Papers

Ensuring the microbiological safety of large, heterogeneous water distribution systems (WDS) typically requires managing appropriate levels of disinfectant residuals including chlorine. WDS include complex fluid interactions that are nonlinear and noisy, making such maintenance a challenging problem for traditional control algorithms. This paper proposes an evolutionary framework to this problem based on neuroevolution, multi-objective optimization, and surrogate modeling. Neural networks were evolved with NEAT to inject chlorine at strategic locations in the distribution network at select times. NSGA-II was employed to optimize four objectives: minimizing the total amount of chlorine injected, keeping chlorine concentrations homogeneous across the network, ensuring that maximum concentrations did not exceed safe bounds, and distributing the injections regularly over time. Each network was evaluated against a surrogate model, i.e.\ a neural network trained to emulate EPANET, an industry-level hydraulic WDS simulator that is accurate but infeasible in terms of computational cost to support machine learning. The evolved controllers produced a diverse range of Pareto-optimal policies that could be implemented in practice, outperforming PPO, a standard reinforcement learning method. The results thus suggest a pathway toward improving urban water systems, and highlight the potential of using evolution with surrogate modeling to optimize complex real-world systems.

Objective. Reliable, continuous neural sensing on wearable edge platforms is fundamental to long-term health monitoring; however, for electroencephalography (EEG)-based sleep monitoring, dense high-frequency processing is often computationally prohibitive under tight energy budgets. Approach. To address this bottleneck, this paper proposes NeuroSleep, an integrated event-driven sensing and inference system for energy-efficient sleep staging. NeuroSleep first converts raw EEG into complementary multi-scale bipolar event streams using Residual Adaptive Multi-Scale Delta Modulation (R-AMSDM), enabling an explicit fidelity-sparsity trade-off at the sensing front end. Furthermore, NeuroSleep adopts a hierarchical inference architecture that comprises an Event-based Adaptive Multi-scale Response (EAMR) module for local feature extraction, a Local Temporal-Attention Module (LTAM) for context aggregation, and an Epoch-Leaky Integrate-and-Fire (ELIF) module to capture long-term state persistence. Main results. Experimental results using subject-independent 5-fold cross-validation on the Sleep-EDF Expanded sleep-cassette (SC) subset with single-channel EEG demonstrate that NeuroSleep achieves a mean accuracy of 74.2% with only 0.932 M parameters while reducing sparsity-adjusted effective operations by approximately 53.6% relative to dense processing. Compared to the representative dense Transformer baseline, NeuroSleep improves accuracy by 7.5% with a 45.8% reduction in computational load. Significance. By coupling neuromorphic event encoding with state-aware context modeling, NeuroSleep offers a deployment-oriented framework for single-channel sleep staging that reduces redundant high-rate processing and improves energy scalability for wearable and edge platforms.

Cross-domain image registration requires aligning images acquired under heterogeneous imaging physics, where the classical brightness constancy assumption is fundamentally violated. We formulate this problem through an image formation model I = R(s, a) + epsilon, where each observation is generated by a rendering function R acting on domain-invariant scene structure s and domain-specific appearance statistics a. Registration then reduces to an inverse rendering problem: given observations from two domains, recover the shared structure and re-render it under the target appearance to obtain the registered output. We instantiate this framework as SAS-Net (Scene-Appearance Separation Network), where instance normalization implements the structure-appearance decomposition and Adaptive Instance Normalization (AdaIN) realizes the differentiable forward renderer. A scene consistency loss enforces geometric correspondence in the factorized latent space. Experiments on EuroSAT-Reg-256 (satellite remote sensing) and FIRE-Reg-256 (retinal fundus) demonstrate state-of-the-art performance across heterogeneous imaging domains. SAS-Net (3.35M parameters) achieves 89 FPS on an RTX 5090 GPU. Code: https://github.com/D-ST-Sword/SAS-Net.

Fifth-generation (5G) wireless networks introduce new architectural paradigms, spectrum usage models, and optimization challenges to support enhanced mobile broadband, massive machine-type communications, and ultra-reliable low-latency communications. This survey provides a comprehensive overview of key technologies and design challenges in 5G systems, with a focus on spectrum coexistence and interference management, network dimensioning and planning, cell-free massive MIMO architectures, fronthaul-aware user management, and power allocation strategies. Representative analytical, simulation-based, and optimization-driven approaches are reviewed, fundamental trade-offs are highlighted, and open research challenges relevant to 5G-Advanced and beyond are identified.

Neuroimaging studies of psychiatric disorders often correlate imaging patterns with diagnostic labels or composite symptom scores, yielding diffuse associations that obscure underlying mechanisms. We instead seek to identify root-causal maps – localized BOLD disturbances that initiate pathological cascades – and to link them selectively to symptom dimensions. We introduce a bilevel structural causal model that connects between-subject symptom structure to within-subject resting-state fMRI via independent latent sources with localized direct effects. Based on this model, we develop SOURCE (Symptom-Oriented Uncovering of Root-Causal Elements), a procedure that links interpretable symptom axes to a parsimonious set of localized drivers. Experiments show that SOURCE recovers localized maps consistent with root-causal BOLD drivers and increases interpretability and anatomical specificity relative to existing comparators.

Generative AI (GenAI) has transformed applications in natural language processing and content creation, yet centralized inference remains hindered by high latency, limited customizability, and privacy concerns. Deploying large models (LMs) in mobile edge networks emerges as a promising solution. However, it also poses new challenges, including heterogeneous multi-modal LMs with diverse resource demands and inference speeds, varied prompt/output modalities that complicate orchestration, and resource-limited infrastructure ill-suited for concurrent LM execution. In response, we propose a Multi-Agentic AI framework for latency- and fairness-aware multi-modal LM inference in mobile edge networks. Our solution includes a long-term Global Planning Agent, a short-term Prompt Scheduling Agent, and multiple on-node LM Deployment Agents, all powered by foundation language models. These agents cooperatively optimize prompt routing and LM deployment through natural language reasoning over runtime telemetry and historical experience. To evaluate its performance, we further develop a city-wide testbed that supports network monitoring, containerized LM deployment, intra-server resource management, and inter-server communications. Experiments demonstrate that our solution reduces average end-to-end latency by 42.9% compared with a state-of-the-art online-learning baseline and by 80.6% compared with the full-activation baseline, while maintaining a mean service-type Jain's index of 0.925. Moreover, our solution adapts quickly without fine-tuning, offering a generalizable solution for optimizing GenAI services in edge environments.

LLMs have shown strong in-context learning (ICL) abilities, but have not yet been extended to signal processing systems. Inspired by their design, we have proposed for the first time ICL using transformer models applicable to motor feedforward control, a critical task where classical PI and physics-based methods struggle with nonlinearities and complex load conditions. We propose a transformer based model architecture that separates signal representation from system behavior, enabling both few-shot finetuning and one-shot ICL. Pretrained on a large corpus of synthetic linear and nonlinear systems, the model learns to generalize to unseen system dynamics of real-world motors only with a handful of examples. In experiments, our approach generalizes across multiple motor load configurations, transforms untuned examples into accurate feedforward predictions, and outperforms PI controllers and physics-based feedforward baselines. These results demonstrate that ICL can bridge synthetic pretraining and real-world adaptability, opening new directions for data efficient control of physical systems.

This paper proposes a neural-network-assisted deformable matched filtering (DMF) framework for carrier-less amplitude and phase (CAP) modulation operating under bandwidth-limited channel conditions. Instead of replacing the analytically derived CAP matched filter, the proposed receiver learns a residual deformation of the nominal matched filter based on a compact set of physically motivated signal features extracted from the received waveform. A total of 16 time-domain, frequency-domain, and memory-related features are used to provide a low-dimensional representation of bandwidth-induced pulse distortion. These features are mapped by a fully connected neural network to complex-valued matched filter coefficients, enabling adaptive pulse-shape compensation prior to symbol-rate sampling. The network is trained end-to-end using a differentiable loss function based on error vector magnitude (EVM). Experimental results obtained using a hardware-in-the-loop CAP transmission system demonstrate that the proposed DMF significantly outperforms conventional fixed matched filtering under severe bandwidth constraints, without requiring decision feedback or increasing receiver latency.

X-ray microscopy (XRM) is commonly used to obtain three-dimensional information on internal microstructure, but the imaging pipeline introduces noise, redundancy and information loss at multiple stages. This paper treats the XRM workflow as an information-processing system acting on a finite information budget. Using entropy, mutual information and Kullback-Leibler divergence, we quantify how acquisition, denoising, alignment, sparse-angle sampling, dose variation and reconstruction reshape the statistical structure of projection data and reconstructed volumes. Case studies based on the Walnut 1 dataset illustrate how these processes redistribute information and impose bottlenecks. We summarise the workflow using a unified information budget and show that mutual information provides a reconstruction-agnostic indicator of fidelity, supporting quantitative comparison and optimisation of XRM protocols, particularly under low-dose or time-constrained conditions

Ultrasound imaging is widely used in noninvasive medical diagnostics due to its efficiency, portability, and avoidance of ionizing radiation. However, its utility is limited by the quality of the signal. Signal-dependent speckle noise, signal-independent sensor noise, and non-uniform spatial blurring caused by the transducer and modeled by the point spread function (PSF) degrade the image quality. These degradations challenge conventional image restoration methods, which assume simplified noise models, and highlight the need for specialized algorithms capable of effectively reducing the degradations while preserving fine structural details. We propose DEMIX, a novel dual-encoder denoising framework with a masked gated fusion mechanism, for denoising ultrasound images degraded by mixed noise and further degraded by PSF-induced distortions. DEMIX is inspired by diffusion models and is characterized by a forward process and a deterministic reverse process. DEMIX adaptively assesses the different noise components, disentangles them in the latent space, and suppresses these components while compensating for PSF degradations. Extensive experiments on two ultrasound datasets, along with a downstream segmentation task, demonstrate that DEMIX consistently outperforms state-of-the-art baselines, achieving superior noise suppression and preserving structural details. The code will be made publicly available.

Early clinical assessment of Alzheimer's disease relies on behavior scores that measure a subject's language, memory, and cognitive skills. On the medical imaging side, functional magnetic resonance imaging has provided invaluable insights into the neural pathways underlying Alzheimer's disease. While prior studies have used resting-state functional MRI by extracting functional connectivity matrices, these approaches neglect the temporal dynamics inherent in functional data. In this work, we present a deep state space modeling framework that directly leverages the blood-oxygenation-level-dependent time series to learn a sparse collection of brain regions to predict behavior scores. Our model extracts temporal features that encapsulate nuanced patterns of intrinsic brain activity, thereby enhancing predictive performance compared to traditional connectivity methods. We identify specific brain regions that are most predictive of cognitive impairment through experiments on data provided by the Michigan Alzheimer's Disease Research Center, providing new insights into the neural substrates of early Alzheimer's pathology. These findings have important implications for the possible development of risk monitoring and intervention strategies in Alzheimer's disease.

The inherently complex-valued nature of Polarimetric SAR data necessitates using specialized algorithms capable of directly processing complex-valued representations. However, this aspect remains underexplored in the deep learning community, with many studies opting to convert complex signals into the real domain before applying conventional real-valued models. In this work, we leverage complex-valued neural networks and investigate the performance of complex-valued Convolutional AutoEncoders. We show that these networks can effectively compress and reconstruct fully polarimetric SAR data while preserving essential physical characteristics, as demonstrated through Pauli, Krogager, and Cameron coherent decompositions, as well as the non-coherent $H-\alpha$ decomposition. Finally, we highlight the advantages of complex-valued neural networks over their real-valued counterparts. These insights pave the way for developing robust, physics-informed, complex-valued generative models for SAR data processing.

This paper develops a real-time, search-based aircraft contingency landing planner that minimizes traffic disruptions while accounting for ground risk. The airspace model captures dense air traffic departure and arrival flows, helicopter corridors, and prohibited zones and is demonstrated with a Washington, D.C., area case study. Historical Automatic Dependent Surveillance-Broadcast (ADS-B) data are processed to estimate air traffic density. A low-latency computational geometry algorithm generates proximity-based heatmaps around high-risk corridors and restricted regions. Airspace risk is quantified as the cumulative exposure time of a landing trajectory within congested regions, while ground risk is assessed from overflown population density to jointly guide trajectory selection. A landing site selection module further mitigates disruption to nominal air traffic operations. Benchmarking against minimum-risk Dubins solutions demonstrates that the proposed planner achieves lower joint risk and reduced airspace disruption while maintaining real-time performance. Under airspace-risk-only conditions, the planner generates trajectories within an average of 2.9 seconds on a laptop computer. Future work will incorporate dynamic air traffic updates to enable spatiotemporal contingency landing planning that minimizes the need for real-time traffic rerouting.

This paper presents the design and implementation of data-driven optimal derivative feedback controllers for an active magnetic levitation system. A direct, model-free control design method based on the reinforcement learning framework is compared with an indirect optimal control design derived from a numerically identified mathematical model of the system. For the direct model-free approach, a policy iteration procedure is proposed, which adds an iteration layer called the epoch loop to gather multiple sets of process data, providing a more diverse dataset and helping reduce learning biases. This direct control design method is evaluated against a comparable optimal control solution designed from a plant model obtained through the combined Dynamic Mode Decomposition with Control (DMDc) and Prediction Error Minimization (PEM) system identification. Results show that while both controllers can stabilize and improve the performance of the magnetic levitation system when compared to controllers designed from a nominal model, the direct model-free approach consistently outperforms the indirect solution when multiple epochs are allowed. The iterative refinement of the optimal control law over the epoch loop provides the direct approach a clear advantage over the indirect method, which relies on a single set of system data to determine the identified model and control.

Realistic sound propagation is essential for immersion in a virtual scene, yet physically accurate wave-based simulations remain computationally prohibitive for real-time applications. Wave coding methods address this limitation by precomputing and compressing impulse responses of a given scene into a set of scalar acoustic parameters, which can reach unmanageable sizes in large environments with many source-receiver pairs. We introduce Reciprocal Latent Fields (RLF), a memory-efficient framework for encoding and predicting these acoustic parameters. The RLF framework employs a volumetric grid of trainable latent embeddings decoded with a symmetric function, ensuring acoustic reciprocity. We study a variety of decoders and show that leveraging Riemannian metric learning leads to a better reproduction of acoustic phenomena in complex scenes. Experimental validation demonstrates that RLF maintains replication quality while reducing the memory footprint by several orders of magnitude. Furthermore, a MUSHRA-like subjective listening test indicates that sound rendered via RLF is perceptually indistinguishable from ground-truth simulations.

This paper extends an acoustic feedback cancellation system by incorporating multiple decorrelation methods. The baseline system is based on a frequency-domain Kalman filter implemented in a multi-delay structure. The proposed extensions include a variable time delay line, prediction, distortion compensation, and a simplified reverberation model. Each extension is analyzed, and a practical parameter range is defined. While existing literature often focuses on a single extension, such as prediction, to describe an optimal system, this work demonstrates that each individual extension contributes to performance improvements. Furthermore, the combination of all proposed extensions results in a superior system. The evaluation is conducted using publicly available datasets, with performance assessed through system distance metrics and the objective speech quality measure PSEQ.

The advancement of machine learning in audio analysis has opened new possibilities for technology-enhanced music education. This paper introduces a framework for automatic singing mistake detection in the context of music pedagogy, supported by a newly curated dataset. The dataset comprises synchronized teacher learner vocal recordings, with annotations marking different types of mistakes made by learners. Using this dataset, we develop different deep learning models for mistake detection and benchmark them. To compare the efficacy of mistake detection systems, a new evaluation methodology is proposed. Experiments indicate that the proposed learning-based methods are superior to rule-based methods. A systematic study of errors and a cross-teacher study reveal insights into music pedagogy that can be utilised for various music applications. This work sets out new directions of research in music pedagogy. The codes and dataset are publicly available.

AI music generators have advanced to the point where their outputs are often indistinguishable from human compositions. While detection methods have emerged, they are typically designed and validated in music streaming contexts with clean, full-length tracks. Broadcast audio, however, poses a different challenge: music appears as short excerpts, often masked by dominant speech, conditions under which existing detectors fail. In this work, we introduce AI-OpenBMAT, the first dataset tailored to broadcast-style AI-music detection. It contains 3,294 one-minute audio excerpts (54.9 hours) that follow the duration patterns and loudness relations of real television audio, combining human-made production music with stylistically matched continuations generated with Suno v3.5. We benchmark a CNN baseline and state-of-the-art SpectTTTra models to assess SNR and duration robustness, and evaluate on a full broadcast scenario. Across all settings, models that excel in streaming scenarios suffer substantial degradation, with F1-scores dropping below 60% when music is in the background or has a short duration. These results highlight speech masking and short music length as critical open challenges for AI music detection, and position AI-OpenBMAT as a benchmark for developing detectors capable of meeting industrial broadcast requirements.

This paper investigates the role of large language models (LLMs) in sixth-generation (6G) Internet of Things (IoT) networks and proposes a prompt-engineering-based real-time feedback and verification (PE-RTFV) framework that perform physical-layer's optimization tasks through an iteratively process. By leveraging the naturally available closed-loop feedback inherent in wireless communication systems, PE-RTFV enables real-time physical-layer optimization without requiring model retraining. The proposed framework employs an optimization LLM (O-LLM) to generate task-specific structured prompts, which are provided to an agent LLM (A-LLM) to produce task-specific solutions. Utilizing real-time system feedback, the O-LLM iteratively refines the prompts to guide the A-LLM toward improved solutions in a gradient-descent-like optimization process. We test PE-RTFV approach on wireless-powered IoT testbed case study on user-goal-driven constellation design through semantically solving rate-energy (RE)-region optimization problem which demonstrates that PE-RTFV achieves near-genetic-algorithm performance within only a few iterations, validating its effectiveness for complex physical-layer optimization tasks in resource-constrained IoT networks.

In the context of civilian and military communications, anti-jamming techniques are essential to ensure information integrity in the presence of malicious interference. A conventional time-domain approach relies on computing the Wiener interpolation filter to estimate and suppress the jamming waveform from the received samples. It is widely acknowledged that this method is effective for protecting wideband systems against narrowband interference. In this work, this paradigm is questioned through the design of a $K$-tone jamming waveform that is intrinsically difficult to estimate assuming a $L$-tap Wiener interpolation filter. This design relies on an optimization procedure that maximizes the analytical Bayesian mean squared error associated with the jamming waveform estimate. Additionally, an analytical proof is provided showing that a multi-tone jamming waveform composed of $L/2+1$ tones is sufficient to render the Wiener-filter-based anti-jamming module completely ineffective. The analytical results are validated through Monte Carlo simulations assuming both perfect knowledge and practical estimates of the correlation functions of the received signal.

Under the ideal assumption of uniform propagation, cell-free massive MIMO (CF-mMIMO) provides uniformly high throughput over the network by effectively surrounding each user with its serving access point (AP) set. However, in realistic non-uniform urban propagation environments, it is difficult to consistently select good limited serving AP sets, resulting in significantly degraded throughput, especially for the worst-served (formerly "cell-edge") users. To restore the uniformly good performance of scalable CF-mMIMO in realistic urban networks, we formulate a novel multi-objective optimization problem to jointly achieve high throughput by maximizing the sum data rate, uniform throughput by maximizing Jain's fairness index of the throughput per user, and scalability by minimizing the serving AP set size. We then propose the UnifSrv AP selection algorithms to solve this optimization problem, consisting of a deep reinforcement learning (DRL)-based algorithm UnifSrv-DRL and a heuristic algorithm UnifSrv-heu. We conduct a comprehensive performance evaluation of scalable CF-mMIMO under realistic urban network distributions, propagation, and mobility patterns. Our results show that UnifSrv significantly outperforms the prior benchmark AP selection schemes, and for the first time achieves uniformly high throughput of CF-mMIMO under non-uniform urban propagation. Importantly, our heuristic algorithm achieves equivalent throughput to our DRL one, but with orders of magnitude lower complexity. We thus for the first time propose a practical AP selection algorithm that makes CF-mMIMO viable in realistic urban networks.

Fetal echocardiography is essential for detecting congenital heart disease (CHD), facilitating pregnancy management, optimized delivery planning, and timely postnatal interventions. Among standard imaging planes, the four-chamber view (4CV) provides important information for CHD diagnosis, where clinicians carefully inspect the end-diastolic (ED) and end-systolic (ES) phases to evaluate cardiac structure and motion. Automated detection of these cardiac phases is thus a critical component towards fully automated CHD analysis. However, existing approaches typically rely on manual annotation of ED/ES frames, which is labour-intensive and time-consuming. We present ORBIT (Orientation-Robust Beat Inference from Trajectories), a self-supervised framework that identifies cardiac phases without manual annotations under various fetal heart orientation. ORBIT employs registration as self-supervision task and learns a latent motion trajectory of cardiac deformation, whose turning points capture transitions between cardiac relaxation and contraction, enabling accurate and orientation-robust localization of ED and ES frames across diverse fetal positions. Trained exclusively on normal fetal echocardiography videos, ORBIT achieves consistent performance on both normal (mean absolute error 1.9 frames for ED and 1.6 for ES) and CHD cases (mean absolute error 2.4 frames for ED and 2.1 for ES), outperforming existing annotation-free approaches constrained by fixed orientation assumptions. These results highlight the potential of ORBIT to facilitate robust cardiac phase detection directly from 4CV fetal echocardiography.

In this work, we analyze a multi-functional reconfigurable intelligent surface (MF-RIS)-enabled radar and communication coexistence (RCC) system, detailing the key aspects of its phase synthesis codebook generation and the implemented localization algorithm for real-time user tracking based on density-based spatial clustering of applications with noise (DBSCAN), which features a Kalman filter for the prediction of user mobility. We derived a 3GPP-compatible radar cross-section (RCS) and re-radiation pattern-based channel model for the described MF-RIS system, supplementing it with channel measurements. We obtained large and small-scale characteristics, including path loss, shadow fading, Rician K-factor, cluster powers, and RMS delay spread. The study finds that Sub-6 GHz indoor propagation is largely free of blind spots, even with a blocked line-of-sight (LoS) path. Therefore, the proposed channel model includes non-line-of-sight (NLoS) paths, including the ones created by the MF-RIS. We also performed an experimental evaluation of the channel throughput in a fifth generation (5G) new radio (NR) single user multiple-input-multiple-output (SU-MIMO) system, reporting a 74% reduction in throughput variance and a 12.5% sum-rate improvement within the MF-RIS near-field compared to the no-RIS setup. This result shows that the MF-RIS can minimize delay spread and increase the coherence bandwidth by creating virtual-LoS (vLoS) path for the moving user, thereby effectively hardening wireless MIMO channels.

The computation of positioning, navigation and timing (PNT) via signal of opportunity (SOP), where signals originally transmitted for communication, such as 5G, Wi-Fi, or DVB-S, are exploited due to their ubiquity and spectral characteristics, is an emerging research field. However, relying on these signals presents challenges, including limited knowledge of the signal modulation and the need to identify recurring sequences for correlation. We offer a guide to implement a receiver capable of capturing broadband downlink Ku-band signals from low Earth orbit (LEO) satellites (e.g., Starlink and OneWeb) and estimating the recurring symbols for SOP measurements. The methodology integrates recent approaches in the literature, highlighting the most effective aspects while guiding the replication of experiments even under limitations on the front-end gain and bandwidth. Using the proposed model, we can identify recurring symbols transmitted by Starlink satellites, which are then used to collect Doppler shift measurements over a 600 s interval. A position, velocity, and time (PVT) solution is also computed via least squares (LS), which achieves a positioning error of approximately 268 m after a post-fit refinement.

Spoken content, such as online videos and podcasts, often spans multiple topics, which makes automatic topic segmentation essential for user navigation and downstream applications. However, current methods do not fully leverage acoustic features, leaving room for improvement. We propose a multi-modal approach that fine-tunes both a text encoder and a Siamese audio encoder, capturing acoustic cues around sentence boundaries. Experiments on a large-scale dataset of YouTube videos show substantial gains over text-only and multi-modal baselines. Our model also proves more resilient to ASR noise and outperforms a larger text-only baseline on three additional datasets in Portuguese, German, and English, underscoring the value of learned acoustic features for robust topic segmentation.

The development of efficient and robust dynamic models is fundamental in the field of systems and control engineering. In this paper, a new formulation for the dynamic model of nonlinear mechanical systems, that can be applied to different automotive and robotic case studies, is proposed, together with a modeling procedure allowing to automatically obtain the model formulation. Compared with the Euler-Lagrange formulation, the proposed model is shown to give superior performances in terms of robustness against measurement noise for systems exhibiting dependence on some external variables, as well as in terms of execution time when computing the inverse dynamics of the system.

In contact-rich tasks, while position trajectories are often easy to obtain, appropriate force commands are typically unknown. Although it is conceivable to generate force commands using a pretrained foundation model such as Vision-Language-Action (VLA) models, force control is highly dependent on the specific hardware of the robot, which makes the application of such models challenging. To bridge this gap, we propose a force generative model that estimates force commands from given position trajectories. However, when dealing with unseen position trajectories, the model struggles to generate accurate force commands. To address this, we introduce a feedback control mechanism. Our experiments reveal that feedback control does not converge when the force generative model has memory. We therefore adopt a model without memory, enabling stable feedback control. This approach allows the system to generate force commands effectively, even for unseen position trajectories, improving generalization for real-world robot writing tasks.

Precise magnetic field modeling is fundamental to the closed-loop control of electromagnetic navigation systems (eMNS) and the analytical Multipole Expansion Model (MPEM) is the current standard. However, the MPEM relies on strict physical assumptions regarding source symmetry and isolation, and requires optimization-based calibration that is highly sensitive to initialization. These constraints limit its applicability to systems with complex or irregular coil geometries. This work introduces an alternative modeling paradigm based on multi-layer perceptrons that learns nonlinear magnetic mappings while strictly preserving the linear dependence on currents. As a result, the field models enable fast, closed-form minimum-norm inversion with evaluation times of approximately 1 ms, which is critical for high-bandwidth magnetic control. For model training and evaluation we use large-scale, high-density datasets collected from the research-grade OctoMag and clinical-grade Navion systems. Our results demonstrate that data-driven models achieve predictive fidelity equivalent to the MPEM while maintaining comparable data efficiency, and we further assess their suitability for real-time magnetic control in a closed-loop tracking experiment running at 100 Hz. Furthermore, we demonstrate that straightforward design choices effectively eliminate spurious workspace ill-conditioning frequently reported in MPEM-based calibration. To facilitate future research, we release the complete codebase and datasets open source.

Speech tokenizers are foundational to speech language models, yet existing approaches face two major challenges: (1) balancing trade-offs between encoding semantics for understanding and acoustics for reconstruction, and (2) achieving low bit rates and low token rates. We propose Speech Diffusion Tokenizer (SiTok), a diffusion autoencoder that jointly learns semantic-rich representations through supervised learning and enables high-fidelity audio reconstruction with diffusion. We scale SiTok to 1.6B parameters and train it on 2 million hours of speech. Experiments show that SiTok outperforms strong baselines on understanding, reconstruction and generation tasks, at an extremely low token rate of $12.5$ Hz and a bit-rate of 200 bits-per-second.

This work develops a theoretical framework for safety controller synthesis in discrete-time stochastic nonlinear polynomial systems subject to time-invariant delays (dt-SNPS-td). While safety analysis of stochastic systems using control barrier certificates (CBC) has been widely studied, developing safety controllers for stochastic systems with time delays remains largely unexplored. The main challenge arises from the need to account for the influence of delayed components when formulating and enforcing safety conditions. To address this, we employ Krasovskii control barrier certificates, which extend the conventional CBC framework by augmenting it with an additional summation term that captures the influence of delayed states. This formulation integrates both the current and delayed components into a unified barrier structure, enabling safety synthesis for stochastic systems with time delays. The proposed approach synthesizes safety controllers under input constraints, offering probabilistic safety guarantees robust to such delays: it ensures that all trajectories of the dt-SNPS-td remain within the prescribed safe region while fulfilling a quantified probabilistic bound. To achieve this, our method reformulates the safety constraints as a sum-of-squares optimization program, enabling the systematic construction of Krasovskii CBC together with their associated safety controllers. We validate the proposed framework through three case studies, including two physical systems, demonstrating its effectiveness and practical applicability.