Papers

While music generation models have evolved to handle complex multimodal inputs mixing text, lyrics, and reference audio, evaluation mechanisms have lagged behind. In this paper, we bridge this critical gap by establishing a comprehensive ecosystem for music reward modeling under Compositional Multimodal Instruction (CMI), where the generated music may be conditioned on text descriptions, lyrics, and audio prompts. We first introduce CMI-Pref-Pseudo, a large-scale preference dataset comprising 110k pseudo-labeled samples, and CMI-Pref, a high-quality, human-annotated corpus tailored for fine-grained alignment tasks. To unify the evaluation landscape, we propose CMI-RewardBench, a unified benchmark that evaluates music reward models on heterogeneous samples across musicality, text-music alignment, and compositional instruction alignment. Leveraging these resources, we develop CMI reward models (CMI-RMs), a parameter-efficient reward model family capable of processing heterogeneous inputs. We evaluate their correlation with human judgment scores on musicality and alignment on CMI-Pref along with previous datasets. Further experiments demonstrate that CMI-RM not only correlates strongly with human judgments, but also enables effective inference-time scaling via top-k filtering. Code is available at GitHub (https://github.com/Haiwen-Xia/CMI-RewardBench). Model weights: CMI-RM (https://huggingface.co/HaiwenXia/CMI-RM). Datasets: CMI-Pref-Pseudo (https://huggingface.co/datasets/HaiwenXia/cmi-pref-pseudo) and CMI-Pref (https://huggingface.co/datasets/HaiwenXia/cmi-pref)

Three-photon (3-P) fluorescence microscopy enables deep in vivo imaging with subcellular resolution, but its performance is fundamentally constrained by the maximum permissible laser power required to avoid tissue heating and photodamage. Under these power-limited conditions, fluorescence signal generation, image contrast, and achievable imaging depth are strongly affected by the illumination beam profile and aberration correction strategy. In this paper, we showed that using a fixed illumination beam size was suboptimal across different imaging depths. We further showed that conventional Zernike-based adaptive optics (AO) correction degrades under reduced Gaussian illumination beam sizes due to loss of modal orthogonality. This degradation results in slow convergence, unintended focal and field-of-view shifts, and excessive wavefront deformations. To overcome these limitations, we introduced a depth-adapted AO framework in which both the illumination beam profile and the aberration correction basis were dynamically matched to the imaging conditions. By combining depth-optimised beam underfilling with a bespoke set of illumination-matched aberration modes, we achieved faster and more stable AO convergence, enhanced fluorescence signal and image quality during deep in vivo multi-channel neuroimaging. Together, these results established a practical and robust AO-enabled three-photon microscopy strategy that maximised imaging performance under realistic power constraints.

We introduce Voices of Civilizations, the first multilingual QA benchmark for evaluating audio LLMs' cultural comprehension on full-length music recordings. Covering 380 tracks across 38 languages, our automated pipeline yields 1,190 multiple-choice questions through four stages - each followed by manual verification: 1) compiling a representative music list; 2) generating cultural-background documents for each sample in the music list via LLMs; 3) extracting key attributes from those documents; and 4) constructing multiple-choice questions probing language, region associations, mood, and thematic content. We evaluate models under four conditions and report per-language accuracy. Our findings demonstrate that even state-of-the-art audio LLMs struggle to capture subtle cultural nuances without rich textual context and exhibit systematic biases in interpreting music from different cultural traditions. The dataset is publicly available on Hugging Face to foster culturally inclusive music understanding research.

Reliable grid operation depends on accurate and timely telemetry, making modern power systems vulnerable to communication layer cyberattacks. This paper evaluates how Denial of Service (DoS), Denial of Data (DoD), and False Data Injection (FDI) attacks disrupt the IEEE 14 bus system using a MATLAB only, time stepped simulation framework built on MATPOWER. The framework emulates a 24 hour operating cycle with sinusoidal load variation, introduces attack specific manipulation of load and voltage data, and performs full AC power flow solves with reactive limit enforcement (PV PQ switching). At each timestep, the system logs true and measured voltages, generator P/Q output, system losses, and voltage limit violations to capture transient cyber physical effects. Results show that DoD causes the largest physical distortions and reactive power stress, DoS masks natural variability and degrades situational awareness, and FDI creates significant discrepancies between true and perceived voltages. The study provides a compact, reproducible benchmark for analyzing cyber induced instability and informing future defense strategies.

This article proposes an energy storage-enhanced hydrogen electrolyzer (ESEHE) to provide grid-forming (GFM) services for off-grid renewable power to hydrogen (ReP2H) systems. Unlike conventional ReP2H systems that use a centralized energy storage (ES) plant, the proposed topology directly connects batteries to the DC buses of electrolysis rectifiers. A tailored virtual synchronous machine (VSM) control framework enables the electrolyzer to autonomously provide real and reactive power support. A coordinated frequency-splitting energy extraction strategy is designed to exploit both the battery and the electrolysis stack's electrical double-layer (EDL) effect on different timescales, maximizing active power support while mitigating battery and stack degradation. An adaptive equalization control strategy is further developed to balance the battery state of charge (SOC) among multiple ESEHEs operating in parallel, which optimizes energy distribution and extends battery life. Real-time simulations on StarSim validate the proposed topology and control strategies. Techno-economic analysis shows that, compared with conventional off-grid ReP2H systems based on a centralized ES plant, the ESEHE improves overall energy efficiency by 0.23% and reduces the initial total converter investment cost by roughly 6%, mainly due to the elimination of bidirectional AC/DC conversion and its associated losses in the centralized ES plant.

Precise motion control of underactuated surface vessels is a crucial task in various maritime applications. In this work, we develop a nonlinear motion control strategy for surface vessels inspired by the pursuit guidance philosophy. Any sufficiently smooth path can be seen as a continuum of virtual targets moving along a specified path, which the pursuer is trying to catch. Contrary to the traditional path-following methods, this work develops an integrated guidance and control approach capable of following any smooth path (unlike the ones composed of a finite number of straight lines and circles). The approach relies on steering the vehicle such that its velocity vector aligns with the line-of-sight (the line joining the moving virtual target and the surface vessel), resulting in a tail-chase scenario. This leads to a path-following behavior. This integrated approach also overcomes the disadvantages inherent in the traditional two-loop-based approaches. Additionally, the proposed work takes into account the asymmetric actuator constraints in the design, which makes the design close to realistic scenarios. Furthermore, the control law has been derived within a nonlinear framework using sliding mode, and thus remains applicable for a wider envelope. The stability of the proposed control strategy is formally proven. Numerical simulations for various specified paths validate the controller's accurate path-following performance.

Active noise control (ANC) is an effective approach to noise suppression, and the filtered-reference least mean square (FxLMS) algorithm is a widely adopted method in ANC systems, owing to its computational efficiency and stable performance. However, its convergence speed and noise reduction performance are highly dependent on the step size parameter. Common step-size algorithms-such as normalized and variable step-size variants-require additional computational resources and exhibit limited adaptability under varying environmental conditions. To address this challenge, a novel Monte Carlo gradient meta-learning (MCGM) approach is proposed herein to determine an appropriate step size, into which a forgetting factor is incorporated to mitigate the impact of initial zero effect. Compared to other algorithms, the proposed method imposes no additional computational burden on FxLMS operations. Numerical simulations involving real-world acoustic paths and noise signals further confirm its effectiveness and robustness.

The integration of large language models (LLMs) into wireless networks has sparked growing interest in building autonomous AI agents for wireless tasks. However, existing approaches rely heavily on manually crafted prompts and static agentic workflows, a process that is labor-intensive, unscalable, and often suboptimal. In this paper, we propose WirelessAgent++, a framework that automates the design of agentic workflows for various wireless tasks. By treating each workflow as an executable code composed of modular operators, WirelessAgent++ casts agent design as a program search problem and solves it with a domain-adapted Monte Carlo Tree Search (MCTS) algorithm. Moreover, we establish WirelessBench, a standardized multi-dimensional benchmark suite comprising Wireless Communication Homework (WCHW), Network Slicing (WCNS), and Mobile Service Assurance (WCMSA), covering knowledge reasoning, code-augmented tool use, and multi-step decision-making. Experiments demonstrate that \wap{} autonomously discovers superior workflows, achieving test scores of $78.37\%$ (WCHW), $90.95\%$ (WCNS), and $97.07\%$ (WCMSA), with a total search cost below $\$ 5$ per task. Notably, our approach outperforms state-of-the-art prompting baselines by up to $31%$ and general-purpose workflow optimizers by $11.1%$, validating its effectiveness in generating robust, self-evolving wireless agents. The code is available at https://github.com/jwentong/WirelessAgent-R2.

This paper investigates the power control problem in wireless networks by repurposing pre-trained large language models (LLMs) as relational reasoning backbones. In hyper-connected interference environments, traditional optimization methods face high computational cost, while standard message passing neural networks suffer from aggregation bottlenecks that can obscure critical high-interference structures. In response, we propose PC-LLM, a physics-informed framework that augments a pre-trained LLM with an interference-aware attention bias. The proposed bias tuning mechanism injects the physical channel gain matrix directly into the self-attention scores, enabling explicit fusion of wireless topology with pre-trained relational priors without retraining the backbone from scratch. Extensive experiments demonstrate that PC-LLM consistently outperforms both traditional optimization methods and state-of-the-art graph neural network baselines, while exhibiting exceptional zero-shot generalization to unseen environments. We further observe that topology-relevant relational reasoning is concentrated in shallow layers, whereas deeper layers encode task-irrelevant semantic noise. Motivated by this finding, we develop a lightweight adaptation strategy that reduces model depth by 50%, significantly lowering inference cost while preserving state-of-the-art spectral efficiency.

In this work, we present a test-driven, agentic framework for synthesizing a deployable low-level robot controller for navigation tasks. Given a 2D map with an image of an ultrasonic sensor-based robot, or a 3D robotic simulation environment, our framework iteratively refines the generated controller code using diagnostic feedback from structured test suites to achieve task success. We propose a dual-tier repair strategy to refine the generated code that alternates between prompt-level refinement and direct code editing. We evaluate the approach across 2D navigation tasks and 3D navigation in the Webots simulator. Experimental results show that test-driven synthesis substantially improves controller reliability and robustness over one-shot controller generation, especially when the initial prompt is underspecified. The source code and demonstration videos are available at: https://shivanshutripath.github.io/robotic_controller.github.io.

This paper considers the flexibility degradation problem caused by excessive flexible ramping product (FRP) requirements with high variable energy resource (VER) penetration}. Based on the rolling-window co-optimization model of energy and FRP, theoretical analysis of this paper reveals a unit dispatch transfer effect, in which high FRP requirements under forecast-based dispatch (FBD) constrain real-time flexibility and distort economic efficiency. To alleviate this effect, a regulated forecast-based dispatch (RFBD) approach is proposed, which moderately caps VER outputs and enhances system flexibility. Simulation results demonstrate that the proposed approach effectively lowers FRP requirements and reduces operating cost compared with FBD.

In-vivo environments, magnetically actuated soft robots offer advantages such as wireless operation and precise control, showing promising potential for painless detection and therapeutic procedures. We developed a trileg magnetically driven soft robot (TMR) whose multi-legged design enables more flexible gaits and diverse motion patterns. For the silicone made of reconfigurable soft robots, its navigation ability can be separated into sequential motions, namely squatting, rotation, lifting a leg, walking and so on. Its motion and behavior depend on its bending shapes. To bridge motion type description and specific low-level voltage control, we introduced TMR-VLA, an end-to-end multi-modal system for a trileg magnetic soft robot capable of performing hybrid motion types, which is promising for developing a navigation ability by adapting its shape to language-constrained motion types. The TMR-VLA deploys embodied endoluminal localization ability from EndoVLA, and fuses sequential frames and natural language commands as input. Low-level voltage output is generated based on the current observation state and specific motion type description. The result shows the TMR-VLA can predict how the voltage applied to TMR will change the dynamics of a silicon-made soft robot. The TMR-VLA reached a 74% average success rate.

Artificial intelligence (AI) is driving unprecedented growth in data center (DC) scale and power demand. AI workloads impose highly dynamic, difficult-to-forecast power profiles on the utility grid, creating reliability and stability challenges that conventional DC architectures are not designed to address. This paper provides a critical review of energy storage systems (ESSs) as the key enabling technology for reliable grid integration of AI DCs. We organize the review around a four-layer hierarchical taxonomy, namely chip-level buffering, rack/server-level ESSs, facility-level uninterruptible power supply (UPS) systems, and grid-scale battery energy storage systems (BESSs), supplemented by non-battery technologies including fuel cells (FCs) and thermal energy storage (TES). Each layer is analyzed with respect to response timescale, power and energy ratings, operational role, integration challenges, and coordination requirements. Key findings include: (i) AI DC load profiles differ fundamentally from traditional loads in their sub-second variability, making conventional ESS dispatch strategies insufficient; (ii) hierarchical, coordinated ESS deployment across all layers is necessary for effective load smoothing and grid support; and (iii) significant gaps remain in simulation tools, degradation modeling, load forecasting, and optimal multi-layer sizing. This review identifies open research challenges and future directions at the intersection of AI computing infrastructure and power system integration.

Reconfigurable intelligent surface (RIS) constitutes a disruptive technology for enhancing vehicular communication performance through reconfigurable propagation environments. In this paper, we propose an adaptive channel estimation framework and hybrid beamforming optimization strategy for RIS-aided vehicular multiple-input multiple-output (MIMO) systems operating in high-mobility scenarios. To address severe Doppler effects and rapid channel variations, we design a velocity-aware pilot scheme that progressively estimates cascaded channels across two timescales, leveraging tensor decomposition and adaptive grouping of passive elements. This framework dynamically balances channel estimation accuracy and spectral efficiency, significantly reducing training overhead. Furthermore, we develop a low-complexity hybrid beamforming algorithm for both narrowband single vehicle user equipment (VUE) and broadband multi-VUE systems. For single-VUE scenarios, we derive closed-form active beamforming solutions and optimize passive beamforming via alternating optimization. For multi-VUE broadband systems, we jointly optimize subcarrier allocation, power distribution, and beamforming to maximize system throughput while mitigating inter-carrier interference (ICI) caused by Doppler spread, subject to quality-of-service (QoS) constraints and RIS hardware limitations. Our simulation results demonstrate that the proposed methods achieve substantial performance gains in channel estimation efficiency, beamforming robustness, and system throughput compared to conventional schemes, particularly under high mobility conditions.

Meta-Reinforcement Learning (Meta-RL) commonly generalizes via smoothness in the task encoding. While this enables local generalization around each training task, it requires dense coverage of the task space and leaves richer task space structure untapped. In response, we develop a geometric perspective that endows the task space with a "hereditary geometry" induced by the inherent symmetries of the underlying system. Concretely, the agent reuses a policy learned at the train time by transforming states and actions through actions of a Lie group. This converts Meta-RL into symmetry discovery rather than smooth extrapolation, enabling the agent to generalize to wider regions of the task space. We show that when the task space is inherited from the symmetries of the underlying system, the task space embeds into a subgroup of those symmetries whose actions are linearizable, connected, and compact–properties that enable efficient learning and inference at the test time. To learn these structures, we develop a differential symmetry discovery method. This collapses functional invariance constraints and thereby improves numerical stability and sample efficiency over functional approaches. Empirically, on a two-dimensional navigation task, our method efficiently recovers the ground-truth symmetry and generalizes across the entire task space, while a common baseline generalizes only near training tasks.

Hearables are becoming ubiquitous, yet their sound controls remain blunt: users can either enable global noise suppression or focus on a single target sound. Real-world acoustic scenes, however, contain many simultaneous sources that users may want to adjust independently. We introduce Aurchestra, the first system to provide fine-grained, real-time soundscape control on resource-constrained hearables. Our system has two key components: (1) a dynamic interface that surfaces only active sound classes and (2) a real-time, on-device multi-output extraction network that generates separate streams for each selected class, achieving robust performance for upto 5 overlapping target sounds, and letting users mix their environment by customizing per-class volumes, much like an audio engineer mixes tracks. We optimize the model architecture for multiple compute-limited platforms and demonstrate real-time performance on 6 ms streaming audio chunks. Across real-world environments in previously unseen indoor and outdoor scenarios, our system enables expressive per-class sound control and achieves substantial improvements in target-class enhancement and interference suppression. Our results show that the world need not be heard as a single, undifferentiated stream: with Aurchestra, the soundscape becomes truly programmable.

Capacity expansion models are frequently used to inform multi-billion dollar grid infrastructure decisions, a context in which there is significant uncertainty surrounding the future need for and performance of such infrastructure. However, despite much academic literature on the topic, virtually no grid planning processes use capacity expansion models that endogenously consider uncertainty, an oversight which frequently leads to short-sighted infrastructure decisions. This is partially due to a technology transfer gap, but it is also due to a lack of methods that work at large scale. In this paper we introduce a method for endogenizing uncertainty into capacity expansion models, a variant of adaptive robust optimization, that addresses this gap. We apply the method to a real-world capacity expansion planning problem, that of the State of California, and compare its performance to that of traditional adaptive robust optimization. We find that both the traditional method and our method identify increased transmission investment as a key lever for increasing robustness and adaptability, while helping to avoid downside risks that current deterministic planning processes may be exposing ratepayers to. Our method performs similarly to the traditional method in terms of outcomes, while significantly reducing computational complexity, making it scalable to real-world planning problems.

Model predictive control offers a powerful framework for managing constrained systems, but its repeated online optimization can become computationally prohibitive. Multiparametric programming addresses this challenge by precomputing optimal solutions offline, enabling real-time control through simple function evaluation. While extensively developed for discrete-time systems, this approach suffers from combinatorial growth in solution complexity as discretization is refined. This paper presents a systematic continuous-time multiparametric framework for linear-quadratic optimal control that directly solves Pontryagin's optimality conditions without discretization artifacts. Through two illustrative examples, we demonstrate that continuous-time formulations yield solutions with substantially fewer critical regions than their discrete-time counterparts. Beyond this reduction in partition complexity, the continuous-time approach provides deeper insight into system dynamics by explicitly identifying switching times and eliminating discretization artifacts that obscure the true structure of optimal control policies. Knowledge of the switching structure also accelerates online optimization methods by providing analytical information about the solution topology. Clear step-by-step algorithms are provided for identifying switching structures, computing parametric switching times, and constructing critical regions, making the continuous-time framework accessible for practical implementation.

Global Navigation Satellite System (GNSS) radio-frequency interference (RFI), including spoofing and jamming, threatens maritime positioning, navigation, and timing. Automatic Identification System (AIS) reports provide GNSS-derived vessel positions, offering an opportunistic wide-area monitoring source; however, raw AIS streams contain communication-integrity artifacts, such as duplicated ship identifiers and stale-data retransmissions, that can mimic trajectory distortion or reporting outages, and isolated single-vessel anomalies cannot be unambiguously attributed to GNSS interference. This paper proposes an AIS-based framework that mitigates communication-integrity artifacts, generates kinematic-consistency and transmission-continuity anomaly cues, and groups retained cues into multi-vessel events using spatiotemporal Density-Based Spatial Clustering of Applications with Noise (ST-DBSCAN). Applied to approximately 966 million AIS messages from Korean coastal waters, the framework retained 48 spoofing-consistent distortion events and 98 jamming-consistent reporting-outage events. These results suggest that AIS can serve as a complementary source for maritime GNSS RFI monitoring.

Recent multimodal systems often rely on separate expert modality encoders which cause linearly scaling complexity and computational overhead with added modalities. While unified Omni-models address this via Mixture-of-Expert (MoE) architectures with specialized experts and routing, they still inflate parameter counts and introduce routing overhead. In this paper, we propose Omni-C (Omni-Compress), a single dense Transformer-based encoder that learns competitive shared representations across heterogeneous modalities–images, audio, and text–through unimodal contrastive pretraining on large-scale unaligned data. By maximizing parameter sharing in the backbone and using lightweight modality-specific projection heads, Omni-C effectively mitigates inter-modality conflicts without requiring MoE, paired supervision, or routing. This design supports efficient deployment on memory-constrained systems via sequential modality processing and low-memory inference, eliminating the need for parallel expert loading or specialized hardware. Experiments show Omni-C achieves performance comparable to expert models in unimodal and cross-model tasks, with modest zero-shot degradation on audio and text that is largely recovered through lightweight linear probing or parameter efficient fine-tuning. The unified architecture substantially reduces inference memory usage compared to multi-encoder baselines, advancing efficient and scalable multimodal learning.

We propose MEBM-Speech, a multi-scale enhanced neural decoder for speech activity detection from non-invasive magnetoencephalography (MEG) signals. Built upon the BrainMagic backbone, MEBM-Speech integrates three complementary temporal modeling mechanisms: a multi-scale convolutional module for short-term pattern extraction, a bidirectional LSTM (BiLSTM) for long-range context modeling, and a depthwise separable convolutional layer for efficient cross-scale feature fusion. A lightweight temporal jittering strategy and average pooling further improve onset robustness and boundary stability. The model performs continuous probabilistic decoding of MEG signals, enabling fine-grained detection of speech versus silence states - an ability crucial for both cognitive neuroscience and clinical applications. Comprehensive evaluations on the LibriBrain Competition 2025 Track1 benchmark demonstrate strong performance, achieving an average F1 macro of 89.3% on the validation set and comparable results on the official test leaderboard. These findings highlight the effectiveness of multi-scale temporal representation learning for robust MEG-based speech decoding.

We propose MEBM-Phoneme, a multi-scale enhanced neural decoder for phoneme classification from non-invasive magnetoencephalography (MEG) signals. Built upon the BrainMagic backbone, MEBM-Phoneme integrates a short-term multi-scale convolutional module to augment the native mid-term encoder, with fused representations via depthwise separable convolution for efficient cross-scale integration. A convolutional attention layer dynamically weights temporal dependencies to refine feature aggregation. To address class imbalance and session-specific distributional shifts, we introduce a stacking-based local validation set alongside weighted cross-entropy loss and random temporal augmentation. Comprehensive evaluations on LibriBrain Competition 2025 Track2 demonstrate robust generalization, achieving competitive phoneme decoding accuracy on the validation and official test leaderboard. These results underscore the value of hierarchical temporal modeling and training stabilization for advancing MEG-based speech perception analysis.

We introduce Whisper-RIR-Mega, a benchmark dataset of paired clean and reverberant speech for evaluating automatic speech recognition (ASR) robustness to room acoustics. Each sample pairs a clean LibriSpeech utterance with the same utterance convolved with a real room impulse response from the RIR-Mega corpus, with stratified splits by reverberation time (RT60) and direct-to-reverberant ratio (DRR). We evaluate five Whisper models (tiny through large-v3) on 1600 test samples and report word error rate (WER) and character error rate (CER) under clean and reverberant conditions. Reverberation consistently degrades performance across all model sizes; the reverb penalty in WER ranges from 2.31 to 15.50 percentage points depending on the model. Whisper-large-v3 shows the smallest penalty; Whisper-tiny shows the largest. We release the dataset, evaluation code, and baseline results to support reproducible research on robust ASR.

Colluding language-model agents can hide coordination in messages that remain policy-compliant at the surface level. We present CLBC, a protocol where generation and admission are separated: a message is admitted to transcript state only if a small verifier accepts a proof-bound envelope under a pinned predicate $\Pi$. The predicate binds policy hash, public randomness schedule, transcript chaining, latent schema constraints, canonical metadata/tool fields, and deterministic rejection codes. We show how this protocol yields an upper bound on transcript leakage in terms of latent leakage plus explicit residual channels, derive adaptive composition guarantees, and state a semantic lower bound when policy-valid alternatives remain choosable. We report extensive empirically grounded evidence: aggregate evaluation satisfies all prespecified thresholds; strict lane decoder advantage is bounded at 0.0000 with MI proxy 0.0636; adaptive-colluder stress tests remain below attacker thresholds; and baseline separation shows large gaps between reject-by-default semantics and audit-only controls. We further quantify operational tradeoffs. Strict full-proof mode has median turn latency 27.53s (p95 28.08s), while sampled proving reduces non-proved-turn latency to 0.327ms. The central finding is that bottlenecks alone are insufficient: security claims depend on verifiable admission semantics that are online, deterministic, and fail-closed.

The recently introduced notions of ranking functions and closure certificates utilize well-foundedness arguments to facilitate the verification of dynamical systems against $\omega$-regular properties. A ranking function and a closure certificate are real-valued functions defined over states and state pairs of a dynamical system whose zero superlevel sets are inductive state invariant and inductive transition invariant, respectively. The search for such certificates can be automated by fixing a specific template class, such as a polynomial of a fixed degree, and then using optimization techniques such as sum-of-squares (SOS) programming to find it. Unfortunately, such certificates may not be found for a fixed template. In such a case, one must change the template; for example, increase the degree of the polynomial. In this paper, we consider a notion of multiple functions in the form of vector certificates. Taking inspiration from the literature on vector barrier certificates as generalizations of standard barrier certificates for safety verification, we propose vector co-Büchi ranking functions and vector closure certificates as nontrivial generalizations of ranking functions and closure certificates, respectively. Both notions consist of a set of functions that jointly overapproximate an inductive invariant by considering each function to be a linear combination of the others. The advantage of such certificates is that they allow us to prove properties even when a single function for a fixed template cannot be found using standard approaches. We present an SOS programming approach to search for these functions and demonstrate the effectiveness of our proposed method in verifying $\omega$-regular specifications in several case studies.

In this study, we present a meat freshness classification framework from Red-Green-Blue (RGB) images that supports both packaged and unpackaged meat datasets. The system classifies four in-distribution (ID) meat classes and uses an out-of-distribution (OOD)-aware abstention mechanism that flags low-confidence samples as No Result. The pipeline combines U-Net-based segmentation with deep feature classifiers. Segmentation is used as a preprocessing step to isolate the meat region and reduce background, producing more consistent inputs for classification. The segmentation module achieved an Intersection over Union (IoU) of 75% and a Dice coefficient of 82%, producing standardized inputs for the classification stage. For classification, we benchmark five backbones: Residual Network-50 (ResNet-50), Vision Transformer-Base/16 (ViT-B/16), Swin Transformer-Tiny (Swin-T), EfficientNet-B0, and MobileNetV3-Small. We use nested 5x3 cross-validation (CV) for model selection and hyperparameter tuning. On the held-out ID test set, EfficientNet-B0 achieves the highest accuracy (98.10%), followed by ResNet-50 and MobileNetV3-Small (both 97.63%) and Swin-T (97.51%), while ViT-B/16 is lower (94.42%). We additionally evaluate OOD scoring and thresholding using standard OOD metrics and sensitivity analysis over the abstention threshold. Finally, we report on-device latency using TensorFlow Lite (TFLite) on a smartphone, highlighting practical accuracy-latency trade-offs for future deployment.

In large-scale LLM pre-training systems with 100k+ GPUs, failures become the norm rather than the exception, and restart costs can dominate wall-clock training time. However, existing fault-tolerance mechanisms are largely unprepared for this restart-dominant regime. To address this challenge, we propose SPARe - Stacked Parallelism with Adaptive Reordering - a fault-tolerance framework that masks node failures during gradient synchronization by stacking redundant data shards across parallelism groups and adaptively reordering execution. SPARe achieves availability comparable to traditional replication while maintaining near-constant computation overhead of only 2~3x, even under high redundancy where traditional replication would require linearly inflating overhead. We derive closed-form expressions for endurable failure count and computation overhead, validate them via SimGrid-based discrete-event simulation, and jointly optimize redundancy and checkpointing to minimize time-to-train. At extreme scale with up to 600k GPUs, SPARe reduces time-to-train by 40~50% compared to traditional replication.

Listening to heart and lung sounds - auscultation - is one of the first and most fundamental steps in a clinical examination. Despite being fast and non-invasive, it demands years of experience to interpret subtle audio cues. Recent deep learning methods have made progress in automating cardiopulmonary sound analysis, yet most are restricted to simple classification and offer little clinical interpretability or decision support. We present StethoLM, the first audio-language model specialized for cardiopulmonary auscultation, capable of performing instruction-driven clinical tasks across the full spectrum of auscultation analysis. StethoLM integrates audio encoding with a medical language model backbone and is trained on StethoBench, a comprehensive benchmark comprising 77,027 instruction-response pairs synthesized from 16,125 labeled cardiopulmonary recordings spanning seven clinical task categories: binary classification, detection, reporting, reasoning, differential diagnosis, comparison, and location-based analysis. Through multi-stage training that combines supervised fine-tuning and direct preference optimization, StethoLM achieves substantial gains in performance and robustness on out-of-distribution data. Our work establishes a foundation for instruction-following AI systems in clinical auscultation.

This paper introduces inspection through GLASS, a Geometric Look-Angle Shaping Strategy for enclosed regions using unmanned aerial vehicles. In doing so, the vehicles guidance command is constructed through a bounded, geometry-consistent shaping of the look angle relative to a desired standoff path. By embedding a smooth, hyperbolic-tangent-type shaping function within a polar geometric framework, GLASS ensures global existence of the guidance dynamics. It avoids the far-field limitations inherent to conventional formulations. Lyapunov stability analysis establishes asymptotic convergence to a prescribed inspection standoff under explicit curvature feasibility conditions, along with analytical settling-time characteristics. The proposed strategy incorporates maximum turn-rate constraints without inducing singularities throughout the workspace. High-fidelity six-degree-of-freedom quadrotor simulations demonstrate the effectiveness of GLASS in representative enclosed inspection scenarios, highlighting a practically viable guidance framework for autonomous enclosed inspection missions.

Accurate reference trajectories are required to validate autonomous agricultural robots and highly automated off-road vehicles under real-world field conditions. In practice, robotic total stations provide millimeter-level prism center coordinates, but the point of interest on the vehicle is typically displaced by a lever arm, ranging from decimeters to multiple meters. Roll and pitch motions, as typically observed in off-road machinery, therefore introduce horizontal point of interest errors far exceeding the measurement accuracy of robotic total stations observations. This paper presents the design, implementation, and validation of a Smart Prism prototype that augments a robotic total station prism with an inertial measurement unit to enable real-time tilt compensation. The prototype integrates an STM32H7 microcontroller and a Murata SCH16T-series IMU and estimates roll and pitch angles using an adaptive complementary filter. The tilt-compensated point of interest coordinates are obtained by transforming a calibrated lever arm from the body frame into the navigation frame and combining it with robotic total station prism positions. To support vehicle-side integration, the system can transmit prism and tilt-compensated point of interest coordinates on the Controller Area Network bus, allowing the point of interest to be treated as a virtual position sensor (e.g., co-located with a rear-axle reference point). Experiments with a fixed ground reference point, using a prism to point of interest lever arm of approximately 1.07m and manual roll/pitch excursions of up to 60 deg, yield three-dimensional root-mean-square errors between 2.9mm and 23.6mm across five test series. The results demonstrate that IMU-based tilt compensation enables reference measurements suitable for validating centimeter-level navigation systems under dynamic field conditions.