Unbalanced optimal transport (UOT) is a natural extension of optimal transport (OT) allowing comparison between measures of different masses. It arises naturally in machine learning by offering a robustness against outliers. The aim of this work is to provide convergence rates of the regularized transport cost and plans towards their original solution when both measures are weighted sums of Dirac masses.
In this paper, we address a bicriteria network design problem that arises from practical applications in urban and rural public transportation planning. We establish the problem's complexity and demonstrate inapproximability results, highlighting the inherent difficulties in finding optimal solutions. Additionally, we identify special cases where approximability can be achieved, providing valuable insights for practitioners. Our proofs leverage complexity results related to directed network design problems, an area that has received limited attention in the existing literature. By investigating these complexity results, we aim to fill a critical gap and enhance the understanding of the interplay between bicriteria decision-making and network design challenges.
Neural networks (NNs) have a great potential for future neutrino telescopes such as IceCube-Gen2, the planned high-energy extension of the IceCube observatory. IceCube-Gen2 will feature new optical sensors with multiple photomultiplier tubes (PMTs) designed to provide omnidirectional sensitivity. Neural networks excel at handling high-dimensional problems and can naturally incorporate the increased complexity of these new sensors. Additionally, their fast inference time makes them promising candidates for handling the high event rates expected from IceCube-Gen2. This contribution presents potential applications of neural networks in the IceCube-Gen2 in-ice optical array. First, we introduce a method to simulate the IceCube-Gen2 optical modules' photon acceptance using a NN that leverages the modules' inherent symmetries. Secondly, we present the status of neutrino NN-based reconstruction efforts, including the adaptation of a novel IceCube technique that combines normalizing flows with transformer NNs. Finally, we describe current progress in noise cleaning applications based on node classification with graph neural networks (GNNs), a method that has already shown promising results for the forthcoming low-energy extension, IceCube-Upgrade.
Ensuring reliable and low-latency communication in offshore wind farms is critical for efficient monitoring and control, yet remains challenging due to the harsh environment and lack of infrastructure. This paper investigates a flying base station (FBS) approach for wide-area monitoring and control in the UK Hornsea offshore wind farm project. By leveraging mobile, flexible FBS platforms in the remote and harsh offshore environment, the proposed system offers real-time connectivity for turbines without the need for deploying permanent infrastructure at the sea. We develop a detailed and practical end-to-end latency model accounting for five key factors: flight duration, connection establishment, turbine state information upload, computational delay, and control transmission, to provide a holistic perspective often missing in prior studies. Furthermore, we combine trajectory planning, beamforming, and resource allocation into a multi-objective optimization framework for the overall latency minimization, specifically designed for large-scale offshore wind farm deployments. Simulation results verify the effectiveness of our proposed method in minimizing latency and enhancing efficiency in FBS-assisted offshore monitoring across various power levels, while consistently outperforming baseline designs.
The next generation neutrino telescope, IceCube-Gen2, will be sensitive to the astrophysical and cosmogenic flux of neutrinos across a broad energy range, from the TeV to the EeV scale. The planned design includes 8 cubic kilometers of ice instrumented with approximately 10,000 optical sensors, a surface array, and a radio array of antennas embedded in the ice laid out sparsely over 500 km^2. The radio array provides sensitivity to ultra-high energy neutrinos using independent radio stations that can trigger on Askaryan emission from neutrino interactions in the ice. In this contribution, we present the design for the radio array along with its planned implementation, which is expected to increase sensitivity to neutrinos with energies beyond 100PeV by at least an order of magnitude over existing arrays. Furthermore, we will quantify the expected science output by presenting measurement forecasts for the main science cases of diffuse flux and point source discovery, as well as cross-section and flavor measurements.
Background: Accurate forecasting of lung tumor motion is essential for precise dose delivery in proton therapy. While current markerless methods mostly rely on deep learning, transformer-based architectures remain unexplored in this domain, despite their proven performance in trajectory forecasting. Purpose: This work introduces a markerless forecasting approach for lung tumor motion using Vision Transformers (ViT). Two training strategies are evaluated under clinically realistic constraints: a patient-specific (PS) approach that learns individualized motion patterns, and a multi-patient (MP) model designed for generalization. The comparison explicitly accounts for the limited number of images that can be generated between planning and treatment sessions. Methods: Digitally reconstructed radiographs (DRRs) derived from planning 4DCT scans of 31 patients were used to train the MP model; a 32nd patient was held out for evaluation. PS models were trained using only the target patient's planning data. Both models used 16 DRRs per input and predicted tumor motion over a 1-second horizon. Performance was assessed using Average Displacement Error (ADE) and Final Displacement Error (FDE), on both planning (T1) and treatment (T2) data. Results: On T1 data, PS models outperformed MP models across all training set sizes, especially with larger datasets (up to 25,000 DRRs, p < 0.05). However, MP models demonstrated stronger robustness to inter-fractional anatomical variability and achieved comparable performance on T2 data without retraining. Conclusions: This is the first study to apply ViT architectures to markerless tumor motion forecasting. While PS models achieve higher precision, MP models offer robust out-of-the-box performance, well-suited for time-constrained clinical settings.
Digital Twin (DT) technology has emerged as a transformative approach in healthcare, but its application in personalized patient care remains limited. This paper aims to present a practical implementation of DT in the management of chronic diseases. We introduce a general DT framework for personalized care planning (DT4PCP), with the core components being a real-time virtual representation of a patient's health and emerging predictive models to enable adaptive, personalized care. We implemented the DT4PCP framework for managing Type 2 Diabetes (DT4PCP-T2D), enabling real-time collection of behavioral data from patients with T2D, predicting emergency department (ED) risks, simulating the effects of different interventions, and personalizing care strategies to reduce ED visits. The DT4PCP-T2D also integrates social determinants of health (SDoH) and other contextual data, offering a comprehensive view of the patient's health to ensure that care recommendations are tailored to individual needs. Through retrospective simulations, we demonstrate that integrating DTs in T2D management can lead to significant advancements in personalized medicine. This study underscores the potential of DT technology to revolutionize chronic disease care.
Accurate survival prediction is critical in oncology for prognosis and treatment planning. Traditional approaches often rely on a single data modality, limiting their ability to capture the complexity of tumor biology. To address this challenge, we introduce a multimodal deep learning framework for survival analysis capable of modeling both single and competing risks scenarios, evaluating the impact of integrating multiple medical data sources on survival predictions. We propose SAMVAE (Survival Analysis Multimodal Variational Autoencoder), a novel deep learning architecture designed for survival prediction that integrates six data modalities: clinical variables, four molecular profiles, and histopathological images. SAMVAE leverages modality specific encoders to project inputs into a shared latent space, enabling robust survival prediction while preserving modality specific information. Its parametric formulation enables the derivation of clinically meaningful statistics from the output distributions, providing patient-specific insights through interactive multimedia that contribute to more informed clinical decision-making and establish a foundation for interpretable, data-driven survival analysis in oncology. We evaluate SAMVAE on two cancer cohorts breast cancer and lower grade glioma applying tailored preprocessing, dimensionality reduction, and hyperparameter optimization. The results demonstrate the successful integration of multimodal data for both standard survival analysis and competing risks scenarios across different datasets. Our model achieves competitive performance compared to state-of-the-art multimodal survival models. Notably, this is the first parametric multimodal deep learning architecture to incorporate competing risks while modeling continuous time to a specific event, using both tabular and image data.
Mandibular Angle Split Osteotomy (MASO) is a significant procedure in oral and maxillofacial surgery. Despite advances in technique and instrumentation, its success still relies heavily on the surgeon's experience. In this work, a human-robot collaborative system is proposed to perform MASO according to a preoperative plan and under guidance of a surgeon. A task decomposition methodology is used to divide the collaborative surgical procedure into three subtasks: (1) positional control and (2) orientation control, both led by the robot for precise alignment; and (3) force-control, managed by surgeon to ensure safety. Additionally, to achieve patient tracking without the need for a skull clamp, an optical tracking system (OTS) is utilized. Movement of the patient mandibular is measured with an optical-based tracker mounted on a dental occlusal splint. A registration method and Robot-OTS calibration method are introduced to achieve reliable navigation within our framework. The experiments of drilling were conducted on the realistic phantom model, which demonstrated that the average error between the planned and actual drilling points is 1.85mm.
Understanding and predicting mobility dynamics in transportation networks is critical for infrastructure planning, resilience analysis, and traffic management. Traditional graph-based models typically assume memoryless movement, limiting their ability to capture sequential dependencies inherent in real-world mobility patterns. In this study, we introduce a novel higher-order network framework for modeling memory-dependent dynamics in transportation systems. By extending classical graph representations through higher-order Markov chains and de Bruijn graph structures, our framework encodes the spatial and temporal ordering of traversed paths, enabling the analysis of structurally and functionally critical components with improved fidelity. We generalize key network analytics, including betweenness centrality, PageRank, and next-step prediction, to this higher-order setting and validate our approach on the Sioux Falls transportation network using agent-based trajectory data generated with MATSim. Experimental results demonstrate that higher-order models outperform first-order baselines across multiple tasks, with the third-order model achieving an optimal balance between predictive accuracy and model complexity. These findings highlight the importance of incorporating memory effects into network-based transportation analysis and offer a scalable, data-driven methodology for capturing complex mobility behaviors in infrastructure systems.
Medical Image Segmentation (MIS) plays a crucial role in medical therapy planning and robot navigation. Prototype learning methods in MIS focus on generating segmentation masks through pixel-to-prototype comparison. However, current approaches often overlook sample diversity by using a fixed prototype per semantic class and neglect intra-class variation within each input. In this paper, we propose to generate instance-adaptive prototypes for MIS, which integrates a common prototype proposal (CPP) capturing common visual patterns and an instance-specific prototype proposal (IPP) tailored to each input. To further account for the intra-class variation, we propose to guide the IPP generation by re-weighting the intermediate feature map according to their confidence scores. These confidence scores are hierarchically generated using a transformer decoder. Additionally we introduce a novel self-supervised filtering strategy to prioritize the foreground pixels during the training of the transformer decoder. Extensive experiments demonstrate favorable performance of our method.
Autonomous vehicles rely on global standard-definition (SD) maps for road-level route planning and online local high-definition (HD) maps for lane-level navigation. However, recent work concentrates on construct online HD maps, often overlooking the association of global SD maps with online HD maps for hybrid navigation, making challenges in utilizing online HD maps in the real world. Observing the lack of the capability of autonomous vehicles in navigation, we introduce \textbf{O}nline \textbf{M}ap \textbf{A}ssociation, the first benchmark for the association of hybrid navigation-oriented online maps, which enhances the planning capabilities of autonomous vehicles. Based on existing datasets, the OMA contains 480k of roads and 260k of lane paths and provides the corresponding metrics to evaluate the performance of the model. Additionally, we propose a novel framework, named Map Association Transformer, as the baseline method, using path-aware attention and spatial attention mechanisms to enable the understanding of geometric and topological correspondences. The code and dataset can be accessed at https://github.com/WallelWan/OMA-MAT.
The integration of wireless power transfer (WPT) with Internet of Things (IoT) offers promising solutions for sensing applications, but faces significant challenges when deployed in hard-to-access areas such as high-temperature environments. In such extreme conditions, traditional fixed WPT infrastructure cannot be safely installed, and batteries rapidly degrade due to hardware failures. In this paper, we propose an uncrewed aerial vehicle (UAV)-assisted data collection and WPT framework for batteryless sensor (BLS) networks deployed in these challenging environments. Specifically, we consider a practical scenario where a UAV first transfers energy to BLS nodes via WPT, enabling these nodes to subsequently transmit their collected data to the UAV through orthogonal frequency-division multiple access (OFDMA). Then, we formulate a multi-objective optimization problem that aims to maximize the fair data collection volume while minimizing the UAV energy consumption through joint optimization of transmit power allocation and flight trajectory planning. Due to the non-convex nature and dynamic characteristics of this problem, conventional optimization methods prove inadequate. To address these challenges, we propose an enhanced soft actor-critic algorithm with parameter-free attention, prioritized experience replay, and value-based reward centering (SAC-PPV), thereby improving the exploration efficiency and learning stability of the algorithm in complex WPT scenarios. Simulation results demonstrate that the proposed approach consistently outperforms benchmark algorithms under various network configurations.
Query optimization is a fundamental task in database systems that is crucial to providing high performance. To evaluate learned and traditional optimizer's performance, several benchmarks, such as the widely used JOB benchmark, are used. However, in this paper, we argue that existing benchmarks are inherently limited, as they do not reflect many real-world properties of query optimization, thus overstating the performance of both traditional and learned optimizers. In fact, simple but realistic properties, such as joins over string columns or complex filter predicates, can drastically reduce the performance of existing query optimizers. Thus, we introduce JOB-Complex, a new benchmark designed to challenge traditional and learned query optimizers by reflecting real-world complexity. Overall, JOB-Complex contains 30 SQL queries and comes together with a plan-selection benchmark containing nearly 6000 execution plans, making it a valuable resource to evaluate the performance of query optimizers and cost models in real-world scenarios. In our evaluation, we show that traditional and learned cost models struggle to achieve high performance on JOB-Complex, providing a runtime of up to 11x slower compared to the optimal plans.
Ensuring the functional safety of motion planning modules in autonomous vehicles remains a critical challenge, especially when dealing with complex or learning-based software. Online verification has emerged as a promising approach to monitor such systems at runtime, yet its integration into embedded real-time environments remains limited. This work presents a safeguarding concept for motion planning that extends prior approaches by introducing a time safeguard. While existing methods focus on geometric and dynamic feasibility, our approach additionally monitors the temporal consistency of planning outputs to ensure timely system response. A prototypical implementation on a real-time operating system evaluates trajectory candidates using constraint-based feasibility checks and cost-based plausibility metrics. Preliminary results show that the safeguarding module operates within real-time bounds and effectively detects unsafe trajectories. However, the full integration of the time safeguard logic and fallback strategies is ongoing. This study contributes a modular and extensible framework for runtime trajectory verification and highlights key aspects for deployment on automotive-grade hardware. Future work includes completing the safeguarding logic and validating its effectiveness through hardware-in-the-loop simulations and vehicle-based testing. The code is available at: https://github.com/TUM-AVS/motion-planning-supervisor
The construction of Low Earth Orbit (LEO) satellite constellations has recently attracted tremendous attention from both academia and industry. The 5G and 6G standards have identified LEO satellite networks as a key component of future mobile networks. However, due to the high-speed movement of satellites, ground terminals often experience frequent and high-latency handovers, which significantly deteriorate the performance of latency-sensitive applications. To address this challenge, we propose a parallel handover mechanism for mobile satellite networks that can considerably reduce handover latency. The main idea is to employ plan-based handovers instead of measurement-based handovers to avoid interactions between the access and core networks, thereby eliminating the significant time overhead associated with traditional handover procedures. Specifically, we introduce a novel network function named the Satellite Synchronized Function (SSF), which is designed to be fully compliant with the standard 5G core network. In addition, we propose a machine learning model for signal strength prediction, coupled with an efficient handover scheduling algorithm. We have conducted extensive experiments, and the results demonstrate that our proposed handover scheme can reduce handover latency by 21\times compared to the standard NTN handover scheme and two other existing handover approaches, along with significant improvements in network stability and user-level performance.
SRL, defined as learners' ability to systematically plan, monitor, and regulate their learning activities, is crucial for sustained academic achievement and lifelong learning competencies. Emerging Artificial Intelligence (AI) developments profoundly influence SRL interactions by potentially either diminishing or strengthening learners' opportunities to exercise their own regulatory skills. Recent literature emphasizes a balanced approach termed Hybrid Human-AI Regulated Learning (HHAIRL), in which AI provides targeted, timely scaffolding while preserving the learners' role as active decision-makers and reflective monitors of their learning process. Nevertheless, existing digital tools frequently fall short, lacking adaptability, focusing narrowly on isolated SRL phases, and insufficiently support meaningful human-AI interactions. In response, this paper introduces the enhanced \flora Engine, which incorporates advanced Generative Artificial Intelligence (GenAI) features and state-of-the-art learning analytics, explicitly grounded in SRL and HHAIRL theories. The \flora Engine offers instrumentation tools such as collaborative writing, multi-agents chatbot, and detailed learning trace logging to support dynamic, adaptive scaffolding tailored to individual needs in real time. We further present a summary of several research studies that provide the validations for and illustrate how these instrumentation tools can be utilized in real-world educational and experimental contexts. These studies demonstrate the effectiveness of \flora Engine in fostering SRL and HHAIRL, providing both theoretical insights and practical solutions for the future of AI-enhanced learning context.
We present GO-CBED, a goal-oriented Bayesian framework for sequential causal experimental design. Unlike conventional approaches that select interventions aimed at inferring the full causal model, GO-CBED directly maximizes the expected information gain (EIG) on user-specified causal quantities of interest, enabling more targeted and efficient experimentation. The framework is both non-myopic, optimizing over entire intervention sequences, and goal-oriented, targeting only model aspects relevant to the causal query. To address the intractability of exact EIG computation, we introduce a variational lower bound estimator, optimized jointly through a transformer-based policy network and normalizing flow-based variational posteriors. The resulting policy enables real-time decision-making via an amortized network. We demonstrate that GO-CBED consistently outperforms existing baselines across various causal reasoning and discovery tasks-including synthetic structural causal models and semi-synthetic gene regulatory networks-particularly in settings with limited experimental budgets and complex causal mechanisms. Our results highlight the benefits of aligning experimental design objectives with specific research goals and of forward-looking sequential planning.
We present initial results from a planned 10 year survey of Ca II H & K emission, using observations made with the ARC 3.5m Telescope at Apache Point Observatory. The primary goal of the survey is to investigate activity cycles in low mass stars. The sample includes stars chosen from the legacy Mount Wilson survey carried out by Olin Wilson more than 50 years ago, together with newly identified planet-host stars and a select sample of early-mid M dwarfs. This paper presents the first four years of data, comprising 1040 observations of 271 stars, with a specific focus on K and M stars. We identify a subsample of 153 stars for continuing observations over the full 10 year survey. Early results indicate that our data are consistent with the MWO cycle periods over a time span of more than 50 years; that there is a bifurcation in activity in the late K range with separate populations of low and high activity stars at lower masses; and that M dwarf planet hosts tend to be mainly found in the population of low activity stars, even in the unbiased (by activity) TESS sample, potentially indicating a link between activity and planet formation. We have also found indications of possible cyclic variability in some of the lower mass stars in the sample. Our ultimate goal is to link the activity cycle and rotation periods in a robust sample of stars spanning FGKM spectral types and to investigate the implications for the underlying magnetic dynamo.
Recent decision-making systems are increasingly complicated, making it crucial to verify and understand their behavior for a given specification. A promising approach is to comprehensively explain undesired behavior in the systems modeled by Markov decision processes (MDPs) through formal verification and causal reasoning. However, the reliable explanation using model-based probabilistic causal analysis has not been explored when the MDP's transition probabilities are uncertain. This paper proposes a method to identify potential causes of undesired behaviors in an uncertain parametric MDP (upMDP) using parameter sampling, model checking, and a set covering for the samples. A cause is defined as a subset of states based on a probability-raising principle. We show that the probability of each identified subset being a cause exceeds a specified threshold. Further, a lower bound of the probability that the undesired paths visit the subsets is maximized as much as possible while satisfying a nonredundancy condition. While computing these probabilities is complicated, this study derives probabilistically approximately correct lower bounds of both probabilities by the sampling. We demonstrate the effectiveness of the proposed method through a path-planning scenario.
The NASA Great Observatories Maturation Program is a development plan to efficiently and effectively develop large, strategic astrophysics missions. Suborbital rocket and balloon programs have long been a key development tool for enabling large missions in NASA astrophysics. We review the significance of these suborbital missions in the preceding decades to demonstrate their contributions to the Great Observatories Maturation Program for the Habitable Worlds Observatory and beyond. We show that suborbital instruments have obtained new science observations of astrophysical sources across the electromagnetic spectrum, matured high-priority component technologies, and served as a training ground for principal investigators of Explorer-class astrophysics satellites. A brief discussion of emerging CubeSat and SmallSat missions and their place in the NASA astrophysics portfolio is also provided.
The DarkSide-20k dark matter direct detection experiment will employ a 21 m^2 silicon photomultiplier (SiPM) array, instrumenting a dual-phase 50 tonnes liquid argon Time Projection Chamber (TPC). SiPMs are arranged into modular photosensors called Tiles, each integrating 24 SiPMs onto a printed circuit board (PCB) that provides signal amplification, power distribution, and a single-ended output for simplified readout. 16 Tiles are further grouped into Photo-Detector Units (PDUs). This paper details the production of the Tiles and the quality assurance and quality control (QA-QC) protocol established to ensure their performance and uniformity. The production and QA-QC of the Tiles are carried out at Nuova Officina Assergi (NOA), an ISO-6 clean room facility at LNGS. This process includes wafer-level cryogenic characterisation, precision flip-chip bonding, wire bonding, and extensive electrical and optical validation of each Tile. The overall production yield exceeds 83.5%, matching the requirements of the DarkSide-20k production plan. These results validate the robustness of the Tile design and its suitability for operation in a cryogenic environment.
We present 4KAgent, a unified agentic super-resolution generalist system designed to universally upscale any image to 4K resolution (and even higher, if applied iteratively). Our system can transform images from extremely low resolutions with severe degradations, for example, highly distorted inputs at 256x256, into crystal-clear, photorealistic 4K outputs. 4KAgent comprises three core components: (1) Profiling, a module that customizes the 4KAgent pipeline based on bespoke use cases; (2) A Perception Agent, which leverages vision-language models alongside image quality assessment experts to analyze the input image and make a tailored restoration plan; and (3) A Restoration Agent, which executes the plan, following a recursive execution-reflection paradigm, guided by a quality-driven mixture-of-expert policy to select the optimal output for each step. Additionally, 4KAgent embeds a specialized face restoration pipeline, significantly enhancing facial details in portrait and selfie photos. We rigorously evaluate our 4KAgent across 11 distinct task categories encompassing a total of 26 diverse benchmarks, setting new state-of-the-art on a broad spectrum of imaging domains. Our evaluations cover natural images, portrait photos, AI-generated content, satellite imagery, fluorescence microscopy, and medical imaging like fundoscopy, ultrasound, and X-ray, demonstrating superior performance in terms of both perceptual (e.g., NIQE, MUSIQ) and fidelity (e.g., PSNR) metrics. By establishing a novel agentic paradigm for low-level vision tasks, we aim to catalyze broader interest and innovation within vision-centric autonomous agents across diverse research communities. We will release all the code, models, and results at: https://4kagent.github.io.
The widely anticipated outburst of recurrent nova T Coronae Borealis (T CrB), which is near the end of its 80-year cycle, provides an excellent opportunity to search for neutrinos from novae. Novae are an energetic class of transients, which have been studied for hundreds of years. Because many of them are located nearby, novae provide an excellent astrophysical laboratory to study shock-powered emission in our own backyard. Several recent novae have previously been detected in GeV gamma rays, and the 2021 outburst of RS Ophiuchi was detected up to TeV energies, with evidence for a hadronic origin of the observed emission. Previous searches for GeV-TeV neutrinos from novae, predicted to occur alongside their gamma-ray emission, have been performed using data from the IceCube Neutrino Observatory. However, no significant neutrino signals from novae have yet been observed. We present plans for follow-up of T CrB in real time with IceCube, using datasets spanning GeV to PeV neutrino energies. Due to its closer distance and higher optical flux, which has been well measured in two historical eruptions, the expected neutrino signal from T CrB is several times stronger than that from RS Ophiuchi. Furthermore, T CrB is located in the Northern sky at a declination where IceCube's sensitivity is an additional factor of a few better than at the location of RS Ophiuchi, which is beneficial to this search.
ColdADC is a custom ASIC digitizer implemented in 65 nm CMOS technology using specialized techniques for long-term reliability in cryogenic environments. ColdADC was developed for use in the DUNE Far Detector complex, which will consist of four liquid argon time projection chambers. Each contains 17 kilotons liquid argon as the target material in order to measure neutrino oscillations. Approximately 40,000 ColdADC ASICs will be installed for DUNE in the first two large detectors and will be operated at cryogenic temperatures during the experiment without replacement. The lifetime of the ColdADC is a critical parameter affecting the data quality and physics sensitivity of the experiment. A measurement of the lifetime of the ColdADC was carried out, and the results shown in this paper assure orders of magnitude longer lifetime of the ColdADC than the planned operation time of the detectors.
We consider the problem of decomposing a piecewise constant function on the circle into a sum of indicator functions of closed circular disks in the plane, whose number and location are not a priori known. This represents a situation where an agent moving on the circle is able to sense its proximity to some landmarks, and the goal is to estimate the number of these landmarks and their possible locations -- which can in turn enable control tasks such as motion planning and obstacle avoidance. Moreover, the exact values of the function at its discontinuities (which correspond to disk boundaries for the individual indicator functions) are not assumed to be known to the agent. We introduce suitable notions of robustness and degrees of freedom to single out those decompositions that are more desirable, or more likely, given this non-precise data collected by the agent. We provide a characterization of robust decompositions and give a procedure for generating all such decompositions. When the given function admits a robust decomposition, we compute the number of possible robust decompositions and derive bounds for the number of decompositions maximizing the degrees of freedom.
Accurate prognosis of non-small cell lung cancer (NSCLC) patients undergoing immunotherapy is essential for personalized treatment planning, enabling informed patient decisions, and improving both treatment outcomes and quality of life. However, the lack of large, relevant datasets and effective multi-modal feature fusion strategies pose significant challenges in this domain. To address these challenges, we present a large-scale dataset and introduce a novel framework for multi-modal feature fusion aimed at enhancing the accuracy of survival prediction. The dataset comprises 3D CT images and corresponding clinical records from NSCLC patients treated with immune checkpoint inhibitors (ICI), along with progression-free survival (PFS) and overall survival (OS) data. We further propose a cross-modality masked learning approach for medical feature fusion, consisting of two distinct branches, each tailored to its respective modality: a Slice-Depth Transformer for extracting 3D features from CT images and a graph-based Transformer for learning node features and relationships among clinical variables in tabular data. The fusion process is guided by a masked modality learning strategy, wherein the model utilizes the intact modality to reconstruct missing components. This mechanism improves the integration of modality-specific features, fostering more effective inter-modality relationships and feature interactions. Our approach demonstrates superior performance in multi-modal integration for NSCLC survival prediction, surpassing existing methods and setting a new benchmark for prognostic models in this context.
A common class of algorithms for informative path planning (IPP) follows boustrophedon ("as the ox turns") patterns, which aim to achieve uniform area coverage. However, IPP is often applied in scenarios where anomalies, such as plant diseases, pollution, or hurricane damage, appear in clusters. In such cases, prioritizing the exploration of anomalous regions over uniform coverage is beneficial. This work introduces a class of algorithms referred to as bounom\=odes ("as the ox grazes"), which alternates between uniform boustrophedon sampling and targeted exploration of detected anomaly clusters. While uniform sampling can be designed using geometric principles, close exploration of clusters depends on the spatial distribution of anomalies and must be learned. In our implementation, the close exploration behavior is learned using deep reinforcement learning algorithms. Experimental evaluations demonstrate that the proposed approach outperforms several established baselines.
Purpose: This work presents the development and pilot implementation of a comprehensive remote dosimetry audit for Ir-192 High Dose Rate interstitial brachytherapy, integrating experimental and computational dosimetry procedures into a unified workflow. TG43 and Model Based Dose Calculations Algorithms (MBDCAs) are both considered. Methods: A compact, water-equivalent phantom was designed to hold two catheters, ten Optically Stimulated Luminescent Dosimeters (OSLDs) and two radiochromic films, enabling point and 2D dose measurements. A user-selected treatment plan was created using a clinical Treatment Planning System (TPS), tailored to the optimal dose range of the dosimeters. A computational dosimetry audit test was also performed via Monte Carlo (MC) simulations, enabling independent 3D dose calculations for the same plan and phantom geometry. All dosimetry results were compared to TPS calculations (TG43 and an MBDCA) using the Gamma Index (GI) test, dose difference maps, and dose-volume histogram comparisons, wherever applicable. The protocol was designed to minimize clinical workload. Results: This study was completed within ten days of phantom delivery to the clinic. If necessary, measurements were corrected using appropriate correction factors determined through side studies. GI passing criteria were adapted to the uncertainty of each dosimetry system. Excellent agreement was found between MBDCA and experimental or MC results. Within the volume of interest, TG43 systematically overestimated dose compared to MC (median difference: 2.16%), attributed to missing scatter and phantom material. Conclusion: Despite the labor-intensive workflow, this protocol supports remote Ir-192 audits with acceptable uncertainties. Combining experimental and computational methods enhances the robustness of the audit. This hybrid approach shows clear advantages for rigorous dosimetry auditing programs.
Graphical User Interface (GUI) agents powered by Large Vision-Language Models (LVLMs) have emerged as a revolutionary approach to automating human-machine interactions, capable of autonomously operating personal devices (e.g., mobile phones) or applications within the device to perform complex real-world tasks in a human-like manner. However, their close integration with personal devices raises significant security concerns, with many threats, including backdoor attacks, remaining largely unexplored. This work reveals that the visual grounding of GUI agent-mapping textual plans to GUI elements-can introduce vulnerabilities, enabling new types of backdoor attacks. With backdoor attack targeting visual grounding, the agent's behavior can be compromised even when given correct task-solving plans. To validate this vulnerability, we propose VisualTrap, a method that can hijack the grounding by misleading the agent to locate textual plans to trigger locations instead of the intended targets. VisualTrap uses the common method of injecting poisoned data for attacks, and does so during the pre-training of visual grounding to ensure practical feasibility of attacking. Empirical results show that VisualTrap can effectively hijack visual grounding with as little as 5% poisoned data and highly stealthy visual triggers (invisible to the human eye); and the attack can be generalized to downstream tasks, even after clean fine-tuning. Moreover, the injected trigger can remain effective across different GUI environments, e.g., being trained on mobile/web and generalizing to desktop environments. These findings underscore the urgent need for further research on backdoor attack risks in GUI agents.