Human-Computer Interaction
Papers
Fresh from arXiv, typeset properly. Highlighted words trade on the market.
Saturday 15 August
This paper presents the design and preliminary evaluation of KR A1, a humanoid-robot interaction prototype that integrates personalized voice synthesis, contextual dialogue processing, and emotion-aware response generation for potential mental-healthcare applications. The system architecture combines a locally hosted large language model (Qwen2.5-3B-Instruct, served via Ollama) for natural-language understanding and response generation, OpenVoice v2 for zero-shot speaker-adaptive speech synthesis, and a two-tier emotion-processing pipeline comprising a keyword heuristic and LLM-based inference. Face recognition (InsightFace) and vector memory retrieval (FAISS with sentence-transformer embeddings) enable user-personalized interaction. The experimental evaluation reported here is limited to the voice-cloning component: cloned voices were evaluated against reference recordings using eight acoustic similarity measures - mel-frequency cepstral coefficients, pitch, spectral centroid, spectral contrast, spectral rolloff, zero-crossing rate, formants, and speaker-embedding cosine similarity - supplemented by a same-speaker versus cross-speaker baseline comparison. All tests were conducted on a small, informal development sample; no statistical significance testing against a held-out baseline was performed. The emotion-recognition accuracy, end-to-end response latency, and therapeutic or clinical impact of the integrated system have not yet been formally evaluated and are identified as directions for future work. All testing was informal and self-conducted by the development team; no patients or external human participants were involved, and the system has not been evaluated or deployed in a clinical setting.
Computers and Society
OpenHamoye: A Multi-Scale Open Geospatial Intelligence Framework for Decision Support Across Nigeria
Decision-making in urban planning, infrastructure prioritization and commercial strategy benefits from granular spatial information. In Nigeria, official statistics are often published at coarse scales and with temporal lag. We present OpenHamoye, an open geospatial intelligence framework that integrates nine public remote-sensing and crowd-sourced datasets into reproducible, multi-scale proxy indicators for development conditions across Nigeria. OpenHamoye organizes the country into kilometre-scale analytical cells within a six-tier hierarchy: country, state, LGA, metropolitan area, corridor and cell. The framework constructs six composite indices: market potential, growth index, development momentum, infrastructure stress, commercial readiness, and housing pressure. Inputs include WorldPop, VIIRS Black Marble, ESA WorldCover, Google Open Buildings, Copernicus vegetation indices, OpenStreetMap, Overture Maps, GRID3, and geoBoundaries. We report a multi-scale assessment covering 37 states, 774 LGAs, and 8 metropolitan areas, using GRID3 facility points (46,146 health facilities and 204,513 schools) and Google Open Buildings data (479,088 building-positive cells). Results show large within-state disparities in the night-time-light proxy: the coefficient of variation across LGAs exceeds 4.5 in Borno and 3.8 in Katsina. The framework also flags high-scoring candidate cells in metros such as Benin City. OpenHamoye outputs are proxy-based screening signals for planning, infrastructure prioritization, resilience screening, market intelligence, and service-access context; they do not measure census population, GDP, income, poverty, welfare, firm revenue, transaction volume, commercial demand, infrastructure capacity, facility access, service quality, investment return, or official planning designation. Derived tables, index definitions, and reproducibility materials are released through GitHub.
Read the full abstract
Computers and Society · Cross-listed from physics.soc-ph
Economic Corridor Detection from Multi-Source Earth Observation and Commercial Geospatial Data Across Nigeria's Kilometre-Scale Grid
Identifying economic corridors at national scale is challenging where firm registries, transaction flows, and local economic-output statistics are incomplete or unavailable. This paper develops a reproducible multi-proxy screening framework for detecting economic-corridor candidates across Nigeria using a kilometre-scale analytical grid. The framework combines built-up land cover from ESA WorldCover, night-time light radiance from NASA VIIRS Black Marble, building-footprint density from Google Open Buildings, a static urban-intensity proxy derived from population, built-up share and night-time light level, and commercial point-of-interest density from Overture Places and OpenStreetMap. A dual proxy gate requiring both built-up evidence, defined as built-up fraction above 0.1, and night-time luminosity, defined as radiance above 1 nW cm-2 sr-1, identifies active corridor cells, which are then clustered into corridor objects and assigned interpretable typologies. The analytical frame contains 762,751 cells: national, state, and geopolitical-zone summaries use 749,191 canonical national-grid cells, while 13,560 configured metropolitan area-of-interest overlay cells are reserved for an eight-metro comparison. Applied to Nigeria, the framework identifies 11,243 active corridor cells on the national grid, representing 1.50% of cells, and 598 corridor objects spanning 10,682 retained cells and approximately 36.3 million people. Corridor activity concentrates around Lagos and other southern commercial geographies, while the method also surfaces emerging and constrained corridor candidates for review. All outputs are proxy-based screening signals for spatial planning, infrastructure prioritisation, and market-intelligence triage; they do not measure GDP, firm revenue, transaction volume, freight movement, infrastructure capacity, service adequacy, or official corridor designation.
Read the full abstract
Systems and Control
Holonic Graceful Transitions Across Centralized, Decentralized, Distributed, and Local Control in DER-rich Cyber-Power Distribution System
The increasing penetration of distributed energy resources (DERs) in distribution system necessitates adaptive coordination frameworks. These frameworks must remain optimal during normal operation and resilient under cyber physical disturbances. Existing coordinated control approaches are typically deployed as static architectures with limited ability to adapt when communication degrades, local instability emerges, and operating conditions become spatially heterogeneous. This work addresses the gap by proposing a DER service-independent edge autonomous holonic adaptive coordination framework. In this framework, each DER controller executes its own control actions and transitions among centralized, distributed, decentralized, and local autonomous coordination modes without always relying on static coordination commands from the grid operator. The framework preserves coordination continuity across modes by retaining local controller states while reconfiguring only the coordination topology, information exchange pattern, and fallback action associated with the active DER service. Graceful transitions are enabled through dwell timers, rate limiting, and safety overrides to prevent dynamic instability during mode changes. Volt VAR control is used as a representative distribution automation application to validate the proposed architecture in a cyber-physical Hardware-in-the-loop (HIL) testbed. The proposed approach is evaluated under diverse cyber-physical event-based scenarios, showing region-confined adaptation under localized disturbance, reduced coordination traffic during coordination switching, and node-confined mitigation under cyber attack through edge anomaly detection and neighbor-corroborated impact estimation.
Read the full abstract
Computation and Language
Hard Negatives Reveal What Easy Negatives Hide: Cross-Lingual Harmfulness Representations Degrade with Resource Tier Under Hard Negatives
Safety alignment in large language models is trained primarily in English, and recent work reports that the underlying harmfulness representation survives translation: English-trained probes separate harmful from harmless prompts almost as well in low-resource languages as in English. This has been taken as evidence that cross-lingual refusal failures mainly reflect calibration rather than representation quality. We show that this conclusion depends on the choice of negative examples. Across nine languages spanning three resource tiers, we replicate near-perfect transfer (AUROC > 0.98) when harmless prompts come from an unrelated distribution (easy negatives). With XSTest contrast prompts, which are benign but surface-similar to harmful requests (hard negatives), transfer collapses in low-resource languages while remaining largely stable in high-resource languages. On Qwen2.5-7B-Instruct, mean AUROC drop increases from 0.003 in English to 0.017 in high-resource, 0.042 in mid-resource, and 0.276 in low-resource languages. The pattern replicates on Aya Expanse. Back-translation chrF controls and a matched-chrF comparison across three languages reduce the likelihood that translation quality explains the effect. The collapse remains after controlling for chrF (partial r = 0.70, p = 0.03). Tokenizer fertility correlates with the collapse and explains part of the resource-tier effect, but not all of it. The results show that easy-negative transfer can coexist with substantial degradation under hard negatives. Easy-negative evaluation alone therefore cannot establish that the harmfulness representation survives translation.
Read the full abstract
Artificial Intelligence
Reporting Under Pressure: Separating Factual and Tonal Sycophancy in LLM Statistical Analysis
Large language models are increasingly asked to analyze data and report what the results mean, a task distinct from the belief- or preference-alignment settings studied in most sycophancy research. We test whether editorial framing in the prompt, ranging from a neutral request to an explicit instruction to search exhaustively for reasons to discredit or to support a finding, changes not just the tone but the substance of a model's report. Across a 4 x 4 factorial design crossing four framing conditions with four ground-truth data patterns (a genuine effect, a confound that mimics an effect but fails a robustness check, a well-powered null, and an underpowered null), we collect 480 responses and score each along two independent dimensions: whether its factual claim about the data diverged from the correct interpretation, and whether only its tone diverged while the claim stayed correct. Factual misrepresentation is concentrated in two cells: brutally critical framing applied to a genuine effect, where the model talks itself into unwarranted skepticism (97% of responses), and significance-seeking framing applied to an underpowered null, where the model overstates confidence in a null conclusion the data cannot support (100% of responses). Tone shifts far more broadly than factual content does, with critical framing producing a defensive, hedge-heavy register across every data pattern regardless of what the data show, while significance-seeking framing shifts tone only where the data leave genuine ambiguity. A confound present in the data itself blocks both kinds of shift almost entirely under every framing condition tested. These results indicate that the risk of framing-induced distortion in LLM-assisted data analysis is neither uniform across framings nor uniform across data patterns, and that a model can hold a correct conclusion in place while its tone shifts substantially around it.
Read the full abstract
Computer Vision
AWM-VLA: AlignedWorld Modeling for Efficient and Explainable Vision-Language-Action Policies
Vision-language-action (VLA) models have become a powerful paradigm for generalist robotic manipulation, yet they are often reactive: the policy maps the current observation directly to an action chunk without reasoning about the long-term consequences of its decisions. Prior attempts to endow policies with world models either reconstruct future frames in pixel space—expensive and dominated by task-irrelevant detail—or decouple the world model from the policy, weakening control. We present AWM-VLA, a unified framework that embeds aligned world modeling directly inside a diffusion-transformer policy. Following the Future Latent REpresentation Alignment (FLARE) principle, we add learnable future tokens whose intermediate activations are aligned with vision-language embeddings of future observations, enabling the policy to anticipate long-term consequences while generating actions. We extend this paradigm in two ways. First, we introduce an object-centric decoupled alignment objective that predicts future object-level semantics alongside the global future embedding, improving both interpretability and multi-instruction generalization. Second, we balance the global and object-centric alignment terms against the action flow-matching loss through a principled weighting, yielding a controllable accuracy–interpretability trade-off. On RoboCasa and humanoid tabletop manipulation benchmarks, AWM-VLA outperforms prior VLA and world-model baselines by up to 21% in success rate, improves generalization to novel objects and instructions, and produces object-centric rationales that are preferred by human raters in 83 of cases. Our approach adds only a few learnable tokens to the policy and is compatible with any diffusion or flow-matching policy, making aligned world modeling an inexpensive, broadly applicable component of generalist manipulation.
Read the full abstract
Systems and Control
Human-Centricity in Industry 5.0: A Survey of Worker Sensing, Adaptive Operations, and Human-in-the-Loop Systems
Industry 5.0 (I5.0) repositions manufacturing around a human-centric vision in which cyber-physical systems must adapt to the worker rather than the other way around. Despite significant advances in worker state monitoring technologies, including markerless computer vision, wearable physiological sensors, and machine-learning-based fatigue and cognitive load estimation, a critical integration gap persists: these monitoring capabilities rarely inform real-time operational decisions, and workers almost never participate in the adaptive processes that affect their tasks and workloads. This paper presents a structured survey of the state-of-the-art across three interconnected dimensions: (1) worker state monitoring in industrial settings, encompassing physical fatigue, cognitive load, and ergonomic risk assessment; (2) human factors integration in Operations Management (OM) models, examining how scheduling, task allocation, and production planning incorporate worker-related objectives; and (3) Human-in-the-Loop (HITL) architectures, analyzing the extent to which worker feedback closes the adaptive decision loop. The survey reveals that while each dimension has matured independently, the integration across them remains largely absent. Based on the reviewed literature, no publicly documented and validated system connects real-time sensor-driven worker state estimation with adaptive OM decisions and worker feedback in a unified closed-loop architecture. This gap is identified as the central challenge for realizing the I5.0 vision, and a five-direction research agenda is proposed.
Read the full abstract
Networking
ST-NDT: A Topological Framework for Reducing Communication Overhead in Network Digital Twins
Next-generation networks (NGN) are becoming increasingly complex, especially with the increasing size of topologies. To manage this complexity requires the adoption of frameworks that enable real-time monitoring, optimization, and “what-if” case scenario analysis. Network Digital Twins (NDTs) have emerged as a key enabler technology for supporting these capabilities because of their ability to provide digital replicas of physical networks operations, enabling scenario testing without interfering with the live network. Although NDTs are considered a key enabler technology, maintaining continuous synchronization with the Physical Twin (PT) introduces considerable communication overhead and excessive bandwidth utilization, limiting sustainability in resource-constrained networks. This paper proposes a topological signal processing framework for sparse network monitoring in NDTs. We represent network edge flows as signals defined on a cell complex, and exploit the Hodge spectral structure of the graph to identify a minimal set of maximally informative sensor edges, thereby reducing the measurement overhead between the physical network and its digital twin replica. Results show that the proposed Sparse Topological Network Digital Twin (ST-NDT) framework consistently outperforms all four baseline frameworks across most of the edge monitoring budgets. For example, at a 20% monitoring budget (48 of 243 edges monitored), ST-NDT achieves improvements of 10.85%, 11.24%, 7.36%, and 6.15% over degree-based, core-number, betweenness centrality, and random-selection frameworks, respectively.
Read the full abstract
Computation and Language
TSS: Target-Side Sparsification for Speculative Decoding in Domain-Specific Large Language Models
Speculative decoding accelerates large language model inference through collaboration between a lightweight draft model and a target verifier. Existing methods mainly improve the draft side, while the target model is typically kept dense and unchanged. We show that, under domain-specific inference, full-depth target verification is not always the optimal choice. Counter-intuitively, skipping selected target layers can reduce verification cost while simultaneously increasing draft acceptance and preserving, or even improving, downstream task performance. Based on this observation, we propose TSS, a target-side sparsification framework for speculative decoding. TSS employs an acceptance- and metric-aware breadth search to explore multi-layer skip configurations without imposing a fixed priority between the two objectives. The selected configurations are stored in a domain-to-configuration mapping and applied by a lightweight skip controller, allowing one complete target model to support multiple sparse verification paths without retraining or permanent parameter pruning. Experiments on Spec-Bench across multiple domains, model scales, and speculative decoding methods show consistent improvements in draft acceptance and downstream task performance. In Translation setting, TSS increases the average accept length from 2.70 to 4.53 (+67.8%), improves BLEU from 0.131 to 0.237 (+80.9%), and raises end-to-end throughput from 75.6 to 127.3 tokens/s, corresponding to a 1.68X speedup.
Read the full abstract
Computer Vision
SPEANet: Structural Prior Enhanced Attention Network for Parameter-Efficient Remote Sensing Object Detection
Remote sensing object detection (RSOD) requires compact backbones capable of preserving weak geometric cues under extreme scale variation and background clutter. Fixed structural operators provide complementary contour and frequency responses without introducing learnable operator coefficients. However, directly injecting these responses can amplify content-irrelevant textures, while applying a uniform operator design across the hierarchy may be poorly matched to stage-specific representation requirements. We propose the Structural Prior Enhanced Attention Network (SPEANet), a parameter-efficient RSOD backbone that integrates fixed operators through stage-specific prior extraction and context-conditioned response modulation. SPEANet assigns smoothed contour and multi-order directional modeling to shallow, high-resolution features, while employing a compact approximation-detail interaction mechanism in deeper stages. Learned spatial gates regulate the resulting prior responses before residual fusion. Experiments on five benchmarks, together with evaluations across seven detection frameworks on DOTA-v1.0, achieve a favorable accuracy-parameter trade-off. With Oriented R-CNN, SPEANet achieves 78.55% mAP on DOTA-v1.0, 72.24% mAP on DOTA-v1.5, and 67.30% mAP on DIOR-R using 23.0M total parameters, including a 5.97M-parameter backbone.
Read the full abstract
Computational Engineering
Compositional Aeroelastic Operators for Morphing Flexible Multibody Aircraft: A Geometric Framework with Structural Verification
Morphing flexible multibody aircraft require structural strain, aerodynamic geometry, surface velocity, and generalized loading to remain compatible as joints and flexible components change configuration. A compositional formulation is developed around an assumed material attachment between each lifting surface and a geometrically exact beam. Separating the component root pose from the section field shows that the body strain and elastic potential of a component depend on its own elastic coordinates, while upstream motion enters kinetic terms and external-load pullbacks. At element level, an exact relative logarithm $d$ supplies strain and potential energy, whereas a reference-anchored section coordinate $\sigma$ supplies deformed section geometry. Finite-order expansions retain the finite reference geometry exactly and truncate only endpoint perturbations. The attachment map then generates surface points, tangents, normals, velocities, and force Jacobians from common section kinematics. Euler–Poincare beam balance, moving-surface potential-flow relations, graph cotangent assembly, and the associated semidiscrete power identity are stated in a common twist–wrench convention. Collocation, pressure, equivalent-load, and structural-station sites are distinguished to expose their approximation errors. Verification gives the expected $N_d+1$ convergence order for degree-$N_d$ relative-log expansions. In a geometrically nonlinear cantilever comparison, a cubic static-manifold correction reduces mean full-record displacement error from $0.479$ to $0.255$ over four completed load cases. These results provide structural and interface-level evidence rather than validation of a complete aircraft aeroelastic prediction.
Read the full abstract
Artificial Intelligence
StateSight: Benchmarking Latent Spatial-State Reconstruction in Vision-Language Models
Vision-language models are increasingly used for multimodal question answering, yet their ability to reconstruct latent spatial structure from a single image remains difficult to isolate. Broad benchmarks often combine perception, optical character recognition, domain knowledge, linguistic priors, and reasoning in the same evaluation. We introduce StateSight, a procedurally generated benchmark for cube-net opposite-face reasoning, occluded cube-tower counting, and 4-neighbor connected-component counting. Each task family contains 300 single-image prompts with deterministic oracle labels and exact-match scoring. OpenAI GPT-5.5, using the API model identifier gpt-5.5, achieved 59.3%, 33.3%, and 28.3% accuracy across the three tasks, while Claude Sonnet 5 achieved 53.3%, 18.7%, and 7.3%. All final direct runs had zero format errors. A 30-participant human baseline on 60 items exceeded both models on every task, with mean accuracies of 80.8%, 68.8%, and 64.3%. Visible-derivation analysis identified recurring errors in image-state reconstruction and reasoning procedure. We also introduce StateSight-Steps, a companion dataset of 900 interleaved image-text examples and 3,600 deterministic intermediate visual states. The results show that format-valid responses can mask failures to recover the spatial structure required for verifiable visual inference.
Read the full abstract
Computation and Language
Are LLMs Safe Beyond Text: Do Emojis Expose Gaps in Safety Evaluation
Safety evaluations of large language models (LLMs) predominantly rely on text-based adversarial prompts, potentially overlooking vulnerabilities arising from alternative input representations. This work examines emoji-augmented prompts as a test case for this gap, evaluating 50 prompts across four open-source LLMs (Mistral 7B, Qwen 2 7B, Gemma 2 9B, Llama 3 8B). Results show substantial variation in robustness: Gemma 2 9B and Mistral 7B exhibit non-zero success rates (10%), Llama 3 8B 6%, while Qwen 2 7B shows complete resistance (0% success rate). A chi-square test ($\chi^2 = 32.94, p < 0.001$) confirms significant differences in outcome distributions. These findings indicate that robustness is sensitive to input representation, and that evaluations restricted to standard text prompts may underrepresent model vulnerabilities.
Read the full abstract
Data Structures and Algorithms
A Deterministic Constant-Competitive Algorithm for Dynamic Mixture-of-Experts Serving
Dynamic Mixture-of-Experts Serving allocates k replica GPUs among m experts as workloads change. At each round, the online algorithm sees the current workload, chooses integral replica counts, and pays bottleneck service cost plus replica movement. It does not know future workloads. Huang, Lou, and Xiao gave an O(sqrt(log k))-competitive randomized algorithm for this problem. We prove a deterministic O(1)-competitive algorithm. For every number of experts and every k>=1, the algorithm satisfies ALG_det <= 10 C_PB OPT + (5 C_PB + 8) k + 16, where C_PB is the absolute constant from Chasing Positive Bodies at resource augmentation one and covering sparsity two. Consequently, CR_det(k)<=10 C_PB for every k>=1, so CR_det(k)=Theta(1). The multiplicative factor does not depend on the number of experts, replica budget, horizon, or workload values. Thus randomization is not needed for the asymptotic guarantee. The proof has two layers. A finite tangent envelope, summable positive resets, and a nonexpansive balanced projection reduce reciprocal-max service costs to a deterministic exact-budget fractional path. A new deterministic rounding theorem converts every such path to integral allocations with service distortion three and movement bounded by the fractional movement plus 6k. The complete reduction, rounding theorem, causal composition, and quantified main theorem are machine-checked in Lean 4 relative to the positive-body result as the sole scientific source premise. The theorem concerns the allocation model above. It does not include network topology, shared-edge congestion, or routing decisions.
Read the full abstract
Computation and Language
Language models suffer from a curse of ambiguity
Large language models increasingly rely on sampling as a driver of their own improvement, making the fidelity of their learned distributions more critical than ever. Yet, not all distributions are equally easy to learn. In this work, we identify a curse of ambiguity: in large language models, and more broadly in all neural networks that produce discrete probability distributions, the more ambiguous a next-token distribution is, the harder it is to learn accurately. Through an extensive theoretical analysis, we trace this curse to architectural and learning roots. More ambiguous distributions require more capacity to be stored, larger embeddings to be represented, more steps to be fitted, and amplify token-sampling noise. We validate these findings on synthetic tasks with controlled ground truth and observe the same signatures in language models trained on real data. Our results provide a new perspective on the statistical capabilities of large language models and a practical framework for when to trust their output distribution.
Read the full abstract
Machine Learning
Detecting Money Laundering in Rwandan Mobile Money: A Machine Learning Framework
Mobile money has widened financial access across Sub-Saharan Africa and enlarged the surface for money-laundering and terrorism-financing (ML/TF) activity in ecosystems dominated by high-volume, low-value transactions. Rwanda is a case in point: several million active mobile-money users, telecom-led wallets on the MTN and Airtel networks, and a Financial Intelligence Centre (FIC) supervising transaction streams whose scale exceeds static rule-based monitoring. This paper develops and evaluates a transaction-monitoring framework aligned to the Rwandan AML/CFT regime under (i) extreme class imbalance (~0.1% prevalence), (ii) scarce and delayed labels, and (iii) bounded investigator capacity. Using SAML-D, a synthetic dataset of 9,504,852 transactions with 17 laundering typologies, we engineer account-centric behavioural features (rolling velocity, net-flow directionality, counterparty diversity, burstiness) and benchmark supervised classifiers (Logistic Regression, Random Forest, LightGBM), unsupervised anomaly detectors (Isolation Forest, Local Outlier Factor), a dense autoencoder, and a late-fusion meta-learner. Evaluation is operational: PR-AUC, recall at a calibrated ~90%-precision point, recall at top-K%, and alerts per 10,000. On the chronologically held-out test period, LightGBM attains PR-AUC = 0.0469, capturing 64 laundering cases at precision ~0.89 with 0.51 alerts per 10,000; the fusion stacker reaches PR-AUC = 0.0477 at precision ~0.91 and 0.46 alerts per 10,000, recovering 59 true positives. We map score bands to Rwanda-relevant analyst workflows and STR/SAR escalation, and outline a staged path from synthetic prototyping to real-data validation with the National Bank of Rwanda and FIC. The contribution is operational: a governance-aware pipeline and evaluation protocol calibrated to the constraints of an African mobile-money regulator, not a new algorithm.
Read the full abstract
Robotics
GUIDER: Evaluating Goal-Free Human Intent Inference for Teleoperated Manipulation on Real-Robot Data
This paper presents an evaluation of a goal-free probabilistic framework for human intent inference during robotic manipulation. We deploy the Global User Intent Dual-phase Estimation for Robots (GUIDER) on data collected from a robotic arm to test the manipulation phase across various assistance scenarios, including making tea and fetching medicine. To support operation, we add online probability updates, workspace limits, support-plane filtering, and a grasping mode that prioritizes feasible grasp regions, all of which are tested on the recorded data while preserving its original temporal conditions. Across 20 manipulation steps in three scenarios, GUIDER estimated human intent within the correct grasp-candidate set in all cases and achieved a time to confident prediction of 3.7 s, a remaining time before first grasp of 49.6 s, a prediction stability of 96.4%, and a runtime of 4.857/4.474 s (mean/median) per perceptual phase of intent.
Read the full abstract
Artificial Intelligence
Measuring Reward Hacking and Reasoning-Answer Decoupling Under Position-Confounded Optimization
When a reward is correct on every training example yet consistent with more than one goal, a model can acquire an unintended one, a failure known as goal misgeneralization. Endpoint accuracy on the training distribution cannot tell the two apart, because solving the task and exploiting a surface feature can satisfy the reward equally well. We treat this as a measurement problem: what does a benchmark score measure once a model has been optimized against a correct but confounded signal? We train language models with GRPO on multiple-choice math problems where the correct answer is always option A, then evaluate on an unseen test set with unbiased answer positions. Across Qwen2.5, Llama 3.x and Gemma-3 models, biased training often drives option-A rates above 0.90 in smaller models and collapses unbiased accuracy toward chance, so accuracy stops measuring math ability and instead measures an answer-position policy. We further find reasoning-answer decoupling: capable models generate reasoning that reaches the correct numeric answer while still selecting A. We track this with numeric extraction and an LLM judge (GPT-4.1-mini; Qwen2.5-3B decoupling rate is about 0.66). The broken construct generalizes beyond the training domain: biased models inflate A-rates on out-of-domain MMLU and value-laden prompts. Continued training on unbiased data reverses the in-domain shift unevenly and only partially reverses the out-of-domain one, so a model can appear restored on its training distribution while remaining biased on unseen inputs. Reasoning-answer decoupling rate, together with answer distributions and out-of-domain behavior, separates capability loss from a learned, transferable shortcut.
Read the full abstract
Computation and Language
Semantic Space of Parts of Speech
Parts of speech categorization is understood in the European linguistic tradition as crisp categorization, which is also reflected in corpus linguistics, where each disambiguated token is assigned exactly one POS. However, the assigned categories are largely determined by arbitrary decisions distilled into annotation manuals. Since some words stand between parts of speech in their semantics or typical syntax, and some parts of speech are closer to each other than others, POS categorization seems inherently fuzzy. We analyze this fuzziness using word2vec embeddings, training a neural network to reduce their high dimensionality to three dimensions relevant for determining parts of speech. This creates a three-dimensional space onto which we map several thousand words, revealing which are prototypical and which lie on the boundaries, and visualizing relationships between parts of speech. The study uses Universal Dependencies POS tags for French, Czech, Finnish, Russian, and English.
Read the full abstract
Human-Computer Interaction
Everything Is a VisionBlock: Conversational Authoring over Git-Versioned Content for Spatial Computing
Spatial applications compile their content into shipped binaries, so every change costs a build-and-redeploy cycle. We present the VisionBlock system, which splits an application into an engine – a generic binary with a fixed set of capabilities (render panels, volumes, and immersive scenes; fetch data; run gestures) – and themes: complete applications expressed as trees of VisionBlocks, units of declarative content the engine renders. Themes are data: creating, changing, or publishing one never touches the binary. Authoring is a chat – each turn produces a VisionBlock's next version – and versioning is plain git. The model is five-dimensional: dimensions 1-3 are space (panel, volume, room); dimension 4 is time (git history – revert to roll back, branch to try variants); dimension 5 is the principal (the per-user domain: the same path resolves differently per person). The engine renders one point, (x, y, z, version, principal). One consequence follows per non-spatial axis: iteration collapses to chat turns and reverts; ownership and permission are properties of content; and together they make applications items – grantable, forkable, sellable subtrees, an economy of apps inside one binary. A blockchain explorer, a document reader, an immersive showroom all run on the same engine; none requires a deploy to change. This paper presents the design; a production implementation is underway, and a subsequent version will report implementation and evaluation.
Read the full abstract
Robotics
Accelerating Optimization over Graphs of Convex Sets via Neural Network Approximations
Motion planning problems such as collision-free navigation and contact-rich manipulation can be naturally formulated as optimization problems that couple discrete decisions with continuous trajectories. The Graphs of Convex Sets (GCS) framework offers a practical solution to these problems. It represents discrete decisions as nodes of a graph and encodes continuous trajectories in the edges connecting them. However, the resulting optimization subproblems can become computationally prohibitive for online replanning. In this work, we propose a learning-based strategy to mitigate this limitation. Specifically, we replace the costly convex relaxation step required by nominal GCS with a single forward pass through a Graph Attention Network that predicts a set of highly probable candidate paths through the graph. A lightweight ranking network then orders these candidates by their estimated trajectory cost. Evaluating them in this order, we terminate our search early while still recovering a near-optimal motion plan. We validate the resulting pipeline across diverse robotic tasks, including collision-free motion planning for a 3D quadrotor and a 7-DoF manipulator, and planning through contact for planar pushing. Across both convex and non-convex cost and constraint settings, our approach yields up to two orders of magnitude speedup over nominal GCS while maintaining a 100% success rate, at the cost of some suboptimality in the recovered solutions. Code implementations and video demonstrations can be found at https://neural-gcs.github.io/.
Read the full abstract
Databases
NeuRoute: Logit-Guided Neural Routing for Billion-Scale Vector Search with Sub-Hour Index Construction
Building approximate nearest neighbor (ANN) indexes at billion scale is often dominated by expensive global clustering or graph construction, making time-to-index a first-order systems concern. We present NeuRoute, a learned hashing index that turns short binary codes into an effective routing primitive for large-scale vector search. NeuRoute trains a lightweight neural network encoder with a selective similarity-preserving objective to produce well-balanced binary addresses. During construction, NeuRoute organizes vectors into buckets by their codes and performs bucket-local clustering in the encoder's low-dimensional space to form centroids. At query time, NeuRoute exploits the encoder logits as an uncertainty signal: it uses deviation-to-threshold scores to prioritize uncertain-bit perturbations for query-adaptive multi-bucket probing, scores bucket-local centroids by their distances to the query to form a compact candidate cluster set, and applies centroid-stage gating with heap-quality-driven early stopping to prune low-value clusters before exact refinement. On billion-scale benchmarks, NeuRoute achieves strong accuracy-throughput trade-offs with fast index construction: on BigANN-1B it reaches $90.3\%$ Recall@10 at 2,414 QPS and is $1.7\times$ faster than OPQ+IVF-PQ (refine) at comparable accuracy, while completing end-to-end training+construction in under an hour on both BigANN-1B and Deep1B-1B. These results show that logit-guided neural routing can make hashing competitive as a lightweight ANN indexing framework at billion scale. Source code and artifacts are available at https://github.com/XingqiaoWang/NeuRoute.
Read the full abstract
Robotics
MM-BEV: Enhancing Timeliness by Computing Where and When it Matters
Multimodal bird's-eye-view (BEV) perception combines LiDAR depth accuracy with dense camera semantics, but its high computational cost and imperfect sensing conditions make real-time deployment challenging. Existing methods largely compress individual detectors and overlook three opportunities: structured sparsity within camera and LiDAR inputs, timing misalignment between modalities, and the fact that many detected objects do not affect the planner's immediate action. We present MM-BEV, a real-time multimodal BEV system guided by a simple principle: compute where and when it matters. MM-BEV divides perception into mandatory work for safety-critical objects within braking distance of the ego vehicle and with short time-to-collision (TTC), and optional work for less urgent regions. It prioritizes mandatory work and reduces or sheds optional work under tight compute budgets. MM-BEV integrates four mechanisms: (1) a criticality-ranked temporal ROI selector based on motion-extrapolated detections from prior frames; (2) sparse, ROI-aware feature extraction using shared-shape camera crops at context-adaptive resolution and ROI-aware LiDAR voxelization; (3) a latency-aware coordinator that adapts LiDAR sweeps, image resolution, and keyframes according to scene dynamics and TTC; and (4) an asynchronous scheduler that decouples sensing from inference and skips stale frames. On nuScenes, MM-BEV reduces inference latency by 1.96x and end-to-end latency by 2.93x, with no loss in geometry-critical recall and only a 0.2 percentage-point drop in safety-critical recall. On a Clearpath Husky A300 equipped with an Ouster-128 LiDAR, BEV cameras, and a Jetson AGX Orin, MM-BEV further reduces mean latency by 2.11x, demonstrating its potential for real-world autonomous systems.
Read the full abstract
Artificial Intelligence
OTel: Building Domain-Specialized Telecom LLM Foundations for Intelligent Networks
Frontier AI models have advanced rapidly, but they still struggle with telecom-specific tasks. We present Open Telco (OTel), an open telecom AI resource with derived datasets for retrieval, reranking, instruction tuning, and safety/abstention, plus 30 full-parameter post-trained baselines across embedding, reranking, and language models. The community has already engaged substantially with the resource: as of May 3, 2026, the released models have been downloaded over 16 million times, and the project has received 157+ pieces of media coverage worldwide. Building on prior open telecom datasets and benchmarks, OTel provides documented telecom data sources, held-out evaluation partitions, trained embedding models, rerankers, context-grounded LLMs, and safety/abstention data in one unified resource. OTel post-training improves performance across all three model families: embedding retrieval reaches 93.5% NDCG@10, reranking reaches 0.952 MRR@10, and language-model correctness reaches 88.2%. We release OTel as a reproducible starting point and invite the community to expand the data, improve embedding and reranking models, and build stronger context-grounded telecom LLMs.
Read the full abstract
Artificial Intelligence
Does the Proof Prove It That Way? Faithful Formalization of Elements Proofs
In formal verification, both the autoformalization of statements and automated proof search have been studied extensively. While automated proof search can produce a formal proof that compiles, the generated proof does not necessarily reflect how the natural-language argument arrives at its conclusion–a property we refer to as faithfulness. With faithfully formalized proofs, one can check the reasoning behind a human- or AI-written argument, and assist mathematicians in formalizing their proof sketches. However, it is particularly challenging due to misalignment of formal proof tactics and natural language reasoning. In this work, we rigorously describe a set of five necessary conditions a faithful formal proof must satisfy, and introduce Pistis, an agentic, oracle-guided proof search that produces formal Lean proofs that satisfy them. At its core is a novel faithfulness-preserving divide-and-conquer search, which we name OrderDecompose, that tracks citation dependencies and blocks unfaithful shortcuts, paired with a refutation search, that surfaces gaps and errors in the natural language proof source. OrderDecompose completes proofs that baselines cannot close even within a 12-hour budget, and its artifacts compile over 33$\times$ as fast as prior work's. We apply Pistis on the first three books of Euclid's Elements, producing high-quality artifacts containing faithful formal proofs. Under a blinded human study and an LLM-as-a-judge protocol on rigorous rubrics, Pistis-generated proofs are favored over prior works–2.89$\times$ and 5.2$\times$ as often by human reviewers and the LLM judge, respectively. It further uncovers gaps in Euclid's proofs and their translation, and can accept or refute natural language proofs written by humans or AI, demonstrating that faithful formalization is useful as a proof-checking tool.
Read the full abstract
Machine Learning
SAGA: Structure-Attended Generative Action Embedding Model that encodes Multi-Surface User Action Sequences
Prior embedding models for sequential recommendation typically operate within a homogeneous action space, limiting their ability to capture cross-surface behavioral signals spanning distinct behavioral domains. We present SAGA, a generative action embedding model that encodes multi-surface user interaction sequences across a Financial Service organization's ecosystems, from checkout, peer-to-peer (P2P) transactions, in-app engagement, email to account actions, into a unified user representation for downstream recommendation tasks. Central to SAGA is a per-field tokenization schema that decomposes each action event into multiple field-level tokens (e.g. product, interaction, surface), enabling field-level attention and per-field training objectives that fused single-token approaches cannot support. Through an offline ablation study on loss formulation, tokenization granularity and training data scope, we isolate the contribution of each design choice. A downstream model integrated with SAGA-generated user embeddings delivers the strongest overall click and conversion lift across diverse downstream touchpoints, compared to all ablated and alternative architectures.
Read the full abstract
Computation and Language
Gated Against One Model, Open to the Next: Option-Only Solvability in Legal Multiple-Choice Benchmarks
Multiple-choice benchmarks are graded on whether a model picks the right option, not on whether it needed the question. Measuring that gap takes care: a model answering A to most items scores above chance wherever the key sits at A, and reads as recognition when it is not. We measure it on UA-JudgeExam: 11,990 four-option items with official keys, published by Ukraine's Higher Qualification Commission of Judges. Shown the options and no question, Claude Haiku 4.5 scores 0.383 against chance, and the leak is concentrated: 11.8% of items are answered blind on all eight option orders, against 0.2 items expected by chance. It is not quotation: search over 280,059 editions of Ukrainian legislation recovers 0.128. Gating those out retains 8,128 items, on which the gating model itself now scores 0.204, while GPT-5.6, which took no part in the selection, still answers 0.515 of them with the question hidden. Scoring eighteen held-out models from eight vendors on the whole set and subtracting each one's position habit, four keep a sizeable excess: GPT-5.6 (+0.265), Gemini 3.1 Pro (+0.240), GPT-5.4 (+0.105), Sonnet 4.6 (+0.081); the rest stay within 0.031, and probes find no recall of the public bank behind it. Without the correction Llama 3.1 8B (0.292 blind, A on 92% of items) would read as a leaker. The gate does select something real: on the items it rejected, eleven of twelve models score 0.518-0.789, every interval clear of what the same model scores on the items it kept. But that signal is one model's, and filtering on it does not transfer upward. Rewriting distractors instead overshoots to 0.168, below chance and as exploitable. The same probe on LEXam returns chance: every option there points into the stem, none longer than 33 characters. Item format decides whether the problem can arise; capability decides how much is extracted. We release the corpus, predictions and harness.
Read the full abstract
Information Theory
New Approximations of Non-Separable MIMO Channels by Separable Channels for Accurate Ergodic Capacity Analysis
In recent years, owing to the high accuracy in characterizing non-separable channels prevalent in next-generation wireless applications, the classical Weichselberger channel model has gained widespread adoption in multiple-input multiple-output (MIMO) systems. However, its non-separable structure also introduces severe analytical complexity, leading to a lack of tractable mathematical frameworks in the literature and thus raises an urgent need for further research. To address the aforementioned analytical complexity, we first derive the nearest separable (double-correlated Rayleigh) fading model to the Weichselberger model under the Kullback-Leibler divergence (KLD), a problem equivalent to rank-1 nonnegative matrix factorization under the Itakura-Saito (IS) distance criterion. The results of our asymptotic analysis in the high-SNR regime reveal that the KLD-enabled approximation achieves a tighter capacity estimate than the conventional Kronecker model, especially in sparse and non-regular scattering environments. Yet, a key limitation of the KLD-enabled model is its tendency to mischaracterize the channel capacity in the low-SNR regime due to its inability to preserve total channel power. As a more robust alternative, we introduce a novel moment matching method (MMM) aimed at mapping the exact channel statistics to those of a Wishart distribution. Both the KLD-enabled and MMM-enabled separable channel directly enable the use of exact closed-form expressions for the ergodic capacity. Numerical results demonstrate that the MMM-enabled model consistently improves upon the capacity accuracy of the conventional Kronecker model across all SNR regimes.
Read the full abstract
Computer Vision
NumerosityVLM: A Cognitively Inspired Benchmark for Interpreting Numerosity Representations in Vision-Language Models
Vision-language models (VLMs) achieve strong performance on high-level multimodal tasks, yet numerosity perception, a cognitive ability that emerges in human infants before language acquisition, remains poorly understood in current models, as existing counting benchmarks entangle numerosity with correlated visual factors. We introduce a cognitively inspired diagnostic benchmark, NumerosityVLM, comprising 10,800 synthetic images across six controlled conditions. The benchmark orthogonally manipulates object size, spatial arrangement, and numerosity, while progressively ablating texture, shape, and color. Evaluating seven VLMs in a zero-shot setting, multi-factor analysis reveals that model architecture explains the largest proportion of performance variance (partial $\omega^{2}=0.325$), far exceeding visual conditions. Layer-wise probing further shows that linearly separable numerosity signals consistently emerge at early stages of the vision encoder, while performance differences across evaluated models are primarily associated with the language model component. Code and data are publicly available at https://github.com/fuy3/NumerosityVLM-Benchmark, and https://huggingface.co/datasets/fuy3/NumerosityVLM.