Papers

We present E2E-WAVE, the first end-to-end learned waveform generation system for underwater video multicasting. Acoustic channels exhibit 20–46% bit error rates where forward error correction becomes counterproductive – LDPC increases rather than decreases errors beyond its decoding threshold. E2E-WAVE addresses this by embedding semantic similarity directly into physical layer waveforms: when decoding errors are unavoidable, the system preferentially selects semantically similar tokens rather than arbitrary corruption. Combining VideoGPT tokenization (1024x compression) with a trainable waveform bank and fully differentiable OFDM transmission, E2E-WAVE achieves +5 dB (19.26%) PSNR and +0.10 (14.28%) SSIM over the strongest FEC-protected baseline in less challenging underwater channel (NOF1) while delivering real-time 16 FPS video at 128x128 resolution over 2.3 kbps channels – impossible for conventional digital modulation. The performance gap only increases in harsher channels (BCH1, NCS1). Trained on a single channel, E2E-WAVE generalizes to unseen underwater environments without retraining, while HEVC fails at sub-5 kbps rates and SoftCast's AWGN assumptions collapse on frequency-selective channels.

This study presents a physically informed hybrid time-frequency and machine learning (STFT-ML) framework for arc stability monitoring in electric arc welding systems. The primary current signal is modeled as a stochastic representation of plasma dynamics and transformed into a structured feature space using localized spectral energy distributions. Within this framework, the Arc Stability Index (ASI), spectral entropy (Hs), and harmonic distortion (THDarc) are defined as energy-based descriptors and integrated with complementary time-domain features to capture both spectral redistribution and temporal variability. Experimental evaluation demonstrates that the SVM-RBF classifier achieves a hold-out accuracy of 94.4%. However, cross-validation results (85.6% for Leave-One-Out and 87.5% +/- 9.4 for 10-fold) and a 95% confidence interval of [81.65%, 92.50%] provide a more realistic assessment of generalization performance. Receiver Operating Characteristic (ROC) and Precision-Recall (PR) analyses further confirm strong class separability, particularly for stable and extinction regimes, while transient states remain more challenging due to their non-stationary nature. Compared to high-dimensional deep learning approaches, the proposed framework significantly reduces computational complexity and inference latency, enabling real-time deployment in resource-constrained environments. The results indicate that spectral energy redistribution around the fundamental frequency serves as a reliable precursor to arc instability. The main contribution of this work lies in the development of a computationally efficient and physically interpretable feature representation framework that bridges time-frequency analysis and machine learning-based classification for industrial diagnostic applications.

Ensuring strict safety guarantees is the paramount challenge for emerging 5G/6G wireless systems, particularly as they increasingly govern mission-critical applications ranging from autonomous UAV swarms to industrial automation. While deep reinforcement learning (DRL) offers a promising solution for complex resource allocation, standard algorithms frequently violate essential constraints, such as QoS mandates and power limits, posing unacceptable risks of system failure and regulatory non-compliance. We propose Safe-Deep Q-Learning, a novel algorithm that simultaneously addresses all three challenges: it handles mixed-integer nonconvex problems by approximating the Q-function, adapts to stochastic dynamics, and enforces dual-timescale constraints using integrated Lagrangian methods. Our framework features adaptive penalty scaling and constraint violation tracking, specifically tailored for wireless environments, and is designed to operate in both distributed and centralized architectural modes. We prove convergence to optimal constraint-satisfying policies under mild conditions and demonstrate robustness through dual variable stabilization. Validation on unmanned aerial vehicle (UAV) swarm control network and post-disaster emergency communications applications shows that Safe-Deep Q-Learning achieves stringent adherence to safety bounds with near-zero violation rates, significantly outperforming existing constrained RL baselines, establishing its effectiveness for safety-critical wireless deployments.

We consider a class of quadratic systems, primarily motivated by incompressible fluid flows, where the nonlinearities are generalized lossless: they do not produce or dissipate energy, as measured by a generalized quadratic metric. Our goal is to compute trapping regions, which are forward invariant sets that certify ultimate boundedness. The key contribution is a novel parameterization of the generalized lossless condition that enables optimization of trapping regions for a broader class of quadratic systems. We also formulate the conditions for ellipsoidal trapping regions, whereas spherical regions have been the focus of prior works. We provide three numerical examples, which demonstrate the improvements offered by the proposed approach relative to existing methods.

Transmission expansion planning (TEP) plays a critical role in ensuring power system reliability and facilitating the integration of renewable energy resources. However, this process requires planners to constantly deal with significant uncertainty. While multistage stochastic TEP models provide a robust framework for identifying investment plans under uncertainty, the rapid growth in problem size hinders their computational tractability. To address this challenge, this paper develops a hybrid machine learning-optimisation framework for stochastic TEP. The proposed approach uses investment decisions and uncertainty scenarios as input features to train surrogate neural networks, which are then reformulated as mixed-integer linear constraints and embedded within an optimisation model. The surrogate model approximates expected operational costs to inform TEP decisions, reducing the burden arising from large operational problems. Case study applications on IEEE test systems demonstrate that, after training, the proposed approach achieves near-optimal investment costs while reducing total computational time by up to a factor of around 13 compared to a single full-optimisation stochastic formulation. This enables performing extensive multi-scenario analysis and stress testing that would otherwise be computationally prohibitive at scale.

Renewable energy adoption has increased significantly in recent years. However, the growing penetration of renewable generation that depends on the weather introduces additional variability, making accurate net load forecasting increasingly challenging. In this study, long short-term memory (LSTM) and fully connected neural network (FCNN) models are employed to forecast net load one hour ahead using direct and indirect approaches. In the direct approach, net load is predicted directly, whereas in the indirect approach, total load, wind generation, and solar generation are predicted separately and then used to derive net load. Installed renewable generation capacity is also incorporated as an input feature to capture changes in renewable generation capability over time. A 24-hour input window and a chronological 90%/5%/5% training, validation, and test split are employed, with preprocessing fitted exclusively on the training data. On 1,754 common test timestamps, the direct FCNN slightly outperforms the indirect FCNN, whereas the indirect LSTM achieves the best overall performance, with a MAPE of 2.46%, an RMSPE of 3.59%, and an R2 of 98.7%. These results demonstrate that component-wise forecasting can improve LSTM-based net load prediction when the characteristics of renewable generation and installed capacity information are represented appropriately.

The growing reliance on large-scale speech data has made privacy protection a critical concern. However, existing anonymization approaches often degrade data utility, for example by disrupting acoustic continuity or reducing vocal diversity, which compromises the value of speech data for downstream tasks such as Automatic Speech Recognition (ASR), Text-to-Speech (TTS), and Speech Emotion Recognition (SER). Current evaluation practices are also limited, as they mainly rely on direct testing of anonymized speech with pretrained models, providing only a partial view of utility. To address these issues, we propose a novel two-stage framework that protects both linguistic content and acoustic identity while maintaining usability. For content privacy, we employ a generative speech editing model to seamlessly replace personally identifiable information (PII), and for voice privacy, we introduce F3-VA, a flow-matching-based anonymization framework with a three-stage design that produces diverse and distinct anonymized speakers. To enable a more comprehensive assessment, we evaluate privacy using both acoustic- and content-based speaker verification metrics, and assess utility by training ASR, TTS, and SER models from scratch. Experimental results show that our framework achieves stronger privacy protection with minimal utility degradation compared to baselines from the VoicePrivacy Challenge, while the proposed evaluation protocol provides a more realistic reflection of the utility of anonymized speech under privacy protection.

This paper develops a semidefinite-programming-based method for online feedback control of nonlinear systems using a state-dependent representation. We formulate sequences of time-varying SDPs whose optimal solutions jointly yield a stabilizing feedback controller and a Lyapunov certificate satisfying stability conditions and quadratic performance specifications. We further establish compact conditions certifying recursive feasibility of the resulting SDP sequences and derive estimates of the region of attraction. Numerical examples on representative nonlinear systems illustrate the flexibility and effectiveness of the proposed method.

Large Language Models have revolutionized natural language processing, achieving unprecedented success across a vast range of tasks. However, their practical application in long-context scenarios is severely hampered by the formidable memory footprint of the Key-Value cache. While channel pruning has emerged as a promising compression strategy, existing methods evaluate channel importance in isolation, fundamentally ignoring the inter-channel interactions that collectively dictate model performance. This oversight leads to suboptimal pruning decisions. To address this, we introduce GRACE (GRaph-guided Adaptive Channel Elimination), a novel framework that reframes KV cache compression as a graph-based optimization problem. GRACE models channels as nodes and their interactions as weighted edges, enabling the identification of a near-optimal channel subset for pruning by minimizing the reconstruction error of the attention weight matrix. Furthermore, GRACE incorporates an adaptive protection mechanism that shields salient key channels from removal, ensuring a robust autoregressive decoding process. Extensive experiments show that GRACE can reduce KV cache size by 60% with negligible performance degradation, consistently outperforming the state-of-the-art method.

We introduce a new framework for room acoustics modelling based on a state-space model of the boundary integral equation representing the sound field in a room. Whereas state-space models of linear time-invariant systems are traditionally constructed by means of a state vector and a 4-tuple of system matrices, the state-space representation introduced in this work consists of a state function representing the pressure distribution at the room boundary, and a 4-tuple of integral operators. We refer to this representation as a boundary integral operator state-space (BIOSS) model and provide a physical interpretation for each of the integral operators. As many mathematical operations on vectors and matrices translate to functions and operators, the BIOSS representation can be manipulated to obtain two transfer function representations, having either a feedback or a parallel feedforward structure. Consequently, various equivalent representations for room acoustics are obtained in the BIOSS framework, in the time or frequency domain, and in continuous or discrete space. We discuss two future directions for how the proposed framework can be fertile for research on room acoustics modelling. Firstly, we identify equivalences between the BIOSS framework and various existing room acoustics models (boundary element models, delay networks, geometric models), which may be used to establish relations between existing models and to develop novel room acoustics models. Secondly, we postulate on how concepts from state-space theory, such as observability, controllability, and state realization, can be used for developing new inference and control methods for room acoustics.

Unmixing reveals the spatial distribution and spectral details of different constituents, called endmembers, in a hyperspectral image. Because unmixing has limited ground truth requirements, can accommodate mixed pixels, and is closely tied to light propagation, it is a uniquely powerful tool for analyzing hyperspectral images. However, spectral variability inhibits unmixing performance, the proper way to determine the number of endmembers is ambiguous, and the clarity of the endmembers degrades as more are included. Hierarchical structure is a possible solution to all three problems. Here, hierarchical unmixing is defined by imposing a hierarchical abundance sum constraint on Deep Nonnegative Matrix Factorization. Binary Linear Unmixing Tactile Hierarchies (BLUTHs) solve the hierarchical unmixing problem with a simple network architecture. Sparsity modulation unmixing growth tailors the topology of a BLUTH to each scene. The structure imposed by BLUTHs allows endmembers with varying levels of spectral contrast to be revealed, mitigating the challenge of spectral variability. The performance of BLUTHs exceeds state-of-the-art unmixing algorithms on laboratory scenes, particularly with regard to abundance estimation, while their performance remains competitive on remote sensing scenes. In addition, ocean color unmixing by BLUTHs is demonstrated on hyperspectral scenes from the HYPSO and PACE satellites.

The paper proposes new algorithms for computing the regularization path of linear-model fitting (i) with L1 costs for the independent variables and L2 costs for the dependent variables, and (ii) with L2 costs for the independent variables and L1 costs for the dependent variables. Unlike previously proposed algorithms, the proposed algorithms for (i) and (ii) are exact duals of each other. The heart of the proposed algorithms is parametric Gaussian message passing (i.e., Kalman-type forward-backward recursions) in the pertinent factor graphs. We first develop such algorithms for linear models with general (but preferably low-dimensional) state space representations. We then specialize to specific state space representations of generic linear models; in this case, the structure of the factor graph allows to refine the proposed algorithms such that the computational complexity drops to linear scale.

This work introduces Structured 3D-SVD as a practical framework for the reconstruction, compression, and analysis of biological volumetric data. Inspired by the logic of matrix singular value decomposition (SVD), the proposed approach represents third-order volumetric data in the spatial domain and supports progressive reconstruction through ordered quasi-singular coeffients. The experimental evaluation was carried out on two biological volumetric datasets: one full-volume scan of a fish and another of a brain. The results show that Structured 3D-SVD achieves reconstruction quality close to that of Tucker decomposition while requiring shorter computation times and outperforms canonical polyadic decomposition (CPD) in both accuracy and runtime. In addition, a progressive reconstruction analysis shows that relatively low truncation levels are sufficient to preserve the main volumetric structures, while higher truncation levels lead to more detailed reconstructions.

Electroencephalography (EEG) foundation models have shown strong potential for learning generalizable representations from large-scale neural data, yet their clinical deployment is hindered by distribution shifts across clinical settings, devices, and populations. Test-time adaptation (TTA) offers a promising solution by enabling models to adapt to unlabeled target data during inference without access to source data, a valuable property in healthcare settings constrained by privacy regulations and limited labeled data. However, its effectiveness for EEG remains largely underexplored. In this work, we introduce NeuroAdapt-Bench, a systematic benchmark for evaluating test-time adaptation methods on EEG foundation models under realistic distribution shifts. We evaluate representative TTA approaches from other domains across multiple pretrained foundation models, diverse downstream tasks, and heterogeneous datasets spanning in-distribution, out-of-distribution, and extreme modality shifts (e.g., Ear-EEG). Our results show that the evaluated TTA methods yield inconsistent gains and often degrade performance, with gradient-based approaches particularly prone to heavy degradation and optimization-free methods showing greater stability. For the evaluated EEG foundation models and representative TTA methods, these findings highlight the limitations of directly applying existing TTA techniques to EEG and underscore the need for domain-specific adaptation strategies.

Clinical ultrasound analysis demands models that generalize across heterogeneous organs, views, and devices, while supporting interpretable workflow-level analysis. Existing methods often rely on task-wise adaptation, and joint learning may be unstable due to cross-task interference, making it hard to deliver workflow-level outputs in practice. To address these challenges, we present USTri, a tri-stage ultrasound intelligence pipeline for unified multi-organ, multi-task analysis. Stage I trains a universal generalist USGen on different domains to learn broad, transferable priors that are robust to device and protocol variability. To better handle domain shifts and reach task-aligned performance while preserving ultrasound shared knowledge, Stage II builds USpec by keeping USGen frozen and finetuning dataset-specific heads. Stage III introduces USAgent, which mimics clinician workflows by orchestrating USpec specialists for multi-step inference and deterministic structured reports. On the FMC\_UIA validation set, our model achieves the best overall performance across 4 task types and 27 datasets, outperforming state-of-the-art methods. Moreover, qualitative results show that USAgent produces clinically structured reports with high accuracy and interpretability. Our study suggests a scalable path to ultrasound intelligence that generalizes across heterogeneous ultrasound tasks and supports consistent end-to-end clinical workflows. The code is publicly available at: https://github.com/MacDunno/USTri.

An inherent assumption of perfect tracking in iterative learning control (ILC) is that there exists an ILC input such that the generated output can track the desired trajectory reference. This assumption may fail in practice, which gives rise to desired but untrackable tasks. This paper gives an end-to-end ILC design for repetitive untrackable tasks in closed-loop systems. The reference input is trial-to-trial updated together with the ILC feedforward input based on the measurement data. This two-player behavior of the closed-loop ILC system is investigated from a cooperative game perspective. A sufficient condition for the two-player end-to-end ILC to have a lower cost than the one-player norm optimal ILC (NOILC) is discovered. Finally, a numerical example is given to verify the effectiveness of the developed method.

In modern large-scale networked systems, rapidly solving optimization problems while utilizing communication resources efficiently is critical for addressing complex tasks. However, distributed optimization over bandwidth-limited networks must cope with two obstacles: limited-bandwidth communication, and slow convergence when the objective is ill-conditioned. In this paper, we consider unconstrained distributed optimization over strongly connected directed networks, where nodes exchange quantized (i.e., compressed) messages. We propose QANM (Quantized Averaged Nesterov Momentum), an algorithm that combines Nesterov momentum with a finite-time quantized averaging protocol. At every iteration, QANM enables all nodes to reach exact agreement, without requiring doubly stochastic weight matrices or undirected communication. For strongly convex and smooth objectives, we show that the objective gap of QANM decays at the accelerated rate $1-\kappa^{-1/2}$ (where $\kappa$ is the condition number), up to a residual of order $\Delta^2$ determined by the quantization level $\Delta$. This improves the condition-number dependence of the transient iteration complexity from $\kappa$ to $\sqrt{\kappa}$ compared with gradient descent that utilizes the same quantized averaging mechanism. Finally, we evaluate QANM on a distributed sensor fusion problem, showing that our algorithm exhibits the fastest initial error reduction compared with other methods from current literature.

Fast execution of contact-rich manipulation is critical for practical deployment, yet providing fast demonstrations for imitation learning (IL) remains challenging: humans cannot demonstrate at high speed, and naively accelerating demonstrations alters contact dynamics and induces large tracking errors. We present a method to autonomously refine time-accelerated demonstrations by repurposing Iterative Reference Learning Control (IRLC) to iteratively update the reference trajectory from observed tracking errors. However, applying IRLC directly at high speed tends to produce larger early-iteration errors and less stable transients. To address this issue, we propose Incremental Iterative Reference Learning Control (I2RLC), which gradually increases the speed while updating the reference, yielding high-fidelity trajectories. We validate on real-robot whiteboard erasing and peg-in-hole tasks using a teleoperation setup with a compliance-controlled follower and a 3D-printed haptic leader. Both IRLC and I2RLC achieve up to 10x faster demonstrations with reduced tracking error; moreover, I2RLC improves spatial similarity to the original trajectories by 22.5% on average over IRLC across three tasks and multiple speeds (3x-10x). We then use the refined trajectories to train IL policies; the resulting policies execute faster than the demonstrations and achieve 100% success rates in the peg-in-hole task at both seen and unseen positions, with I2RLC-trained policies exhibiting lower contact forces than those trained on IRLC-refined demonstrations. These results indicate that gradual speed scheduling coupled with reference adaptation provides a practical path to fast, contact-rich IL.

This paper presents an online reinforcement-learning framework for safe gain scheduling of a nonlinear quadcopter controller. Rather than learning thrust and torque commands directly, the proposed method selects gain vectors online from a finite library of pre-certified stabilizing controllers, thereby preserving the structure of the underlying snap-based control law. Safety is enforced by restricting the policy to admissible gains that maintain forward invariance of a prescribed safe state set, while dwell-time constraints prevent excessively fast switching. To reduce the action-space dimension, translational gains are shared across spatial axes by exploiting the isotropic structure of the translational dynamics, whereas yaw gains are scheduled independently. A deep Q-network learns to adjust feedback authority according to the current flight condition, using aggressive gains during large transients and milder gains near hover. High-fidelity nonlinear simulations demonstrate accurate trajectory tracking, bounded attitude motion, reduced control effort near convergence, and stable hover regulation under online safe gain scheduling.

Modern Graphics Processing Unit (GPU)-backed services must satisfy strict latency service-level objectives (SLOs) while controlling spare-capacity costs. In multi-tenant GPU cloud platforms, this trade-off is inherently dynamic because workload demand is endogenous; specifically, pricing shapes the submissions of heterogeneous tenants, which subsequently impact congestion and delay. We formulate the joint pricing-and-scaling problem as a large-population Stackelberg game problem, and we derive an explicit equilibrium demand map. The resulting closed-loop model reveals a structural failure mode in which delay-insensitive workloads sustain a residual demand floor, making the backlog undrainable under bounded price and service capacity. This observation motivates a computable drainability guardrail that certifies uniformly negative backlog drift in the residual-demand regime. For any fixed price-capacity pair satisfying the drainability guardrail, we establish global convergence to a unique operating point under a checkable step-size condition. Building on this fixed-pair analysis, we further develop an optimizer-agnostic action shield that provides a negative-drift certificate for shielded execution in the residual regime of the dynamic problem and show empirically that it improves safety and robustness for model-free reinforcement learning (RL) in this setting.

Generative semantic communication (SemCom) harnesses pretrained generative priors to improve the perceptual quality of wireless image transmission. Existing generative SemCom receivers, however, rely on maximum a posteriori (MAP) estimation, which fundamentally cannot preserve the data distribution and thus limits achievable perceptual quality. Moreover, current diffusion-based approaches using single-domain guidance face significant limitations: latent-domain guidance is sensitive to channel noise, while image-domain guidance inherits decoder bias. Simply combining both domains simultaneously yields an overconfident pseudo-posterior. In this paper, we formulate semantic decoding as a Bayesian inverse problem and prove that posterior sampling achieves optimal perceptual quality by preserving the data distribution. Building on this insight, we propose alternating dual-domain posterior sampling (ADDPS), a diffusion-based SemCom receiver that alternately enforces latent-domain and image-domain consistency during the sampling process. This alternating strategy decomposes joint posterior sampling into simpler subproblems, avoiding gradient conflicts while retaining the complementary strengths of both domains. Experiments on FFHQ demonstrate that the proposed ADDPS achieves superior perceptual quality compared with existing methods.

This tutorial derives the mathematical foundations of what it means for a carrier waveform to be predictable and non-selective. We focus on Zak-OTFS, where each carrier waveform is a pulse in the delay-Doppler (DD) domain, formally a quasi-periodic localized function with specific periods along delay and Doppler. Viewed in the time domain, the Zak-OTFS carrier is realized as a pulse train modulated by a tone (termed a pulsone). We start by providing physical intuition, describing what it means for the Zak-OTFS carrier waveforms to be geometric modes of the Heisenberg-Weyl (HW) group of discrete delay and Doppler shifts that define the discrete-time communication model. In fact, we show that these geometric modes are common eigenvectors of a maximal commutative subgroup of our discrete HW group. When the channel delay spread is less than the delay period, and the channel Doppler spread is less than the Doppler period, we show that the Zak-OTFS input-output (I/O) relation is predictable and non-selective. Given the I/O response at one DD point in a frame, it is possible to predict the I/O response at all other points, without recourse to some mathematical model of the channel. While it may be intuitive that geometric modes of the HW group are predictable and non-selective wireless carriers, this is not a requirement. We provide a necessary and sufficient condition that depends on the ambiguity properties of the basis of carrier waveforms. In fact, we show that the structure of a pulse train modulated by a Hadamard matrix is common to several families of waveforms proposed for 6G, including Zak-OTFS, AFDM, OTSM and ODDM.

This experimental dataset presents both module-level and cell-level characterization data for lithium-ion battery modules composed of three parallel-connected inhomogeneous cells across a wide range of module-level state of health (M-SoH) and cell-to-cell variation (CtCV). First, 70 cells are aged to establish an inventory with cell-level state of health (C-SoH) ranging approximately from 100% to 80% (80% is considered as the end-of-life for automotive applications). From this inventory, 78 battery modules are then assembled, each exhibiting a distinct M-SoH value (from 100% to 80.98%) and a unique CtCV value (from 0% to 9.31%, defined as population standard deviation of C-SoH within each module). Module-level characterization data are collected under 25{\deg}C at 0.5C and 0.25C conditions, enabling extraction of module-level capacities and supporting diagnostic analyses such as incremental capacity analysis and differential voltage analysis. Before a module is assembled and tested, cell-level characterization tests are conducted for every individual cell within that module under 1C conditions, enabling direct quantification of CtCV and providing accurate labels for cell-level capacities and internal resistances. The dataset is organized with both raw time-series data and processed summary information such as C-SoH, M-SoH, and CtCV for all modules. With the paired module-level and cell-level characterization data, this dataset enables understanding and development of advanced degradation monitoring mechanisms for battery modules with parallel-connected cells in the presence of CtCVs.

This paper explores the coupling of dissimilar dynamic models for system-level analysis. Design, control, and estimation for dynamic systems require accurate and analytically tractable representations. However, development of such representations is impeded by the heterogeneous modeling paradigms used for subsystems and components. Prior efforts to address this provide frameworks to link model forms together, though these can neglect integration of data-driven models or relinquish features that support analysis. This work addresses this by introducing a control-oriented framework to couple two types of representations: physics-based models, and a class of data-driven models. The framework consists of a multistep procedure that transforms subsystem models into matching representations, constructs a fully coupled model using defined coupling terms, and analyzes control properties. A physics-based microgrid with a data-driven data center load model is used to demonstrate the proposed methodology and the foundation it provides to preserve analytical tractability. Results from the case study show that the coupling structure influences the equilibrium points and stability properties of the integrated system.

Existing multimodal safety benchmarks focus solely on visual inputs and cannot assess Omni Large Language Models (LLMs) that process vision, audio, and text. We introduce MCBench, a benchmark with 1196 scenarios spanning four safety categories that require integrating multiple modalities for accurate safety assessment. Each unsafe scenario is paired with a minimally different safe counterpart to assess model sensitivity. Our evaluations of state-of-the-art models reveal significant challenges. Omni LLMs struggle with subtle or non-physical risks but perform better when salient visual or acoustic cues are present. Analysis of reasoning traces shows that, although models can extract modality-specific information, they often fail to integrate these cues effectively for safety judgments. Our findings reveal that current Omni LLMs lack robust cross-modal reasoning in safety-critical settings, underscoring the need for improved architectures and training strategies for multimodal safety.

Medical time series, such as electrocardiograms (ECG) and electroencephalograms (EEG), exhibit complex temporal dynamics and structured cross-channel dependencies, posing fundamental challenges for automated analysis. Conventional convolutional and recurrent models struggle to capture long-range dependencies, while Transformer-based approaches incur quadratic complexity and often introduce redundant interactions that are misaligned with the intrinsic structure of physiological signals. To address these limitations, we propose MedMamba, a principle-driven multi-scale bidirectional state space architecture tailored for medical time series classification. Our design is guided by three key inductive biases of physiological signals: spatial centralization, multi-timescale temporal composition, and non-causal contextual dependency. These principles are instantiated through a lightweight channel-mixing module for cross-channel reparameterization, multi-scale convolutional tokenization for temporal decomposition, and bidirectional Mamba blocks for efficient global context modeling with linear complexity. Extensive experiments on six benchmark datasets spanning EEG, ECG, and human activity signals demonstrate that MedMamba consistently outperforms state-of-the-art methods across diverse modalities. Notably, it achieves 85.97% accuracy on PTB and establishes new state-of-the-art performance on the challenging ADFTD dataset (54.72% accuracy and 52.01% F1-score). Strong results on long-sequence benchmarks, such as SleepEDF, further validate its capability in modeling long-range dependencies. Moreover, MedMamba achieves a speedup of 4.6x in inference, highlighting its practicality for real-time clinical deployment. These results suggest that principle-guided state space modeling offers an effective and scalable alternative to Transformer-based approaches for medical time series analysis.

Wireless radiance field (WRF) reconstruction aims to learn a continuous, queryable representation of radio frequency characteristics over 3D space and direction, from which specific quantities, such as the spatial power spectrum (SPS) at a receiver given a transmitter position, can be predicted. While Gaussian splatting (GS)-based method has surpassed Neural Radiance Fields (NeRF)-based method for this task, existing adaptations largely transplant vision pipelines, limiting physical interpretability and accuracy. We introduce BiSplat-WRF, a planar GS framework that retains the expressiveness of 3D GS while removing unnecessary projections and incorporating global EM coupling and mutual scattering among primitives. Each primitive is a 2D planar Gaussian with 3D coordinates, rendered directly on the angular domain of the SPS. A bilinear spatial transformer (BST) aggregates inter-primitive relations on an angular grid and, via attention, captures long-range electromagnetic dependencies, thereby enforcing globally aware EM interactions that reflect the complex physics of the wireless environment. On spatial spectrum synthesis task, BiSplat-WRF surpasses NeRF-based and prior GS-based baselines with respect to the Structural Similarity Index (SSIM); comprehensive ablation studies validate the contribution of BST. We also provide a larger BiSplat-WRF+ variant that further increases SSIM at a higher computation cost, serving as a strong reference for future studies.

Voice assistants increasingly rely on Speech Language Models (SpeechLMs) to interpret spoken queries and execute complex tasks, yet existing benchmarks lack domain breadth, acoustic diversity, and compositional reasoning complexity to evaluate tool-calling performance. We introduce Audio2Tool, a large-scale dataset comprising approximately 30,000 queries designed to assess tool-calling capabilities of SpeechLMs across three primary domains: Smart Car, Smart Home, and Wearables. Our benchmark features a multi-tier complexity hierarchy, ranging from simple direct commands to complex multi-intent and needle-in-a-haystack extraction to isolate distinct failure modes. To ensure realism, we employ zero-shot voice cloning text-to-speech synthesis and diverse noise profiles to simulate in-the-wild conditions. Evaluations of state-of-the-art SpeechLMs and ASR-LLM pipelines show strong performance on simple commands but significant degradation under compositional and acoustic challenges. Code and dataset are publicly available on the project page: https://audio2tool.github.io/.

Audio deepfakes pose a significant security threat, yet current state-of-the-art (SOTA) detection systems do not generalize well to realistic in-the-wild deepfakes. We introduce a novel In-Context Learning paradigm with comparison-guidance for Audio Deepfake detection (ICLAD). The framework enables the use of audio language models (ALMs) for training-free generalization to unseen deepfakes and provides textual rationales on the detection outcome. At the core of ICLAD is a pairwise comparative reasoning strategy that guides the ALM to discover and filter hallucinations and deepfake-irrelevant acoustic attributes. The ALM works alongside a specialized deepfake detector, whereby a routing mechanism feeds out-of-distribution samples to the ALM. On in-the-wild datasets, ICLAD improves macro F1 over the specialized detector, with up to $2\times$ relative improvement. Further analysis demonstrates the flexibility of ICLAD and its potential for deployment on recent open-source ALMs.

Linear-threshold networks (LTNs) capture the mesoscale behavior of interacting populations of neurons and are of particular interest to control theorists due to their dynamical richness and relative ease of analysis. The aim of this paper is to advance the study of global asymptotic stability in LTNs with asymmetric neural interactions and heterogeneous dissipation under the structural Lyapunov diagonal stability (LDS) condition. To this end, we introduce a one-parameter family of LTNs that preserves the LDS condition and has a parameter-independent equilibrium set. In the fast limit, this family converges to a projected dynamical system (PDS), while in the slow limit, it converges to a discontinuous hard-selector system (HSS). Under LDS, we prove that the fast PDS limit is globally exponentially stable and that the HSS limit is globally asymptotically stable. This alignment suggests that the limiting systems capture essential mechanisms governing stability across the entire LTN family. Together with numerical evidence, these findings indicate that resolving stability at the fast and slow endpoints provides a promising and structurally grounded path toward establishing global stability for LTNs with biologically plausible recurrence and diagonal dissipation.