Papers

This letter addresses budgeted stealthy false-data-injection (FDI) scheduling against remote state estimation, where a resource-constrained adversary may corrupt at most a fraction $\bar{\Gamma}$ of transmissions. Existing event-triggered schedulers invert a Gaussian innovation tail to set the firing threshold and certify stealth by covariance matching; both are exact only under Gaussianity, which real cyber-physical residuals routinely violate. We propose a distribution-free scheduler pairing the worst-case FDI action with a split-conformal calibrated trigger, requiring neither the plant matrices nor any distributional model. We establish exact pathwise stealth against every magnitude-measurable detector, for any firing rule and innovation law; a finite-sample distribution-free bound on the mean firing rate, with almost-sure budget attainment under stationarity and ergodicity; and a steady-state degradation identity linear in a single scalar energy capture $\psi$, maximized by the same order statistic that delivers the budget guarantee. A conditional sign-symmetry condition delimits when the certificate extends to sign-sensitive detectors, the residual exposure being governed by a fourth cumulant that vanishes under Gaussian noise. Monte-Carlo studies and a heavy-duty truck CAN record confirm the bounds and quantify what the Gaussian assumption costs outside its regime.

Security Operations Centers (SOCs) often rely upon on-the-job training focusing on the specific tools and procedures a SOC utilizes. The complexity in tooling can overshadow the underlying reasoning process, hindering an analyst's ability to learn investigative skills. We formulate the concept of "non-perishable knowledge" which corresponds to investigative thinking skills independent of tools. We developed a scenario-driven training method to understand whether such non-perishable knowledge can be imparted through specially designed incident scenarios, where trainees are presented with and solve investigative challenges in a tool-agnostic manner. We designed two such scenarios based on real-world incidents. Human subjects were recruited from a university's student body for ethnographic study to understand how this training method is received by the subjects and how they perform on such tasks. We collected data from 20 hours of documented training with 25 trainees spread across five sessions. Using grounded-theory, we analyzed the data and uncovered factors that inhibit or promote learning of the investigative thinking skills. Our research combines scenario-based training, ethnographic research, and technical analysis to examine how to best train students in the reasoning skills that industry deems vital to SOCs.

The Viterbi algorithm has been previously used to perform forced alignment of audio to text to mine training data from online resources. However, many existing implementations have quadratic time and space complexity, scaling poorly to long input sequences. We propose two optimizations to address this issue. First, we apply the Hirschberg algorithm to perform the alignment in place using linear memory. Second, we model the alignment between speech and text as a constrained random walk, allowing us to prune the search space with arbitrary confidence while accounting for transcription errors. The Hirschberg optimization reduces memory usage from 140 GB to 5 MB for three-hour inputs while producing identical alignments in one-third the time of torchaudio when both run on a CPU. We achieve an additional 2x speedup with pruning on inputs longer than 20 minutes while preserving alignment accuracy in more than 98% of tested cases.

Let $q_m=2m/(m+1)$ and put \[ \beta_0=\frac{3}{2}+\frac{1}{\log 2}=2.9426950408\ldots, \] where $\log$ is the natural logarithm. We give a proof scheme showing that, for every $\varepsilon>0$, there is $C_\varepsilon<\infty$ such that every complex-valued function $f:\{-1,1\}^n\to\C$ of Fourier degree at most $m$ satisfies \[ \left(\sum_{A\subseteq[n]}\abs{\wh f(A)}^{q_m}\right)^{1/q_m} \le C_\varepsilon m^{\beta_0+\varepsilon}\norm{f}_\infty. \] The improvement over the $m^9$ estimate of the earlier draft has two ingredients. The first three Fourier levels are estimated at the scales $m$, $m^{3/2}$, and $m^{5/3}$. These losses are encoded in the weight $M^{\mu_r/r}r^B\min\{r,L_M\}^8$, $L_M\asymp_B \log (M+1)$, with $\mu_r=\lceil3r/2\rceil$ for $r\ge2$. A parity-compatible central window handles $r<L_M$, while a shrinking balanced window handles $r\ge L_M\asymp_B\log(M+1)$, no parity is used in this regime. The leading high-degree loss–gain factor is $\ee\,2^{-B}$; careful uniform bounds close the bootstrap for every $B>1/\log2$. The formula for $\mu_r$ is explained below.

As digital commerce ecosystems expand into low-end consumer electronics (CE), hardware constraints-specifically limited CPU duty cycles and volatile heap fragmentation-become significant bottlenecks for complex transactional flows. Traditional on-device middleware requires high "network chattiness" to manage multi-step state machines, leading to increased latency and potential transaction failure on unstable residential networks. This paper proposes a novel Transactional Backend-for-Front-End (T-BFF) Orchestration Framework designed for resource-constrained embedded platforms. By migrating the transactional state machine and microservice orchestration to a cloud-side layer, we achieve a significant reduction in client-side overhead. Our framework introduces a "Double-Handshake" protocol utilizing non-volatile flash memory for state recovery after hardware reboots. Experimental results on an ARM Cortex-A53 platform demonstrate a 35% reduction in maximum heap usage and a 40% improvement in end-to-end transaction latency. This framework provides a scalable, sustainable blueprint for maintaining transactional integrity on legacy hardware in the 2026 IoT landscape.

As language models (LMs) rise in prominence, there is interest in making them more transparent in order to better understand their internal behavior. Recent interpretability work has focused on using sparse autoencoders (SAEs) to break down neuron activations at a given layer in the LM into human-understandable features, where each feature represents a concept that the model has learned. In this paper, we share work on using SAEs to analyze the behavior of text classification LMs. We present techniques for exploring the relationships between the SAE's features and the model's predictions and errors. We integrate these techniques into SAEfarer, a tool for analyzing concepts learned by text classification LMs. We assess SAEfarer in an expert pilot evaluation with five Ph.D. students.

Learning analytics dashboards (LADs) are intended to help students make sense of their learning data to support reflection and decision-making. However, their visualisations can be complex, particularly for students with low visualisation literacy. Narrative techniques, such as annotated charts and data comics, have been used to communicate insights directly, but not as supplementary materials to empower students to explore their visualisations themselves. In response, we conducted a qualitative study examining how data comics can complement LADs. We interviewed 18 nursing students and 4 of their teachers about a multimodal LAD containing visualisations with data comics explaining them. Analysis showed that data comics were clear, engaging, and helped make complex visualisations more accessible, though they must be carefully designed to avoid overwhelming students with information. The findings suggest that both students and teachers are receptive to data comics as a means of supporting the interpretability of LADs.

Replacing attention in a pretrained language model is a compatibility problem: a plausible substitute may alter representations expected by later layers. TinyCeNN-LM introduces a quality-gated post-training conversion framework using CeNN-inspired cellular-recurrent layers with bounded local processing, compact recurrent memory, routing, fusion, and accept-or-rollback validation. Three implementations are studied: Integrated Memory, MemoryFusion, and PDelta3-GDN2-CLVR+Local32. Strict PDelta3 conversion accepts a layer only when representation and NLL criteria pass fixed thresholds. On SmolLM2-135M, layers 0-2 are accepted with cumulative $\Delta\mathrm{NLL}=+0.01209$, while layer 3 is rejected despite acceptable NLL because representation fidelity fails. On Qwen3.5-0.8B, full-attention layers 3, 7, and 11 are accepted with final $\Delta\mathrm{NLL}=+0.02073$. Integrated Memory keeps perplexity within $-0.07\%$ to $+0.93\%$ while reducing total cache by up to $6.01\%$. A sampled 200-item downstream sanity check gives $28.5\%$–$32.0\%$ overall accuracy for converted Qwen releases. The results support conservative, quality-gated structural conversion rather than universal attention replacement or speedup.

Octopus crawling motivates soft robots that exploit redundancy, yet discovering and organizing diverse coordination modes for adaptation remains challenging. To address this, we introduce a Diffusion-based Uncertainty-aware Optimization (DUO) algorithm that learns demonstration-free crawling controllers for a simulated, muscle-actuated CyberOctopus. This work represents the first application of diffusion-based control to soft multi-arm robots in contact-rich simulations. By embedding a variety of locomotion behaviors within a shared control distribution, this approach enables the simulated octopus to navigate dynamic physical constraints, demonstrating that learned coordination diversity inherently facilitates robust adaptation. The main contributions include: (i) a symmetry-structured policy representation that folds radially equivalent controllers into a canonical directional sector, (ii) an online black-box optimization strategy, the DUO algorithm, that discovers and retains diverse coordination modes, and (iii) a control editing technique that adapts existing controllers to novel actuator constraints without retraining. These results show how learned coordination diversity makes motor abundance a practical resource for adaptation in soft multi-arm robots.

Mixture-of-Experts (MoE) decoding on scratchpad-based tensor accelerators (STA) is dominated by moving expert weights while the compute engines sit idle. This traffic is hard to hide, because the experts are known only after routing, and hard to shrink without losing quality or adding critical-path work. We present CARDAN, which represents each expert-weight matrix as a vector-quantized component plus a shared-basis low-rank component and co-designs this representation with a multi-engine decoding dataflow. The representation separates expert-common from expert-private work, so the dataflow overlaps DMA with computation on several engines. Across five MoE families on AWS Trainium3, CARDAN matches or improves BF16-teacher perplexity across all five models and speeds up batch-one decoding by 1.15-1.31x over AWS dense MoE megakernels, rising to 1.7x at batch size 16.

SQL has been augmented with AI operators, enabling modern data analytics platforms to derive insights from both structured and unstructured data. We observe that while current Text-to-SQL systems can successfully generate these AI-augmented queries, reliably evaluating their correctness remains a critical open challenge. Current metrics, which rely on exact query results and deterministic execution, systematically fail against the flexible, non-deterministic outputs of AI operators. In this paper, we formalize these unique evaluation failure modes and introduce a Multilayered Evaluation Framework that decouples deterministic database logic from flexible AI semantics. We test our approach across both industry (BigQuery) and academic (ThalamusDB) systems. We demonstrate that traditional Execution Accuracy severely penalizes valid queries, achieving as low as a 25% detection rate for correct translations. Furthermore, even a state-of-the-art LLM-based autorater falsely rejects 32% of accurate queries due to the complexity of judging both relational and AI components simultaneously. By validating the standard relational logic and the AI operations separately, our framework achieves state-of-the-art overall accuracy across both platforms (up to 97.2%), proposing a reliable standard for benchmarking AI-powered SQL generators.

Multi-party human-robot collaboration poses a dual challenge: robot decisions should remain interpretable and auditable, while executed actions must satisfy safety constraints during physical interaction. Combining explainable decision-tree policies with control-barrier-function (CBF) filtering provides a promising architecture but creates two learning mismatches in multi-agent reinforcement learning. Safety projection changes the action applied to the environment, while the coupled proposal graph can misalign independently optimized actor updates with a team-level update. We present safety-aligned gradient enforcement (SAGE) to address both mismatches. Its shield-annealed internalization layer (SAIL) uses a differentiable finite-penalty proposal map while retaining the exact CBF quadratic program for execution, preserving constraint-normal sensitivity to internalize repeatedly active safety constraints. Team-averaged Lyapunov policy optimization (TALO) constructs a team-aware update reference and applies a Lyapunov half-space correction to regulate independent actor updates. Physical experiments with two humanoid robots and a human partner demonstrate deployment feasibility. Across nine simulation scenarios, SAGE achieves a 71.0% success rate with 0.5 collision steps per thousand environment steps. Ablations show that direct CBF filtering reduces collision frequency by 98.5% but decreases success from 67.3% to 59.3%. SAIL reduces proposal violation by 48.8% and proposal-execution correction by 85.2%, while TALO reduces the update-consistency gap by 50.8%.

The 30 and 40 GHz instruments of the QUIJOTE radio astronomy experiment use very sensitive receivers which need to be characterized both to ensure their proper functioning and for their calibration. Given the age of the PXI-1031DC device used up until now for this purpose, the design of a new device, the Calibrator TFGI QUIJOTE, has been proposed. This new device has been designed with the idea of being simple and modular, easy to use and upgrade. It is built on a Raspberry Pi 5 system, using an MCC 118 as analog reading module. This system, widely known among both makers and professionals, not only meets the requirements of modularity and ease of use, but also ensures that the device is much more economically competitive than the PXI. A first version of the device has been built and tested, obtaining good results in accuracy and ease of use.

We propose a decoupled, layerwise method for structurally sparsifying the fully connected layers of pretrained neural networks. Rather than penalizing all layers jointly, our approach extracts shallow two-layer subnetworks, normalizes the inner weights, and applies a structured group penalty to the outer weight matrix of each block, processing layers sequentially to prune neurons and reduce the width of each layer. We prove that the constrained decoupled objective is equivalent at optimality to a specific joint penalty on the inner and outer weights, for any positively homogeneous activation, and thus admits a clean projected and proximal formulation. Our central finding is that this decoupled reformulation is more robust than coupled methods. In numerical experiments it provides a wider usable range of the regularization strength and a lower rate of catastrophic over-pruning than the tested joint baseline while maintaining comparable accuracy. We establish these properties in controlled classification and sparse-recovery studies, and examine their scope in a high-dimensional PINN stress test and in the feed-forward layers of OPT-1.3B.

The dominant method of processing sonar data is using image-based representations, requiring the preprocessing of image data on autonomous systems. We propose an alternative data processing method for remote sensing applications via the use of data in Comma-Seperated Value format. Experimentation on our alternative approach shows a reduction of processing time by 91.18%, an improvement in accurate object detection by Machine Learning, and an increase in SNR (Signal-to-noise ratio), PSNR (Peak signal-to-noise ratio), and other evaluation metrics.

Manipulating previously unseen objects remains challenging, as their dynamics depend on latent physical properties, such as friction and mass distribution, that cannot be inferred from perception alone. Prior experience across objects can provide an initial estimate of unseen object dynamics, but this estimate remains uncertain and can degrade further during sim-to-real transfer. Adapting the dynamics through interaction can progressively refine the estimation, however, updating the model may invalidate the planned trajectory. Successful and efficient manipulation therefore requires both rapid dynamics adaptation and a planning strategy that can incorporate this evolution. In this work, we introduce MetaPusher, a meta-learning and adaptive planning framework for nonprehensile manipulation of unseen objects without prior object-specific interactions. A meta-learned dynamics model rapidly adapts from interactions during task execution, while an adaptive kinodynamic planner updates long-horizon plans by reusing and refining its existing search tree. This coupling enables manipulation and adaptation without a separate data collection phase. We evaluate MetaPusher on unseen objects in simulation and in sim-to-real scenarios, comparing against fine-tuning and active learning methods, MPPI-based control, and a reinforcement learning policy. It achieves lower prediction error and improves task success rate by up to 20%.

AI productivity is often measured by task completion time, economic value, or improvements in outcome quality. However, these measures usually treat collaboration as a black box where they capture what output was produced, but not the interaction cost required to produce it. Motivated by economics literature, we introduce a productivity-oriented framework for evaluating human-AI collaboration as outcome quality relative to interaction cost. Across two datasets spanning four tasks, we show that: (1) sessions with identical quality ratings can differ by up to 70 times in interaction cost; (2) quality-cost relationships vary by task, with some tasks rewarding extended interaction and others favoring fast convergence; (3) subjective user ratings are not reliable substitutes for productivity; and (4) productive sessions are characterized by agents probing earlier and users spending less effort repairing the interaction. By distinguishing productive success from costly success, our framework makes interactional cost visible and shows how dialogue analysis can inform the evaluation and design of AI systems.

Let $\mathcal{A}_0$ be the set of finitely generated free groups on some countable alphabet and and define $\mathcal{A}_r$ inductively as the union of $\mathcal{A}_{r-1}$ with the set of amalgamated free products $G_1\ast_{\ZZ}G_2$, where $G_1,G_2\in \mathcal{A}_{r-1}$. We show, for all $r$, that every element of $\mathcal{A}_r$ has linear conjugator length, quantifying a result of Larsen and extending a result of Kharlampovich and Myasnikov. This is achieved by a linear conjugator length bound on a family of iterated HNN extensions of free groups. Also, in an appendix, we adapt a proof of A. Wei$\ss$ to show that all Generalized Baumslag-Solitar Groups have linear conjugator length. As a consequence, we obtain that one-relator groups with non-trivial center have linear conjugator length, giving evidence for a conjecture of Bridson, Riley, and Sale.

Robots carrying out tasks in dark environments need to localize from a single RGB camera, in light so low that the per-pixel signal approaches the sensor's own noise, on a power-constrained onboard computer, in real time. Each of these constraints has matured pipelines, but the intersection does not. Offline low-light reconstruction now recovers structure below -4 dB but is far too slow to run in real time, while the real-time monocular systems a robot can actually carry (DROID-SLAM, DPV-SLAM, etc.) degrade or fail when SNR gets low. We measured how they fail: across the nine lowest darkness levels of our scenes, DROID-SLAM returns a full-length trajectory carrying no information about the camera's motion on all nine, VGGT-SLAM and CUT3R on eight, pi^3 on seven, and DPV-SLAM on four. We present SYS, a monocular pipeline built on a low-light feed-forward pointmap front end with an explicit match gate, which returns three tracked trajectories and no uninformative ones, at the lowest error of any method where it tracks (24-47% of the no-information ceiling against 56-73% for the strongest baseline), and at the narrowest coverage. On a real robot video take in which 86.5% of delivered frames are entirely black, every configuration of ours stops after the lit beginning, while DROID-SLAM and DPV-SLAM each emit a pose for all 1178 frames. Our method contribution is an embedded execution path for the Jetson AGX Orin: running the map, keyframes and backend at 384 pixels with tracking at 256, together with two fixes to the per-frame pose solve, is a replicated Pareto improvement, 1.28x throughput at 0.964x error on one scene and 1.42x at 0.68x on a second, with 47% less peak GPU memory and 29% less energy per pose. We evaluate on a calibrated, bit-exact regenerable noise ladder, on relabelled real-world dark exposures, and on a new dark-room video ladder recorded from a Boston Dynamics Spot robot.

Fine-tuning has emerged as a widely adopted approach for adapting LLMs to a variety of downstream tasks. However, how it reshapes their internal mechanisms remains poorly understood. To address this, we investigate how fine-tuning alters internal representations in LLMs, including attention patterns and layer-wise activations, and examine whether these changes are linked to task-relevant components identified by EAP (e.g., attention heads and logit-level activations) that drive task performance. We find that EAP-identified components are concentrated within specific layers, indicating a degree of functional localisation in how models internalise task-specific behavior. Notably, the distribution of these components across layers is largely uncorrelated with the layers undergoing the most substantial representational changes during fine-tuning. Furthermore, we observe that overlap in EAP-identified components across tasks does not translate into cross-task performance transfer if the tasks are different in nature (e.g. classification vs. generative tasks). More specifically, fine-tuning on one task can lead to a degradation of performance on another when the two tasks exhibit a high degree of overlap in their EAP-identified components.

Training a visuomotor policy calls for abundant demonstrations that closely match the target environment, yet collecting them anew remains expensive. Existing demonstration synthesis methods reduce this cost but remain constrained by high manual effort, limited visual fidelity, or heavy reliance on physics simulators. This paper introduces GaussianFactory, a high-fidelity data engine that mass-produces demonstrations with a single video scan as its only human input and no physics engine in the generation loop. Specifically, GaussianFactory reconstructs the scene as an editable 3D Gaussian Splatting (3DGS) replica and samples from the object-combination tasks the scene affords. For each task, it plans grasps and trajectories purely kinematically on the geometric reconstruction, rendering photorealistic demonstrations that visually match the target environment. Physical dynamics enter the pipeline only where contact force interactions dictate the outcome—during grasp formation, via a learned contact model pretrained once on an interaction dataset. To evaluate the downstream utility of the synthesized demonstrations, we implement an end-to-end scan-to-deployment workflow in two setups: a simulated scene that stands in for the real world to enable reproducibility, and a real-world workspace with a physical UR10e robot. In each setup, a standard diffusion policy trained solely on the synthesized demonstrations achieves 95.1% and 84.2% success rates, respectively.

Scientific sensor data is foundational across disciplines including spacecraft engineering, clinical medicine, and atmospheric science. In each context, pipelines are constructed to ingest raw archives, assess data quality, perform feature engineering and semantic annotation, and record provenance. However, these pipelines are often implemented as monolithic, domain-specific scripts with implicit assumptions and limited reusability across fields. This work introduces SensorWF, a FAIR-annotated workflow framework for generalizable scientific time-series analysis. The framework features a five-module reusable core (M1-M5) with typed input/output contracts. A domain adapter pattern isolates all domain-specific logic within M1, enabling modules M2-M5 to operate identically across disciplines. Domain assumptions are encoded in a machine-readable component registry, enabling reuse across sensor domains without modifying the analytical core. SensorWF also generates runtime PROV-O/ProvONE provenance traces with SHA-256 checksums for all file-path entities and emits SSN/SOSA-aligned OWL ontologies as primary outputs. To assess generalizability, SensorWF is instantiated in three distinct scientific domains: spacecraft telemetry, ambulatory ECG, and atmospheric climate, with synthetic fault injection and multi-detector anomaly detection demonstrated as use-case extensions. Results show that a single codebase, parameterized solely through M1 adapters of approximately 170-500 lines each, supports analytical pipelines across domains with varying sampling rates, channel counts, and fault taxonomies. All code, the component registry, ontology artifacts, and datasets are made available as an open scientific object. Our codebase is publicly available at https://purl.archive.org/sensor-wf.

Recent advances in large language models have improved their effectiveness as back-end components for voice assistants, particularly in intent understanding and context-aware input classification. However, online-hosted models introduce network dependency and variable inference latency, limiting their suitability for time-critical autonomous driving applications. In this work, we address these issues by developing Jarvis, an offline voice assistant for high-level behavioral commands of autonomous vehicles. Its architecture integrates speech recognition and synthesis with natural language command classification into a lightweight, local framework. Jarvis core component is a text-to-command classifier, built using a domain-specific fine-tuning of the Mistral 7B model, demonstrating low-latency inference. Our experimental evaluation demonstrates that our solution outperforms larger online-hosted models, achieving 97.63 % intent recognition accuracy with an average processing latency of 1.39 s, making it well-suited for operations requiring quick response times. To support further research and fine-tuning, we provide an open-source implementation.

Deep reinforcement learning (DRL) has achieved strong performance across a wide range of continuous-control problems. These continuous-control policies, however, are often defined as direct mappings from an observed state to an action or action distribution, requiring a single feed-forward network to construct an optimal control decision in one pass. While effective, this formulation leaves little opportunity for the policy to reconsider or progressively improve an action once an initial prediction has been formed. In this work, we explore an alternative approach: rather than learning only to directly predict an action, can a policy learn to iteratively improve one, and can this iterative process provide advantages during policy learning? We introduce Iterative Action Refinement (IAR), an iterative action-construction method that constructs control actions through a sequence of learned residual corrections. Starting from an initial proposal, a shared refinement network repeatedly conditions on the observed state and the current action proposal, allowing each refinement step to revise the action constructed by preceding steps. The final refined proposal is then used to determine the action executed by the agent. We integrate this iterative action-construction mechanism with Proximal Policy Optimization (PPO), yielding REFINEPPO. We evaluate REFINEPPO across 14 benchmark control tasks, complemented by controlled ablations of refinement depth and update schedules and analyses aimed at understanding why iterative refinement is effective. Across these environments, REFINEPPO matches or exceeds the performance of standard PPO while demonstrating faster convergence on several tasks.

While learning from human motions has enabled highly dynamic humanoid skills such as dancing and martial arts in obstacle-free space, traversal through densely cluttered environments remains underexplored. These spaces are three-dimensional and geometrically constrained, requiring scene-aware locomotion that tightly couples whole-body motion with scene geometry for obstacle avoidance. To address these challenges, we present Moving Through Clutter (MTC), a learning-from-demonstration framework for scene-aware humanoid locomotion. To bypass costly physical scene construction, MTC uses procedurally generated Virtual Reality environments for immersive data collection. To transform these human motions into training-ready humanoid motions, we propose a scene-aware motion retargeting algorithm that converts human demonstrations into humanoid trajectories while strictly enforcing robot-scene clearance to guarantee collision-free traversal. These reference trajectories are then used to train a scene-aware locomotion policy that deploys on a Unitree G1 humanoid. Evaluated on our proposed MTC-Challenge for multi-obstacle traversal, the policy demonstrates a 70.2% collision-free rate across diverse scenarios, successfully traversing complex environments through diverse whole-body skills, including crawling through low-clearance passages and squeezing through narrow gaps.

Modern networks are large in scale and heterogeneous in configuration, making manual policy management increasingly impractical. Intent-Based Networking (IBN) addresses this by automating the translation of high-level operator goals into low-level network configurations. Yet existing IBN systems rely on static heuristics and fixed-feature classifiers that generalize poorly to distribution shifts such as new service definitions or evolving phrasing in operator requests. Large Language Models (LLMs), with strong reasoning and translation capabilities demonstrated across many domains, are a natural candidate for IBN policy generation. However, it is unclear whether LLMs can be reliably applied to this task, nor whether their use mitigates or worsens the underlying security risk. In this work, we show that while fine-tuned LLMs excel at intent translation, they exhibit false negative rates when checking whether a proposed intent violates an existing security policy. The root cause is not a lack of logical reasoning capability, but LLMs lack of persistent grounding in network topology and group hierarchy. Motivated by this finding, we introduce NetInspector, a three-layer agentic framework that enforces a verify-then-act protocol, decoupling information retrieval from reasoning so that the LLM focuses on symbolic reasoning while every policy decision is grounded in verifiable network facts retrieved from a live Environment Layer before approval. On NetInspector-Bench, a 2,224-sample synthetic benchmark spanning campus, enterprise, and WAN topologies, NetInspector reduces FNR by over 30% relative to ungrounded baselines and remains robust under linguistic distribution shifts.

GPU implementations of the Muon optimizer and of first-order semidefinite programming (SDP) solvers share one computational pattern: a matrix factorization is replaced by a fixed-depth polynomial filter applied after normalization. When a few dominant spectral components carry most of the input's scale, normalization pushes the remaining spectrum toward zero, where the filter is least accurate. We propose spectral deflation: estimate the dominant components, remove them, filter the normalized residual with the unchanged filter, and restore them. Deflation preserves the target matrix function, and we prove that it strictly reduces the finite-step error of the classical Newton–Schulz family. Implemented with batched randomized SVD for Muon and warm-started subspace tracking for ADMM, deflation consistently improves GPT-2 pretraining over the corresponding Muon baselines with both the Newton–Schulz and Polar Express mappings, also in wall-clock time, and lowers the KKT residuals of factorization-free ADMM on large-scale SDPs within a similar projection time.

Learning in robotic games is constrained not only by strategic information but also by what the body can still execute. We study this coupling in a hierarchical humanoid-quadruped penalty system in which game-level self-play policies command fixed soccer whole-body controllers (S-WBCs). The humanoid shooting skill is initialized from self-collected motion-capture data, whereas the quadruped saving skill is learned by reinforcement learning. We introduce dynamics-induced commitment mapping (DIC-Map), a body-grounded analysis that estimates continuation capability, identifies the first persistent loss of a terminal alternative, and tests whether the remaining interaction admits a reduced zero-sum game. For symmetric terminal alternatives, the reduced game yields a closed-form bound on optimal strategy concentration determined by the responder's value of deferring. We further show that, when the responder acts through an estimator, equal response values eliminate the direct terminal-allocation gradient and leave an estimator-mediated first-order learning channel. Experiments locate commitment about 0.29 s before contact, and changing only ball speed shifts deferral coverage. Across four responder policies, replacing the estimator raises save rate from 0.240 to 0.472, whereas a comparable gain in read accuracy obtained by waiting raises it only to 0.246. Posterior analysis is used for the equilibrium comparison because the available coverage terms are observational proxies. Project website: https://chris-ruizegeng.github.io/penaltykick/

Coaxial rotor drones have generated considerable interest because of energy efficiency and small size, but they are afflicted with inherent underactuation for roll and pitch control, although systems like swashplates have circumvented this limitation at the cost of greater mechanical complexity. This work presents a novel coaxial drone supplemented by a two-degreesof-freedom pendulum mechanism for active thrust vectoring that offers a less mechanically complicated alternative. We develop a comprehensive Lagrangian dynamic model that does not ignore the inertial contributions of all the components, including body, servo arms, and motor assembly. A Linear Quadratic Regulator(LQR) is designed based on the linearized dynamics around the hover equilibrium. High-fidelity simulations taking actuator dynamics and sensor noise into account validate the proposed architecture. An Extended Kalman Filter (EKF) blends GPS, barometer, and IMU estimates with high accuracy for state estimation. The findings verify the potential and reliability of this approach for power-saving, rapid coaxial UAVs.

In this work, we examine the topology of information flow patterns within attention graphs to effectively distinguish hallucinated from non-hallucinated responses. We analyze the Forman-Ricci curvature to identify structural patterns indicating information bottlenecks in attention graphs. We then introduce a method that captures both semi-local and global information-flow characteristics of attention heads associated with hallucinated responses. We evaluate our approach extensively across several LLMs and established benchmarks. Empirical results demonstrate that our proposed single-pass approach provides consistent improvements over existing attention-based and multi-response baselines across two hallucination-detection benchmarks, while achieving competitive performance across diverse LLM architectures. Further analysis reveals that impaired context sharing among tokens during causal generation is strongly associated with hallucination occurrences in LLMs. In particular, hallucinated responses are consistently characterized by an over-reliance on self-attention, diffused context retrieval from earlier tokens, or information over-squashing, especially in the final transformer layer.