Keyword Mamba: Spoken keyword spotting with state space models

Efficient and robust Bird’s-Eye-View perception via state space models with linear complexity

L2G-Net: Local-to-global feature enhancement via cluster tokens for 3D place recognition.

X2Shape: CT-free 3D multi-organ reconstruction with biplanar X-rays.

A spatiotemporal settlement estimation framework integrating augmented clustering and state-space modeling during subway deep foundation excavation

Physics-informed continuous state space models for probabilistic imaging in complex nonlinear media: A 3d magnetotelluric application

BASSM: Blur-aware selective state space model for non-uniform motion deblurring in percutaneous spinal endoscopy

Bridging heterogeneous state space models and convolutional neural networks for efficient and robust pavement crack segmentation

Enhanced visual state space model for real-time wafer defect detection

TriSSR: Tri-expert fusion in state space models for sequential recommendation

A novel state-space model identification method from a behavioral system-theoretic perspective

Data-driven time-varying state-space modeling for model predictive temperature control of steel strip in continuous annealing lines

This paper presents a robust [Formula: see text] state-feedback controller design for a singular Takagi-Sugeno (T-S) fuzzy model of a synchronous generator, effectively addressing time delays, external disturbances, and algebraic constraints inherent to singular systems. The proposed controller employs a descriptor system formulation that captures both differential and algebraic equations, providing a more accurate representation of power system dynamics than conventional state-space models. Necessary and sufficient conditions for the existence of the [Formula: see text] controller are derived as strict linear matrix inequalities (LMIs), ensuring numerical tractability and guaranteeing closed-loop admissibility. The proposed controller, denoted as HITSFS (Descriptor-based [Formula: see text] control), is rigorously compared against RHITS (Non-descriptor [Formula: see text] control) and NFTSFS (Non-fragile saturation control) under exhaustive validation scenarios, including systematic parameter variations (minimum, nominal, maximum), measurement noise ([Formula: see text]-0.010 p.u.), time delays ([Formula: see text]-0.5 s), and distinct fault conditions (0.3-1.0 p.u.). The proposed HITSFS controller achieves the lowest ISE and peak overshoot across all states, with 24 total wins compared to only 2 for RHITS and 4 for NFTSFS. It demonstrates superior noise immunity (near-zero ISE for [Formula: see text]) and fault recovery, while consuming the least control energy (6.2949 pu[Formula: see text]s), consistently outperforming both baseline controllers across all test scenarios.

ABSTRACT Robust bearing fault diagnosis in electric motor‐driven electromechanical systems remains challenging under varying operating conditions, where changes in speed, load, and torque induce substantial distribution shifts in vibration signals. This study presents an end‐to‐end deep learning framework that integrates learnable multi‐band FIR decomposition, a Dynamic Phase Event Encoder (DPEE), and a Selective state space model–convolutional (SSM–Conv) hybrid backbone within a unified differentiable architecture. Unlike conventional approaches that mainly rely on amplitude‐ or spectrum‐oriented representations, the proposed method introduces a phase‐driven intermediate representation designed to capture both localized fault‐related irregularities and longer‐range temporal dependencies. To reflect more realistic monitoring conditions, evaluation was performed using a condition‐based leave‐one‐operating‐condition‐out protocol, ensuring strict separation between training and test conditions and reducing condition‐level data leakage. On the Case Western Reserve University (CWRU) dataset, the framework achieved consistently strong performance across unseen load conditions, while on the more challenging Paderborn dataset, the results varied depending on the severity of the condition shift. Ablation studies further supported the contribution of both the learnable FIR decomposition and the DPEE module. Additional analyses were also conducted to assess robustness and transferability. Under additive white Gaussian noise, the model remained comparatively stable at mild‐to‐moderate SNR levels but showed noticeable degradation under severe noise. Cross‐dataset experiments between CWRU and Paderborn, including target‐domain fine‐tuning with 5%, 10%, and 20% labeled target data, indicated that limited target supervision can substantially improve adaptation. Overall, the results suggest that the proposed framework is a promising approach for reliability‐oriented bearing condition monitoring under variable operating regimes.

Introduction: Declining mental wellbeing and rising suicidality are major public health concerns. Deliberate self-poisoning is a common form of suicidal behaviour in Western societies. This study examined trends, demographics, and substances involved in deliberate self-poisonings across European poison centres and assessed the feasibility of harmonised data collection. Methods: Retrospective poison centre data from 2017–2022 were collected from four European countries: national data from France, the Netherlands, and Switzerland, and from the Freiburg poison centre in Germany. Using a standardised template, cases were categorised by gender, age group, and substance. Deliberate self-poisonings cases were expressed as a proportion of all poisoning cases. Temporal trends were analysed using exponential smoothing state space models. Results Within six years, poison centres reported 1,783,858 cases, and one in ten (177,921 cases) involved deliberate self-poisonings. Rates varied across countries, ranging from 8% in France to 23% in the Netherlands. Cases with deliberate self-poisoning remained stable between 2017 and 2020, with an increase observed from 2021 to 2022. Females accounted for 70% of cases, rising to 84% among children/adolescents of 5–17 years. Rates among children/adolescents of 5–17 years increased significantly after 2020 in all countries, largely due to a strong increase among females. Substances involved varied by country, but alcohol, benzodiazepines, and antidepressants dominated in adults, while paracetamol and ibuprofen dominated among children/adolescents of 5–17 years. Discussion Common trends in deliberate self-poisoning, particularly in different age groups, are highlighted and challenges of cross-national data harmonisation in European poison centres are demonstrated. Conclusion Deliberate self-poisoning represented a substantial proportion of poisoning cases, with a sharp rise after 2020, especially among children/adolescent females using readily available over-the-counter non-prescription medicines. These findings underscore the need for targeted prevention strategies and demonstrate the value of coordinated data collection. Expanding collaboration across additional countries could enhance monitoring, strengthen trend analyses, and inform public health interventions.

ABSTRACT Introduction Model-Informed Precision Dosing (MIPD) has improved individualized therapy, but in critical illness its reliance on intermittently updated data creates a temporal mismatch between pharmacokinetic (PK) models and rapidly evolving physiology. Areas covered Synthesizing population pharmacokinetic (popPK) and data science literature, we examine the limitations of snapshot-based dosing, using vancomycin as example. We propose the Dynamic Living Digital Twin (DLDT) framework, which integrates high-frequency electronic health record data into adaptive state-space models such as Kalman filtering. Rather than adding more covariates, the DLDT reframes patient physiology as a continuously evolving latent state that can be updated using temporally dense clinical data. Methodological, infrastructural, and regulatory Software as Medical Device (SaMD) barriers are also evaluated. A non-systematic PubMed search identified relevant publications available up to March 2026. Expert opinion The next advance in precision dosing will come from temporally adaptive PK reasoning. In this paradigm, patient physiology is treated as an evolving latent state, and clinical pharmacists may increasingly interpret exposure trajectories rather than isolated dose recommendations. Although technically feasible, implementation remains constrained by data interoperability and workflow integration challenges. Clinical pharmacists may therefore increasingly act as stewards of dynamic model outputs, using anticipated exposure trajectories to preempt PK shifts.

Reconstructing gene regulatory networks (GRNs) with directionality and regulatory types is an important challenge in computational biology. Existing methods often struggle to effectively capture complex topological structures in highly skewed GRNs due to imbalances between local and global information and to the collapse of representation dimensionality. To address these challenges, we propose BMGRN, a unified framework that reconstructs directional and GRNs with regulation types by integrating bidirectional state space modeling with dual contrastive representation learning. Drawing inspiration from sequence modeling, BMGRN employs an enhanced bidirectional Mamba2 architecture to capture long-range dependencies and asymmetric regulatory interactions between genes efficiently. This design enables global information propagation while maintaining directional specificity. Furthermore, a dual contrastive learning mechanism is introduced to alleviate oversmoothing and dimensional collapse, enforcing representation uniformity and discriminability in low-connectivity scenarios. By coupling these representations with a KAN-based convolutional predictor, BMGRN adaptively learns nonlinear dependencies and regulatory modes, thereby improving its modeling capacity for the GRN inference. Experiments on multiple benchmark data sets show that BMGRN attains superior performance, demonstrating great potential for large-scale GRN inference. The code is available at https://github.com/KanZh/BMGRN.

Adaptive closed-loop neuromodulation is an emerging therapeutic paradigm for treating neurological and psychiatric diseases. However, its hardware implementation remains challenged by limited regulation accuracy, high latency & energy overhead, and poor cross-workload compatibility. To address these faced challenges, this article presents a multi-task neural processing system-on-chip (SoC) with three key technologies. First, an established multi-input multi-output linear state-space model (MIMO LSSM) with linear quadratic Gaussian (LQG) control is configured for deterministic on-chip execution to improve regulation accuracy. Second, a processing element (PE)-array-aware compact parameter-encoding scheme is proposed to reduce storage cost and memory-access overhead. Third, a mode-configurable compute fabric (MCCF) is designed to support diverse neuromodulation workloads on a unified hardware fabric. The designed SoC was fabricated in a TSMC 65nm CMOS process. Measured results and performance comparison show that it achieves a maximum energy efficiency of 1.43 TOPS/W (3.08×), a maximum area efficiency of 1.09 GOPS/mm2 (13.29×), and a peak performance of 5.12 GOPS (40.96×). Besides, the SoC has been demonstrated on the BONN and DEAP datasets, achieving accuracies of 99.18% in seizure detection and 92.3% in emotion detection. Overall, the proposed SoC offers a competitive hardware solution for adaptive closed-loop neuromodulation.

State space modeling of commensurate fractional order systems for extracting EEG rhythm

Tracking small-bodied animals in estuarine environments entails significant technological and analytical challenges. Diamond-backed terrapins are small (max 1.4kg) turtles that inhabit salt marshes of the eastern U.S. and the Gulf of Mexico. Terrapin movements have been studied with VHF radio telemetry, acoustic telemetry, and mark and recapture methods, which have indicated maximum straight-line movement distances < 10km and mean home ranges < 1 km2. We deployed 21 Argos satellite tags on adult female terrapins at two sites on Long Island, New York to better understand the spatial ecology of this imperiled species, and to test newly available tracking technology. We processed the location data three ways: (1) we used a location data filter to remove unlikely terrestrial and oceanic locations and applied a state-space model to account for Argos location errors, (2) we applied the state-space model to unfiltered data to determine the effects of not removing unlikely locations, and (3) we used only the highest quality location class 3 (LC 3) locations. We used the data resulting from each of these approaches to calculate four different movement metrics: summer home range size (95% minimum convex polygons (MCPs) and kernel density estimates (50% and 95% KDE, with both reference [href] and least squares cross validation [LSCV] bandwidths)), the total distance traveled from June to August, maximum distance traveled in one day, and daily movement rates. Home ranges estimated from the three processing techniques were similar in size and covered the same spatial areas. Estimates for total distance traveled, daily movement rates, and maximum distance traveled were similar between the state-space modeling techniques, but LC 3 estimated distances were twice as long. Movement metrics and home ranges were similar between the two study sites, despite differences in urbanization and bay size. These results suggest that most movement metrics and home range estimates are fairly insensitive to these different analytical techniques, even at relatively smaller spatial scales. Additionally, our study indicates substantially larger home ranges and longer straight-line movements than VHF telemetry or sonic tag studies, highlighting the utility of satellite tags to improve our understanding of terrapin ecology and conservation.