Neural networks encode inputs as high-dimensional vectors, known as representations, that capture how models process data by encoding task-relevant structure and semantics. Representation alignment refers to the degree to which different models, layers, or training conditions produce similar representations for the same inputs, with important implications for model interpretation, selection, and robustness analysis. Existing approaches to measure alignment primarily rely on geometric properties, such as neighborhood and cluster similarity, offering limited insight into the global organization of representations. In this work, we present TopoAlign, a topology-aware framework for visually comparing model representations from a structural perspective. Leveraging mapper graphs from topological data analysis, TopoAlign jointly analyzes graphs constructed from representations of shared inputs across different models or layers. The framework supports a top-down comparative workflow: it first performs global structure alignment via joint force-directed optimization to produce coordinated graph layouts; it then identifies local correspondences through automated detection of structurally matching regions, visualized with Bubble Sets; and finally it enables fine-grained pattern inspection through motif-based queries and membrane-inspired visualizations. We demonstrate TopoAlign through case studies on language and multimodal models, complemented by expert feedback. Our results show that TopoAlign provides meaningful insights into representation structure and alignment from a topological perspective.

Digital histopathology whole slide images (WSIs) provide gigapixel-scale high-resolution images that are highly useful for disease diagnosis. However, digital histopathology image analysis faces significant challenges due to the limited training labels, since manually annotating specific regions or small patches cropped from large WSIs requires substantial time and effort. Weakly supervised multiple instance learning (MIL) offers a practical and efficient solution by requiring only bag-level (slide-level) labels, while each bag typically contains multiple instances (patches). Most MIL methods directly use frozen image patch features generated by various image encoders as inputs and primarily focus on feature aggregation. However, feature representation learning for encoder pretraining in MIL settings has largely been neglected.
In our work, we propose a novel feature representation learning framework called weakly supervised contrastive learning (WeakSupCon) that incorporates bag-level label information during training. Our method does not rely on instance-level pseudo-labeling, yet it effectively separates patches with different labels in the feature space. Experimental results demonstrate that the image features generated by our WeakSupCon method lead to improved downstream MIL performance compared to self-supervised contrastive learning approaches in three datasets.

Embolization has evolved from a purely mechanical occlusion technique to a multifunctional platform that supports imaging enhancement and therapeutic functions, including localized drug delivery, immunotherapy, and vascular remodeling. Although reviews on embolic agents have been published, the therapeutic performance of embolic platforms has not yet been systematically examined from a materials perspective linking material design and formation mechanisms to drug loading, release behavior, and therapeutic outcomes. In this review, we discuss the embolic system design principles and mechanisms underlying both clinically used and emerging embolic agents, emphasizing liquid/gel systems and microsphere (MS)-based platforms with integrated therapeutic functionality. We highlight how material design governs catheter delivery, vascular penetration, occlusion stability, and controlled drug release, key factors that govern the performance of all embolic categories. For liquid/gel embolics, we summarize clinical formulations alongside their reported outcomes, and we review emerging systems according to their mechanisms of solidification and biological interaction, including thermoresponsive gels, chemically triggered networks, complex coacervates, and shear-thinning nanocomposites. For MS embolics, we summarize clinically used materials and discuss emerging systems, focusing on how polymer chemistry, cross-linking, and network architecture regulate drug loading and release. Finally, we discuss key translational challenges in the emerging embolic systems and highlight opportunities for future embolic platforms that enable more precise and durable therapeutic control.

Pigmentation is a defining feature of melanocytic cells, and melanogenesis, the biosynthesis and compartmentalization of pigment into lysosome-like melanosomes, is tightly linked to cell state and developmental programs. However, the optical complexity of melanin-rich organelles complicates efforts to study melanogenesis in live cells. Traditional optical measurements conflate pigment absorption with light scattering from intracellular granularity, limiting biological insight. Our pipeline, QUILPEN (QUantitative Imaging of Label-free Pigment-associated ENtities), uses a custom LED-array microscope to independently quantify transmitted, scattered, and absorbed light in live, unlabeled melanocytic cells. QUILPEN builds off our labs prior work in LED-based DPC and Quadrant Dark Field, or QDF, which reduce scatterrelated edge-effects. Our imaging workflow produces temporally resolved, multi-channel images to characterize the biophysical signature at single-cell resolution. Using melanoma cell lines with diverse pigmentation states, we show that acral melanoma cells display tightly correlated scatter and absorption signals, consistent with pigment-laden melanosomes driving both granularity and light absorption. By analyzing well-characterized melanoma models and perturbing pigment synthesis, we define how QUILPEN signals relate to melanin production, melanosome abundance, and general organelle content. Because QUILPEN enables longitudinal tracking without labels, it allows direct measurement of melanosome dynamics within individual cells and lineages. By separately quantifying absorption and scatter, QUILPEN reveals the sources of optical heterogeneity in melanocytic cells and links optical signatures to underlying cell physiology.

Motivated by the need for efficient estimation of conditional expectations, we consider a least-squares function approximation problem with heavily polluted data. Existing methods that are powerful in the small noise regime are suboptimal when large noise is present. We propose a hybrid approach that combines Christoffel sampling with certain types of optimal experimental design to address this issue. We show that the proposed algorithm enjoys appropriate optimality properties for both sample point generation and noise mollification, leading to improved computational efficiency and sample complexity compared to existing methods. We also extend the algorithm to convex-constrained settings with similar theoretical guarantees. When the target function is defined as the expectation of a random field, we extend our approach to leverage adaptive random subspaces and establish results on the approximation capacity of the adaptive procedure. Our theoretical findings are supported by numerical studies on both synthetic data and on a more challenging stochastic simulation problem in computational finance.

As the demand for intelligent systems grows, leveraging edge learning and autonomic self-management offers significant benefits for supporting real-time data analysis and resource management in edge environments. We describe and evaluate four distinct task allocation scenarios to demonstrate the autonomics for edge resources management: random execution, autonomic broker-based scheduling, priority-driven execution, and energy-aware allocation. Our experiments reveal that while prioritization-based scheduling minimizes execution times by aligning with task criticality, the energy-aware approach presents a sustainable alternative. This method dynamically adapts task execution based on renewable energy availability, promoting environmentally conscious energy management without compromising operational efficiency. By harnessing renewable energy signals, our findings highlight the potential of edge autonomics to achieve a balance between performance, resource optimization and sustainability. This work demonstrates how intelligent edge-cloud integration can foster resilient smart building infrastructures that meet the challenges of modern computing paradigms.

Purpose: Transcatheter edge-to-edge repair (TEER) and annuloplasty devices are increasingly used to treat mitral valve regurgitation, yet their mechanical effects and interactions remain poorly understood. This study aimed to establish an open-source finite element modeling (FEM) framework for simulating patient-specific mitral valve repairs and to evaluate how TEER, annuloplasty, and combined strategies influence leaflet coaptation and valve mechanics. Methods: Four G4 MitraClip geometries were modeled and deployed in FEBio to capture leaflet grasp and subsequent clip-leaflet motion under physiologic pressurization. CardioBand annuloplasty was simulated by reducing annular circumference via displacement-controlled boundary conditions, and Mitralign suture annuloplasty was modeled using discrete nodal constraints. Simulations were performed for prolapse and dilated annulus cases. Valve competence (regurgitant orifice area, ROA), coaptation/contact area (CA), and leaflet stress and strain distributions were quantified. Results: In prolapse, TEER restored coaptation but increased leaflet stresses, whereas band and suture annuloplasty produced distinct valve morphologies with lower stress distributions. In dilation, TEER alone left residual regurgitation, while annuloplasty improved closure. Combined TEER & band annuloplasty minimized ROA, maximized CA, and reduced stresses relative to TEER alone, though stresses remained higher than annuloplasty alone. Conclusion: This study establishes a reproducible, open-source FEM framework for simulating transcatheter TEER and annuloplasty repairs, with the potential to be extended beyond the mitral valve. By quantifying the mechanical trade-offs of TEER, suture annuloplasty, band annuloplasty, and their combinations, this methodology highlights the potential of virtual repair to guide patient selection and optimize surgical planning.

This paper presents a novel end-to-end framework for closed-form computation and visualization of critical point uncertainty in 2D uncertain scalar fields. Critical points are fundamental topological descriptors used in the visualization and analysis of scalar fields. The uncertainty inherent in data (e.g., observational and experimental data, approximations in simulations, and compression), however, creates uncertainty regarding critical point positions. Uncertainty in critical point positions, therefore, cannot be ignored, given their impact on downstream data analysis tasks. In this work, we study uncertainty in critical points as a function of uncertainty in data modeled with probability distributions. Although Monte Carlo (MC) sampling techniques have been used in prior studies to quantify critical point uncertainty, they are often expensive and are infrequently used in production-quality visualization software. We, therefore, propose a new end-to-end framework to address these challenges that comprises a threefold contribution. First, we derive the critical point uncertainty in closed form, which is more accurate and efficient than the conventional MC sampling methods. Specifically, we provide the closed-form and semianalytical (a mix of closed-form and MC methods) solutions for parametric (e.g., uniform, Epanechnikov) and nonparametric models (e.g., histograms) with finite support. Second, we accelerate critical point probability computations using a parallel implementation with the VTK-m library, which is platform portable. Finally, we demonstrate the integration of our implementation with the ParaView software system to demonstrate near-real-time results for real datasets.

Diffusion has emerged as a powerful framework for generative modeling, achieving remarkable success in applications such as image and audio synthesis. Enlightened by this progress, we propose a novel diffusion-based approach for symbolic regression. We construct a random mask-based diffusion and denoising process to generate diverse and high-quality equations. We integrate this generative processes with a token-wise Group Relative Policy Optimization (GRPO) method to conduct efficient reinforcement learning on the given measurement dataset. In addition, we introduce a long short-term risk-seeking policy to expand the pool of top-performing candidates, further enhancing performance. Extensive experiments and ablation studies have demonstrated the effectiveness of our approach.

We present a unified interpolation scheme that combines compactly-supported positive-definite kernels and multivariate polynomials. This unified framework generalizes interpolation with compactly-supported kernels and also classical polynomial least squares approximation. To facilitate the efficient use of this unified interpolation scheme, we present specialized numerical linear algebra procedures that leverage standard matrix factorizations. These procedures allow for efficient computation and storage of the unified interpolant. We also present a modification to the numerical linear algebra that allows us to generalize the application of the unified framework to target functions on manifolds with and without boundary. Our numerical experiments on both Euclidean domains and manifolds indicate that the unified interpolant is superior to polynomial least squares for the interpolation of target functions in settings with boundaries.

In many applications, one seeks to approximate integration against a positive measure of interest by a positive discrete measure: a numerical quadrature rule with positive weights. One common desired discretization property is moment preservation over a finite dimensional function space, e.g., bounded-degree polynomials. Carathéodory's theorem asserts that if there is any finitely supported quadrature rule with more nodes than the dimension of the given function space, one can form a smaller (and hence more efficient) positive, nested, quadrature rule that preserves the moments of the original rule.

We describe an efficient streaming procedure for Carathéodory-Steinitz pruning, a numerical procedure that implements Carathéodory's theorem for this measure compression. The new algorithm makes use of Givens rotations and on-demand storage of arrays to successfully prune very large rules whose storage complexity only depends on the dimension of the function space. This approach improves on a naive implementation of Carathéodory-Steinitz pruning whose runtime and storage complexity are quadratic and linear, respectively, in the size of the original measure. We additionally prove mathematical stability properties of our method with respect to a set of admissible, total-variation perturbations of the original measure. Our method is compared to two alternate approaches with larger storage requirements: non-negative least squares and linear programming, and we demonstrate comparable runtimes, with improved stability and storage robustness. Finally, we demonstrate practical usage of this algorithm to generate quadrature for discontinous Galerkin finite element simulations on cut-cell meshes.

We investigate model reduction of parametric linear time-invariant (LTI) dynamical systems. When posed in the frequency domain, this problem can be formulated as seeking a low-order rational function approximation of a high-order rational function. We propose to use a standard reduced basis method (RBM) to construct this low-order rational function. Algorithmically, this procedure is an iterative greedy approach, where the greedy objective is evaluated through an error estimator that exploits the linearity of the frequency domain representation. The greedy framework is motivated through theoretical results of rational approximability of functions. This framework provides a principled approach to rational compression of high-order rational functions, and provides a computational pathway for model reduction of parametric LTI systems. 

Identification of patient-specific scar and fibrosis is a critical step in the personalization of cardiac computational models. Late gadolinium enhanced cardiac magnetic resonance imaging (LGE-cMRI) is often used to identify patient anatomy, as well as tissue fibrosis and scar. Automated methods to identify scar from LGE-cMRI exist. Still, there is no clear consensus as to which is best in the context of patient-specific computational modeling of atrial fibrillation. There has been no substantial investigation into the effects that variability in scar may have on downstream patient-specific simulations. This study compares the distribution of scar patterns generated via automated LGE-cMRI analysis alongside human-guided scar identification. We assess the effects each identified scar pattern has on downstream computational modeling outputs by comparing the number of stable re-entrant arrhythmias induced In Silico in atrial fibrillation. We find both substantial disagreement between scar patterns identified via automated and human-guided methods, as well as sensitivity in the arrhythmia simulation outcomes across scar patterns. These results highlight the sensitivity of such computational models to these input parameters and enforce the need for robust personalization tools in the cardiac modeling field.

The development of network-connected computing resources has led to various computing paradigms over the years, each bringing its own set of challenges for creating efficient distributed systems. Currently, there is an increasing need to integrate the evolving Internet of Things (IoT) with the established Cloud infrastructure. This integration often requires adding intermediate layers to address Cloud limitations such as latency, bandwidth, security, cost, and control. This configuration, known as the computing continuum, involves a diverse array of distributed devices with unique characteristics working together to meet the demands of both current and emerging applications. This paper explores the path that has led to the development of the computing continuum, offering a technology-agnostic definition from a historical perspective. It also examines applications that can benefit from the computing continuum and identifies research challenges that need to be addressed to fully realize its potential.

Ventricular arrhythmia frequently complicates myocardial ischemic events, sometimes to devastating ends. Accurate arrhythmia prediction in this setting could improve outcomes, yet traditional models struggle with the temporal complexity of the data. This study employs a Long Short-Term Memory (LSTM) network to predict the time to the next premature ventricular contraction (PVC) using high-resolution experimental data. We analyzed electrograms from 11 large animal experiments, identifying 1832 PVCs, and computed time-to-PVC. An LSTM model (247 inputs, 1024 hidden units) was trained on 10 experiments, with one held out for testing, achieving a validation MAE of 8.6 seconds and a test MAE of 135 seconds (loss 68.5). Scatter plots showed strong validation correlation and a positive test trend, suggesting the potential of this approach.

Artificial intelligence is increasingly pervasive in many sectors. In this regard, IT operations are having a big deal on extracting useful information from the large amount of resources' datasets available (e.g., CPU, memory, disk, energy). The issue is bigger if we consider multiple cloud tiers. Artificial intelligence is a key technology when the main goal is to improve microservice migration through offload management. However, it struggles to facilitate distributed contexts where both data transfer needs to be reduced and data privacy needs to be increased. There is therefore a need for novel solutions that resolve the problem of prediction resource utilization (e.g. CPU) while maintaining data privacy and reducing data communication. In this paper, we present a Bi-LSTM model with attention trained in Federated Learning on CPU historical data. The dataset comes from multiple Microsoft Azure trace. The results are compared with the literature and showcase a good generalization and prediction results for metrics collected by multiple virtual machines. The model is evaluated in terms of R-squared, MSE, RMSE and MAE.

A cell’s ability to sense and respond to the mechanical properties of the extracellular matrix (ECM) is essential for maintaining tissue homeostasis, and its disruption contributes to diseases such as fibrosis, cardiovascular disorders, and cancer. Effective mechanical coupling between the plasma membrane, the underlying filamentous actin (F-actin) cytoskeleton, and integrin-based adhesion complexes (IACs) is required to link ECM mechanics to cell morphology, yet the underlying mechanisms remain incompletely understood. Here, we combine computational modeling and high-resolution imaging to show that integrin–ECM bonds determine F-actin cytoskeleton organization. On soft substrates, short-lived IACs bonds allow rapid actin retrograde flow and dense branching, restricting protrusion and limiting cell spreading. In contrast, stiff substrates or Mn²⁺-mediated integrin activation stabilize adhesions, promote filament alignment, and drive membrane protrusion for cell spreading. These cytoskeletal transitions arise from feedback between adhesion strength and the spatial positioning of the F-actin barbed ends relative to the leading-edge membrane. This positioning determines whether filaments polymerize into linear bundles or branch into dendritic networks, each generating distinct protrusive forces that regulate cell spreading. Collectively, our findings establish integrin–ECM bond stability as a key regulator of F-actin cytoskeleton organization and cell morphology.

Focused ultrasound (FUS) therapies for cancer provide non-invasive precise thermal treatment to tissues. Current FUS treatment systems rely on magnetic resonance imaging (MRI), B-mode ultrasound imaging, or harmonic motion imaging for guidance. MRI has the advantage of quantitatively measuring temperature but the high cost and limited availability of MRI limit ultimate impact. An MR-compatible Ultrasound Through Transmission (UTT) system that can quantitatively assess tissue changes caused by temperature is being investigated as a replacement for the current standard of guidance. Two FUS 256-element transducers are mounted in a container with a sample placed at their coinciding geometric center. A FUS transmitter system is connected to the transmitter transducer, while a FUS research system samples the signal at the receiver transducer. A UTT protocol of 256 sequential single-element transmissions with reception on 256 elements is performed on three different samples: homogeneous gelatin phantoms, gelatin phantoms with attenuative inclusions, and porcine meat samples before and after thermal ablation. Hybrid Angular Spectrum (HAS) acoustic simulations are performed on the unheated gelatin samples segmented into regions with measured properties. Simulations are also done on the heated porcine sample segmented into regions with ablative temperatures based on MR temperature imaging. Measured UTT datasets are compared to HAS-predicted transmission data. Complex regression shows good agreement between the UTT-measured and HAS-predicted datasets. On homogeneous samples, the average complex correlation coefficient across all receiver elements is 0.8897. In 1 cm and 2 cm diameter inclusion samples, the correlation is 0.7994 and 0.6934, respectively. On the porcine meat sample, the pre-ablation average correlation is 0.6636, with a post-ablation average correlation of 0.6595. The data indicate that ablation of the tissue causes measurable changes in the received signal, but our simplified model is inadequate to capture the true tissue changes. Future work is investigating this discrepancy with more advanced modeling. The ultimate goal is to use physics-informed neural networks to predict tissue changes from the UTT received signal.

We examine which decentralized finance architectures enable meaningful regulation by combining financial and computational theory. We show via deduction that a decentralized and permissionless Turing-complete system cannot provably comply with regulations concerning anti-money laundering, know-your-client obligations, some securities restrictions and forms of exchange control. Any system that claims to follow regulations must choose either a form of permission or a less-than-Turing-complete update facility. Compliant decentralized systems can be constructed only by compromising on the richness of permissible changes. Regulatory authorities must accept new tradeoffs that limit their enforcement powers if they want to approve permissionless platforms formally. Our analysis demonstrates that the fundamental constraints of computation theory have direct implications for financial regulation. By mapping regulatory requirements onto computational models, we characterize which types of automated compliance are achievable and which are provably impossible. This framework allows us to move beyond traditional debates about regulatory effectiveness to establish concrete boundaries for automated enforcement.

Transcatheter arterial embolization (TAE) is a minimally invasive technique used to treat hypervascular tumors, hemorrhage, and vascular abnormalities. Though microspheres (MSs) have achieved widespread clinical use as embolic agents, they often lack imaging opacity, optimal morphology and mechanical properties which can lead to unpredictable trajectories, non-target delivery, and suboptimal embolization. This study developed tantalum-loaded calcium alginate (Ta@Ca-Alg) MSs with intrinsic radiopacity, tunable density, and mechanical properties. Ta@Ca-Alg MSs were synthesized using a gas-shearing method and analyzed for size, morphology, swelling behavior, density, radiopacity, and optimized mechanical properties. The results demonstrated that Ta@Ca-Alg MSs maintained a narrow size distribution, with increasing Ta concentration enhancing radiopacity to levels comparable with the clinical contrast agent OMNIPAQUE 350. Density and Young's modulus corresponding to different Ta concentrations were also investigated. Phantom model testing validated effective vessel occlusion and controlled penetration. In vitro hemocompatibility, sterility, and cytotoxicity studies confirmed excellent biocompatibility. These findings suggest that Ta@Ca-Alg MSs are a promising radiopaque embolic agent with optimized radiopacity, density, and mechanical properties, offering excellent potential for TAE procedures.