automation with fewer human queries. CARE is principled, general, modular, and works with any black-box AI model. Our novel adaptive calibration module guarantees risk control at every time step for any rectification module. We further show how CARE improves query efficiency when the AI model is well trained and the human-AI misalignment has a clear structure. Experiments on four safety-critical real-world datasets spanning driving, language, and robotics demonstrate that CARE achieves human-aligned decisions while reducing human queries by 25-81% relative to baselines.

We study which graph classes determine every graph with simple spectrum up to isomorphism by homomorphism counts from their members. We show that such a class has unbounded treewidth and Euler genus and contains a graph with the complete graph on seven vertices as a minor. To prove these conditions, we construct pairs of non-isomorphic cospectral graphs with simple spectrum, each bipartite, nonsingular and of maximum degree at most six. The construction attaches rooted trees to Cai-Furer-Immerman graphs and preserves homomorphism indistinguishability over every minor-closed class. We then prove that subcubic homomorphism counts determine every graph whose adjacency matrix has kernel dimension at most one. This extends the known result for nonsingular graphs and applies to all graphs with simple spectrum. We also prove that two graphs have the same subcubic homomorphism counts if and only if they have the same number of vertices and there is a bijective linear isometry between the images of their adjacency matrices that intertwines the adjacency operators and preserves the coordinatewise trilinear forms. Finally, we prove that homomorphism counts from the single-vertex graph and from subdivisions of any sequence of connected cubic graphs have the same distinguishing power as subcubic homomorphism counts if every complete graph is a minor of some member of the sequence.

Navigable graphs can be built without searching for neighbors: partition the data, evaluate every pair inside each part, and select each point's edges from the candidates. We give such a construction in three separable stages and show that the middle one decides the quality. The pool is any partition with a few memberships per point. The ending turns a point's candidates into out-edges; ours keeps a bounded heap, prunes by occlusion with a per-corpus slack, and appends reverse edges, re-pruning only where a list overflows. The spine is any edge set, exempt from the prune, that keeps the graph reachable from its entry; ours, half-space-proximal edges over a random sample, routes monotonically to every sampled point and replaces a spanning tree at 1/10 to 1/500 of its cost. The ending composes with any partitioner: on PiPNN's own candidate pool it beats PiPNN's ending on each of six corpora from $10^6$ to $10^8$ points, by 3 to 14% in distance evaluations at equal recall, and with 60 to 120 memberships per point the composed build matches or beats a full dense construction at k=10 and k=100 on all six, in 0.5 to 0.9 of its build time, deterministically. The analysis explains why. Once a pool is localised its quality is set by the data: every pool built on GIST lands within 4% of the exact-kNN ceiling, and the pairs a block cover misses are predicted, point by point, by the local clustering of the kNN graph, whose zero-clustering tail sets the memberships a corpus needs and grows with n. All code, patches and logs are public.

Constraint handling methods typically intervene before the optimizer acts, by modifying the objective or the gradient. Yet momentum, adaptive scaling, and structured preconditioning can substantially reshape that signal before it becomes a parameter update. We formulate optimizer relative constrained learning, where constraint compatibility is assessed on the post optimizer update. Building on this view, we introduce ORACLE, which evaluates the native optimizer's realized step through a joint endpoint linearization of heterogeneous constraint families, constructs the resulting alignment in the optimizer's own geometry, bounds its authority, and commits it only after validation. We evaluate ORACLE across eight Partial Differential Equation benchmarks and four optimizers spanning Euclidean, diagonal adaptive, and structured preconditioned geometries, where it improves or matches native optimizer in 94% of configurations. Cross model analysis shows the same behavior in 92% of configurations, while matched comparisons show improvements over alternative constraint-handling methods acting at the objective, gradient, and post-optimizer levels.

Large language models can extract useful signals from heterogeneous enterprise data, but high-recall extraction often produces outputs that are duplicated, uneven in granularity, semantically overlapping, or too numerous for downstream systems and human reviewers to use effectively. We present a dataset-adaptive post-processing architecture developed for Customer Intent Extraction (CIE), where unstructured customer language is transformed into stable, traceable intent units. The approach separates recall-oriented extraction from utility-oriented reduction. Source-specific preprocessing first isolates evidence from multimodal plans, sparse operational records, and structured opportunity data. Candidate intents are then standardized and deduplicated, optionally enriched with metadata for embedding computation, represented in a shared semantic vector space, and grouped using a clustering strategy selected according to the candidate set's characteristics. Cluster-level keywords provide an explainability layer, while singleton reassignment requires agreement between embedding and keyword similarity. Finally, constrained language-model aggregation produces one concise intent per cluster without introducing unsupported concepts, and the resulting unit retains provenance, clustering, embedding, and generation metadata. This treats post-processing not as cosmetic cleanup, but as a semantic reduction layer converting high-recall LLM outputs into reusable enterprise intelligence. We also describe two downstream applications: Machine-Generated Intents, which infer likely objectives for customers lacking direct evidence from peer customers with similar profiles, and intent-guided semantic retrieval and mapping, which uses the stable intent as a query against a downstream decision space, illustrated here by mapping customer intents to business outcomes.

We study long-time mean-field large deviations for granular-media diffusions. Under a dissipativity assumption, for arbitrary $T_N\to\infty$, the occupation measure of the $N$-particle empirical measure satisfies a large deviation principle on $\mathcal{P}(\mathcal{P}(\mathbb{R}^d))$ with speed $NT_N$ and rate function $\mathcal{I}(\Gamma)=\int_{\mathcal{P}(\mathbb{R}^d)}\mathfrak{i}(\mu)\Gamma(d\mu)$. Here $\mathfrak{i}(\mu)$ is one quarter of the free-energy dissipation: $\mathfrak{i}(\mu)=\frac14\left\|\nabla\frac{\delta\mathcal{F}}{\delta\mu}(\mu)\right\|_{L^2(\mu)}^2$, where $\mathcal{F}$ is the free energy of the model. The local cost $\mathfrak{i}(\mu)$ can be interpreted as the per-unit-time Dawson–Gärtner action of the constant path at $\mu$, and $\mathcal{I}$ averages these costs according to $\Gamma$. The main difficulty in the upper bound is that empirical measures are atomic, so their free energy is infinite. We address this by analyzing a regularized free-energy functional along the empirical-measure flow of controlled granular-media diffusions. The lower bound uses approximation by finitely supported occupation measures whose atoms have smooth densities with Gaussian tails. For quadratic potentials, we show that the variational problem associated with an atypically large overall space-time variance exhibits a transition between a Dirac minimizer at a Gaussian profile and non-Dirac minimizers supported on Gaussian profiles with different centers.

Supervisory governors can interfere with the tool-using agents they regulate. We study this possibility in a controlled file-recovery environment where increases in regulatory intensity trigger experimentally imposed tool failures. A cost-blind governor can turn these failures into persistent blocking that prevents task completion. We compare this governor with a backoff rule that reduces intervention probability using a moving average of known induced events. On a hand-coded stochastic-policy agent, the failure pattern appears under both result replacement and execution of corrupted tool arguments. For the persistent policy, adaptive backoff improves completion relative to a fixed weak governor with approximately matched intervention frequency. A Gemini 2.5 Flash experiment comprising 576 episodes across 6 tasks also shows reduced blocking and improved completion under backoff; among the tested settings, intermediate backoff strength achieves the highest observed aggregate success. These results identify an interaction between intervention cost and persistent action blocking, together with a possible mitigation. The cost mechanisms are imposed and their induced events are directly observable to the backoff rule; applicability beyond this controlled environment remains an empirical question.

When Coulomb interactions dominate kinetic energy, dilute two-dimensional carriers form a Wigner solid. Both its melting and disorder-driven localization can drive density-tuned metal-insulator transitions, but their relative roles and interplay have been debated for decades. Atomically thin semiconductors offer a new test because transport and scanning tunneling microscopy can probe the same material, yet poor contacts impede transport where correlations are strongest. Here we introduce contact-free radio-frequency reflectometry that senses sheet resistance through capacitive gate coupling. It resolves the electron and previously inaccessible hole transitions in the same bilayer MoSe2 device, whose carriers share disorder but not mass. Their critical densities differ fourfold yet correspond to comparable Coulomb-to-kinetic-energy ratios, where tunneling microscopy finds solid-liquid coexistence. Magnetoresistance peaks sharply at each transition, reflecting a Pomeranchuk-like, field-driven expansion of the Wigner solid against the correlated liquid. Our results demonstrate interaction-driven transitions in high-quality two-dimensional semiconductors, opening quantitative transport in contact-limited systems.

Antiferromagnets (AFMs) host magnons that can exhibit fully circular, elliptic and linear polarizations of sublattice magnetization dynamics. We analyze theoretically different momentum-space magnon polarization textures and magnon transport in collinear AFMs and altermagnets (AMs). The sensitivity of the polarization texture to a magnon momentum direction determines two regimes of magnon polarization kinetics: The Dyakonov-Perel relaxation or its dephasing due to asynchronous Larmor precession. For an easy-axis AFM, magnetic field parallel to the Néel vector leads to the suppression of the magnetic dipolar interaction-induced Dyakonov-Perel relaxation, accompanied by the increase in the magnon spin-diffusion length and its counter-intuitive growth with temperature. For a perpendicular orientation, instead, relaxation of circular polarization of magnons is enhanced due to Larmor precession. Finally, extremely anisotropic relaxation of magnon polarization is predicted for AMs or AFMs having Dzyaloshinskii-Morya interaction.

Multi-agent path planning (MAPP) in continuous environments often relies on roadmaps to balance safety and search efficiency. However, traditional roadmap generation methods, such as lattice grids or standard sampling-based approaches, frequently face a trade-off between graph density and the likelihood of finding feasible, high-quality solutions. In this paper, we propose a scalable heterogeneous Graph Neural Network (GNN) framework for the automated generation and evaluation of shared multi-agent roadmaps. Our model covers the representation of waypoints, agent locations, and task locations as distinct nodes in a heterogeneous graph, allowing it to reason over global connectivity and inter-agent interactions. By training on occupation density maps aggregated and collected from expert solver trajectories, the GNN learns to identify critical points of interest and prune redundant nodes and edges. This process produces a compact, coordination-aware roadmap that is invariant to task permutations and is reusable for multi-agent pick and delivery tasks. Experimental results demonstrate that our framework can reduce planning effort and can potentially find better solutions, reaching at least 40% reduction in runtime and in graph size for dense roadmaps.

Standard safety evaluations of large language models assess harmful requests written in canonical plain text, while models in real-world deployment routinely receive inputs containing emojis, altered spellings, encoded strings, and character-level variations. This work introduces the Adversarial Surface-Form Robustness Dataset (ASRD), comprising 2,100 prompts across seven distinct surface-form families. Five open-weight language models are evaluated across these prompts, producing 10,500 responses. The Quad-State Evaluation Rubric classifies each response into one of four outcomes: harmful compliance, safe response, comprehension failure, or indeterminate. Emoji and invisible Unicode variations cause almost no comprehension failure, with pooled harmful compliance of 20.27% and 17.20% against a 22.87% baseline that is driven mainly by Mistral 7B, whereas leetspeak, encoded wrappers, and hybrid transformations score 2.40%, 0.13%, and 2.40% while comprehension failure rises to 36.47%, 65.60%, and 34.47%. Inspection of raw model outputs reveals three response behaviors: hallucinated benignity, structural collapse, and language drift. Project page: www.pavanmaddula.com/quadstate

Perceiving structured shapes, such as human faces, from pixels is an inherently ambiguous task in real-world conditions. Yet, shape inference is largely posed as a deterministic regression task predicting fixed spatial coordinates. We find that deterministic regression is brittle when visual evidence is ambiguous or incomplete; under severe occlusions deterministic models exhibit structural collapse, predicting incoherent shapes or reverting to generic averages. To address this, we introduce Shape-Bayes, a probabilistic framework that couples uncertainty-aware visual perception with Bayesian shape reasoning. Rather than forcing point estimates, Shape-Bayes dynamically weights visual evidence against geometric priors to infer a structurally valid shape posterior. Demonstrated on human face shape regression, a rigorous testbed featuring complex non-rigid deformations and strict anatomical constraints, Shape-Bayes comprises: (1) a base model predicting noisy landmarks alongside distilled aleatoric uncertainties; (2) a lightweight Transformer encoding these observations into an adaptive prior over a PCA shape manifold; and (3) a differentiable Bayesian solver computing closed-form posteriors by balancing the noisy predictions against this prior. By guaranteeing complete structural integrity, Shape-Bayes achieves an absolute improvement of up to ~34% IDR over state-of-the-art deterministic models. Simultaneously, it yields highly calibrated uncertainty bounds and reduces relative error by up to 12.5%, establishing a new state-of-the-art for robust 2D face shape regression under severe occlusion. The project page is at https://shape-bayes.github.io.

Medical imaging models often operate as black boxes, limiting interpretability and systematic debugging. We introduce an easy-to-use, plug-and-play framework for concept-based interpretation and model refinement. By aligning a single-modality encoder to BioMedCLIP, we construct a Concept Bottleneck Model (CBM) that enables concept-level interventions. These interventions allow us to isolate causal versus spuriously correlated concepts, validate insights with domain experts, and generate counterfactual samples for targeted fine-tuning. We evaluate our framework on a Mayo Clinic ultrasound dataset and the CheXpert 5x200 chest X-ray dataset. Results demonstrate that concept intervention enables reliable model diagnosis while maintaining, and occasionally improving predictive performance via guided fine-tuning. Our findings highlight the practical value of this framework for controlled, interpretable refinement of clinical deep learning models.

Pairwise comparisons are widely used to infer latent scores and identify top-ranked items. Although sharp statistical guarantees are available under uniform sampling, real data often induce irregular comparison graphs with heterogeneous observation counts across pairs. In this paper, we study an arbitrary fixed comparison graph under the Bradley–Terry–Luce model with minimal assumptions. We prove that both the standard maximum likelihood estimator and Rank Centrality achieve a high-probability entrywise error rate of order $1/\sqrt{\lambda_{\mathcal{D}}}$ up to logarithmic and dynamic-range factors, where $\lambda_{\mathcal{D}}$ is the algebraic connectivity of the count-weighted observation graph. This guarantee is graph-monotone since $\lambda_{\mathcal{D}}$ cannot decrease when additional comparisons are added. Under heterogeneous sampling, where comparison pairs are sampled independently with unequal probabilities, our guarantee improves existing error bounds or requires weaker assumptions. We further extend our analysis to show that both estimators are robust against outcome-adaptive augmentation, where an adversary can choose additional comparison pairs after observing the initial outcomes.

Organisations deciding which assets to migrate to post-quantum cryptography first, and how that work competes with a backlog of classical findings, lack a common unit: post-quantum scores are dimensionless and quantum-only, while classical risk is annualised loss. We express both threats as a cost of delay in currency per year on the same asset. The quantum term is the rate at which deferring migration commits irreversible harvest-now-decrypt-later loss, computed from required confidentiality duration, recordable traffic share and data turnover, using a quantum-arrival law fitted in closed form to published expert anchors. A competing-risks factor couples the two terms so the same loss is not counted twice. We prove that the construction is not a weighted sum of a classical and a quantum score, give a dominance threshold on the classical hazard that is independent of asset value, and show that the induced order is optimal for sequencing work under constant loss rates. We verify the implementation by simulation, adaptive quadrature and a second implementation. On four public systems with parameters assumed from public documentation, the quantum term reorders the asset list beyond input uncertainty for one system only (signal-to-noise 1.36 against 0.14-0.46), moves specific long-lived, quiet, recordable assets decisively, and otherwise changes magnitudes. Asset value and turnover explain 73-91% of the uncertainty in the quantum term; the arrival date explains 3-17%. The case studies are illustrative and are not validated against outcomes.

A combination of remote sensing techniques, open space data and field testing was developed to derive astronomical and meteorological parameters and assess Dark Sky brightness of the hitherto unstudied Schist Villages region in the mountainous Interior Center of Portugal. This study covered the period 2012-2017 and established the methodology for regular light pollution and sky brightness studies in the region using satellite imagery to reduce cost and time and increase accuracy of the final analysis. Long-term Artificial Light At Night (ALAN) data from VIIRS instrument aboard the Suomi-NPP satellite along with cloud cover inferred from MODUS/AQUA satellite data and orographic data, were analyzed in a Geographical Information Systems (GIS) environment. The derived ALAN values were consistent with a preliminary site survey selection, confirming the accuracy of the identified dark sky spots in the region. The analyzed region area spans 1624 Km$^2$ and the darkest site within it showed a sky brightness of 21.4 mag/arcsec$^2$. Ground-based SQM photometer observations of the Night Sky Brightness (NSB) at the zenith validated the satellite derived ALAN data. The devised field testing is itself a low cost solution that can be used by teams in developing countries to easily assess dark sky potential prior to the development of local observatories and astro-tourism activities. This study supported the initial certification of this region as a Dark Sky Starlight Tourist Destination since 2018, an important milestone for dark sky protection in central Portugal.

Modern language model systems operate autonomously over increasingly long contexts containing untrusted text and images. Can an adversarial input continue to steer a model even after that input is removed from its context? We show that attacks can be trained to persist through the key/value (KV) cache of subsequent tokens, allowing adversarial influence to outlive direct access to its source.We consider the Visual Memory Injection (VMI; Schlarmann and Hein, 2026) attack setting, in which an adversarial image that stays in the context plants a hidden backdoor: the model behaves normally until a chosen trigger elicits an attacker-chosen response. We first demonstrate persistence in this setting with optimized soft prompts, which remain effective after we mask the prompt from attention. We then introduce Persistent Visual Memory Injection (P-VMI), which optimizes images to preserve this adversarial behaviour after they are masked from attention. These attacks persist over conversations substantially longer than those used during optimization. On Qwen3-VL-8B-Instruct, P-VMI achieves up to approximately $90\%$ target success in its strongest configuration and remains effective under a stricter removal setting that exposes the image only on the first turn. A cache-swap ablation localizes the persistent influence to the KV cache. Finally, we show that these attacks can be trained to survive compaction that retains the KV cache of a summary generated by the same model, demonstrating that adversarial behaviour can persist in cached state without continued access to its source.

We present BEACON-SP, an ontology-grounded Graph Retrieval-Augmented Generation (GraphRAG) framework for clinician-facing decision support in behavioral health settings such as suicide prevention, where effective assessment requires integrating heterogeneous clinical, behavioral, social, and temporal evidence. BEACON-SP combines patient knowledge graphs with ontology-guided retrieval to support multi-hop reasoning across diagnoses, medications, risk and protective factors, life events, and temporal relationships. The framework is enabled by a comprehensive suicide prevention ontology that integrates the Three-Step Theory, the Integrated Motivational-Volitional Model, and the Suicide Social Determinants of Health Ontology into a unified representation of patient risk factors. We construct ontology-grounded patient knowledge graphs and evaluate BEACON-SP for clinician-facing question answering. Compared with a vector-based retrieval-augmented generation (RAG) baseline on a 1,500-query benchmark spanning 15 clinical categories and 100 patients, BEACON-SP improves completeness, clinical relevance, and evidence grounding under a corrected comparative evaluation protocol, with a small gain on factual accuracy. In paired criterion-level comparisons, GraphRAG is preferred in 76.4% of cases. These results demonstrate the potential of ontology-guided GraphRAG to provide structured, contextualized patient evidence for clinical decision support.

Indexer-based sparse attention reduces the cost of core attention by passing only a fixed, small number of important tokens to it. However, the indexer must still score the entire KV cache at every decoding step. This scoring overhead becomes a major bottleneck as the context length grows. We propose SPIN (Shadow Predictive Indexer) to reduce this indexer overhead. SPIN uses lightweight, history-based prediction to identify important KV blocks, avoiding the need to score the full KV cache at every decoding step. SPIN treats KV blocks and speculative decoding as first-class design and implementation considerations. Across extensive evaluations on long-context and agentic benchmarks, SPIN achieves 30-40% sparsity while preserving task quality. In end-to-end vLLM serving, SPIN improves output throughput by up to 14.9% and reduces median inter-token latency by up to 13.2%.

Martian dust particles are a few micrometers in diameter, comparable to the mean free path of the gas in the hypersonic shock layer. Their Knudsen number ranges from approximately 0.2 to 100, placing them in the transitional and free-molecular regimes, whereas the flow around the vehicle is a continuum. The two-way flow-particle coupling previously developed by the authors for a two-temperature thermochemical nonequilibrium flow is extended to a receding and pyrolyzing heat shield surface. The gas-surface interaction and the in-depth material response are solved on separate time scales. The surface balances are exchanged at every flow step, while the in-depth response is coupled implicitly over physical time windows. The flow-material coupling is assessed against graphite arc-jet measurements. The framework is applied to the Schiaparelli trajectory from 50 to 30 km. Three surface chemistry models (equilibrium, Park-5, and Zhluktov-Abe) are compared for fixed geometry, char-induced recession, and combined char ablation and particle erosion. At the final trajectory point, the stagnation point heat flux differs by 30-75% among the chemistry models. For the equilibrium model, surface recession changes this heat flux by less than 1% but lowers wall temperature and increases pyrolysis gas injection. Particle erosion contributes 1.73 mm of stagnation point recession, exceeding the char-induced recession, and more than doubles the total recession for all three models. To the authors' knowledge, this study provides the first quantitative comparison of particle erosion between rarefied and continuum drag correlations, and the continuum correlation underestimates the stagnation point erosion rate by 31-35%.

Performance issues in software systems are a critical quality issue that can erode user trust, violate service-level agreements, and ultimately affect business efficiency. Consequently, software performance engineering has shifted its focus to developing robust techniques to detect performance regressions as early as possible in the development cycle. Performance regression analysis often relies on time series of performance measurements to detect significant changes in performance behavior. Change Point Detection (CPD) methods have been widely used to automate the identification of such changes in various domains, including finance, healthcare, and performance monitoring. However, the effectiveness of these methods for software performance measurements has not been thoroughly evaluated. In this paper, we present a comprehensive study to evaluate the effectiveness of various CPD methods on real-world software performance measurement datasets. We start by collecting performance measurement data from three large software systems and characterizing the unique properties of performance time series data. Then, we undertake a large-scale effort to annotate and evaluate the consistency of human annotators' identification of potential performance changes. Thereafter, we evaluate the accuracy of twelve distinct CPD methods in detecting potential performance changes, providing insights into their applicability and effectiveness in software performance regression analysis.

Alice and Bob each have a qubit, and they would like to perform a two-qubit gate on them with the assistance of Charlie. The action of this gate depends on a parameter, which either Charlie knows, will receive later as classical information, or is encoded in a program state. We discuss both the case in which Alice and Bob send Charlie their qubits, which he entangles with his state, and the case where Alice and Bob use teleportation to entangle their states with Charlie's state. All of these procedures depend on the ability of Alice and Bob to apply local correction operations that depend on the results of measurements that Charlie makes, and we discuss under what conditions such operations are possible.

An agentic system issues several structurally different kinds of LLM calls. It routes intent, classifies actions, grounds language in a device registry, plans multi-agent pipelines and writes the Python code those pipelines run. The difficulty of these call sites varies by an order of magnitude, yet in practice a single model, chosen for the hardest site, serves all of them. In this work, we evaluate 9 models from 0.8B to a frontier hosted model across the five call sites of a deployed open-source home-automation framework (Wactorz), using its unmodified production prompts and two real Home Assistant installations (280 cases, 2520 scored calls). We find that capability is not ordered the same way at every site, and that larger models are not uniformly better: one 4B model is worse than its 2B sibling at grounded actuation. Paired testing shows the best local model to be statistically indistinguishable from both hosted models at four of five sites. Only code generation separates them, against a small hosted model (p = 0.039) as well as a frontier one (p = 0.002). Aggregate accuracy also hides a safety failure specific to actuation, where small models resolve the accuracy/refusal trade-off in degenerate ways: one model (Gemma4 E2B) actuates on 87.2% of requests for devices the site does not own, while another refuses every request it receives. Routing each site to its best local model reaches 91.8% against 95.4% at no per-call cost. In a live deployment judged by a user, hosting only the two generative sites matches hosting everything (39/43 against 39/43) for 28% of the spend, and the actuation gap the benchmark predicted appears as exactly one case in twenty-six. Benchmark, harness and all records are released at https://github.com/waldiez/slm-callsite-eval.

Modern neural networks are often miscalibrated, with a tendency to overconfidence. Existing train-time calibration methods largely modify task losses or calibration penalties, leaving neighborhood structure in learned representations underexploited. We introduce graph smoothing as a general principle for train-time calibration, which encourages similar predictive distributions across neighboring samples in representation space. We analyze the effects of graph smoothing, deriving bounds that connect predictive divergence between neighboring samples to local confidence variation and to the propagation of pointwise calibration error, and characterize the conditions under which smoothing can or cannot improve calibration. In light of this analysis, we propose \modelNoSpace, a graph-based train-time regularizer that penalizes the Jensen–Shannon divergence between predictive distributions of neighboring samples. We present a thorough empirical analysis, showing that across standard calibration benchmarks, \model improves predictive quality, and the improvement is complementary to post-hoc calibration: after temperature scaling, \model attains the lowest NLL of all evaluated train-time methods in seven of the eight image and tabular settings. These findings demonstrate the value of graph smoothing over learned representations for neural network calibration.

In this paper, we construct and analyze a stochastic model for infectious disease spread in service systems. Our model considers a $M/M/1$ queue where a proportion, $p$, of the arriving population is infectious and the remaining population is susceptible. When an infectious person overlaps in the queue with a susceptible customer for more than $\ell$ units of time, then the susceptible person becomes infected. We provide both a personalized perspective of how an individual susceptible customer is infected and a community perspective of how one infectious customer can infect the others around them. From the personalized perspective, we derive the probability that a susceptible person becomes infected and the probability they are infected by at least $k$ different people. From the community perspective we compute the mean and higher moments of the number of susceptible customers that are infected by one infectious individual. Finally, we use stochastic simulation to verify our analytical results and show that our new formulas accurately describe the behavior our stochastic model of infection and to expand our model into the consideration of different arrival and service regimes.

Exponential directed last passage percolation is a random maximization model connected to tandem queues and interacting particle systems. We study the line-to-point model of the percolation, in which the starting point is allowed to vary inside a line boundary, conditioned on an atypically large last-passage time to a fixed endpoint. Extending recent results for the point-to-point model where the starting point is fixed, we develop heuristic predictions for the conditional limit shape and prove corresponding fluctuation results in certain regimes through steepest-descent analysis of exact multi-point formulas. Our calculations suggest that the conditional fluctuations near diagonal are described by Brownian bridges coupled through an additional shared Gaussian component.

We prove that every affine coordinate subspace of positive dimension and positive codimension fails to be of strong Khintchine type for convergence. Combining this with Huang's criterion for Khintchine-type affine subspaces gives a sufficient Diophantine condition for an affine coordinate subspace to be of Khintchine type but not of strong Khintchine type for convergence.

Vision-language-action (VLA) models map visual observations and language instructions to continuous robot actions. This task requires a transition from representations that describe the scene and instruction to representations that support action generation. Many continuous-action VLAs leave this transition implicit and supervise it mainly through the final action-prediction loss. We introduce PAIR, a framework that learns a shared perception-action representation between these two spaces. During training, a Masked Action Autoencoder encodes expert action chunks into horizon-aligned Action Latent Tokens. A Bridge Module extracts task-relevant features from the current visual-language representations. PAIR aligns these features with the Action Latent Tokens to form Bridge Tokens that preserve task information and capture the structure of expert actions. The Bridge Tokens are then projected into the action-token space and injected into the initial Action Tokens, providing an action-ready starting point for Action Expert refinement. At inference, the autoencoder is removed, and the Bridge Tokens are generated only from the current observation and instruction. Experiments on LIBERO, LIBERO-Plus, and CALVIN ABC-D show gains for the evaluated OpenVLA-OFT and VLA-Adapter models. On LIBERO-Plus, PAIR raises VLA-Adapter's success rate from 59.1% to 64.2%. On CALVIN, it increases VLA-Adapter's average completed sequence length from 4.42 to 4.53. Across seven real-world tasks, PAIR raises OpenVLA-OFT's success rate from 51.4% to 65.0%. Representation analyses show that Bridge Tokens retain task information while making continuous-action information accessible before Action Expert refinement. These results support a shared intermediate representation as a useful interface between perception and action in continuous-action VLAs.

Diffusion models can generate high-quality images, yet aligning their outputs with individual user preferences remains challenging. A key bottleneck is accurately modeling diverse user preferences from limited feedback. Existing approaches often rely on labor-intensive manual preference annotations or vision-language models (VLM) to extract preference information from user interaction histories, introducing substantial annotation or computational costs that limit scalability. We propose an approach that learns personalized reward models directly from users' historical image preference pairs. First, we use an autoencoder to compress hundreds of visual attributes into 50 attribute-anchored preference dimensions and train an evaluator to score images along these dimensions. We then represent each user's preferences as a linear combination of the shared dimension scores, estimating the user-specific weights by maximizing the likelihood of their observed pairwise preferences under the Bradley-Terry model. This formulation reduces per-user adaptation to optimizing a low-dimensional weight vector, simplifying optimization and enabling data-efficient personalization from sparse feedback. The learned personalized rewards guide image generation at inference time while keeping the diffusion model frozen. Experiments on real-user preference data show that our approach achieves approximately 77% held-out pairwise preference prediction accuracy and improves the alignment of generated images with individual user preferences.