Suboptimal Data Research Articles

Collaborative relay for achieving long-term and low-AoI data collection in UAV-aided IoT systems

In Internet of Things (IoT) systems, sensor nodes are frequently placed in remote and unattended locations to monitor environmental data. One significant challenge is ensuring the timely and efficient transmission of data generated by these sensor nodes back to the base station. The use of unmanned aerial vehicles (UAVs) can provide a practical solution to this challenge by acting as mobile relay nodes for facilitating data transmission. In most existing works, UAVs are typically restricted to collecting data within their designated areas and returning to the base station for data offloading, resulting in suboptimal data timeliness due to long-distance flights. A limited number of works have explored the utilization of relay collaboration by UAVs for data collection, enabling efficient and immediate transmission of sensor node data to the base station. Nevertheless, UAVs positioned at significant distances from the base station face challenges in obtaining timely energy replenishment. This makes them unable to effectively support long-duration data collection missions. In order to tackle these challenges, we develop a UAV-aided IoT collaborative data collection mechanism and propose a matching games-based data collection (MGDC) scheme. In this scheme, we begin by identifying convergence nodes within the ground sensor network, responsible for uploading sensor-generated data to passing UAVs. Furthermore, we divide the mission area into multiple subareas based on the number of available UAVs. Subsequently, using a matching game algorithm, we establish relay relationships between UAVs to enable efficient relay transmissions among paired UAVs. To achieve efficient data collection of UAVs, we employ an improved adaptive large neighborhood search (IALNS) algorithm for UAV flight path planning. Finally, we incorporate an alternating charging mode to ensure all UAVs have the opportunity to return to the base station for energy recharge. Through comprehensive experimentation, we confirm the significant enhancement provided by our proposed data collection scheme compared to existing schemes. This scheme effectively reduces system age of information (AoI) and extends the runtime of the system.

LEARNING ON THE EDGE: BENCHMARKING ACTIVE LEARNING FOR THE SEMANTIC SEGMENTATION OF ALS POINT CLOUDS

Abstract. While most research in automatic semantic segmentation of 3D geospatial point clouds is concerned with enhancing respective Machine Learning (ML) models, we aim to shift the focus to be more of a data-centric nature. This means, we consider the creation of respective data sets that ML models learn from as key component, since even the most sophisticated model performs poorly when learning from suboptimal data. In this regard, the straightforward approach of providing labeled data abundantly is prohibitively expensive and just not scalable in times of high-frequency data acquistion cycles, where a dedicated training set should be available for each new epoch, as ML models often lack generalizability. As a remedy, we rely on Active Learning (AL), which is a cost-efficient and quick method to generate required training data at scale. Although AL has been (scarcely) applied in the geospatial domain before, a comprehensive evaluation of its capabilities, including benchmarking of achievable accuracies is lacking. Therefore, we apply the AL concept to both ISPRS’ current point cloud benchmark data sets as well as to a third large scale National Mapping Agency point cloud. Respective experiments are conducted with both a feature-driven Random Forest classifcation approach and a data-driven Submanifold Sparse Convolutional Neural Network classifier. Our experiments verify that by labeling only a fraction of available training points (typically ⪡ 1%), we can still reach accuracies that are at maximum only about 5 percentage points worse compared to leading benchmark contributions.

Differing sensitivity of species distribution modelling algorithms to climate data source

The increasing availability of contemporary climate databases calls for comparative studies testing their performance in ecological modelling. We evaluated how contemporary climate data sources interact with the modelling algorithm and affect the predictive accuracy of Species Distribution Models (SDMs). We modelled the distribution of 37 woody plants in the Iberian Peninsula using three global climate databases: WorldClim (WC), CHELSA (CH) and MERRAclim (MER); a regional climate database (the Iberian Climate Atlas, ICA), and three modelling algorithms: Generalized Linear Models (GLMs), Generalized Additive Models (GAMs), and Partial Least Squared Regressions (PLSRs). The predictive capacity of the models was estimated by performing an external evaluation while controlling for extrapolation issues. The climate database source variable and its interaction with the modelling algorithm variable showed significant effects on a model's predictive performance. Post hoc tests revealed that MER produced the lowest discrimination capacity scores, especially using GLMs as algorithm. PLSR models were significantly better than GLMs when MER was used. GAMs and PLSR models with MER were significantly worse than some ICA and CH models but not worse than any of the WC models. These results indicate that contemporary climate data sources need to be considered as a seedbed of uncertainty in SDMs and that poorly flexible algorithms are unable to deal with suboptimal data. CH is a reliable global climate database and PLSR is a technique worth considering in SDMs. The best practice is to select the most accurate climate data available and to choose the algorithm based on the purpose of the study and on context-dependent technical details.

Learning eco-driving strategies from human driving trajectories

Eco-driving represents a promising avenue for mitigating energy consumption in road transportation. To enhance the applicability of learning-based eco-driving strategies, this study presents a novel framework that employs offline reinforcement learning in eco-driving control. This framework enables a vehicle agent to acquire eco-driving behavior by leveraging real-world human driving trajectories. Specifically, the human driving trajectories, along with the corresponding traffic signal timing scheme, obtained from empirical data, are utilized to construct a comprehensive Markov Decision Process (MDP) dataset for offline policy training. To accommodate learning from sub-optimal human-driving data, a Conservative Q-learning (CQL) algorithm is deployed. Subsequently, the proposed offline learning method is compared with alternative learning-based, model-based, and rule-based approaches, effectively illustrating the feasibility of offline learning and the efficacy of the CQL algorithm. Notably, the energy consumption is demonstrated to be improved by 67.3% compared to a behavioral car-following model, with only marginal compromise to travel efficiency. Furthermore, a sensitivity analysis is conducted, revealing the generalizability of the offline learning-based method across various simulation configurations and even diverse energy consumption models.

Endpoint adjudication in cardiovascular clinical trials.

Endpoint adjudication (EA) is a common feature of contemporary randomized controlled trials (RCTs) in cardiovascular medicine. Endpoint adjudication refers to a process wherein a group of expert reviewers, known as the clinical endpoint committee (CEC), verify potential endpoints identified by site investigators. Events that are determined by the CEC to meet pre-specified trial definitions are then utilized for analysis. The rationale behind the use of EA is that it may lessen the potential misclassification of clinical events, thereby reducing statistical noise and bias. However, it has been questioned whether this is universally true, especially given that EA significantly increases the time, effort, and resources required to conduct a trial. Herein, we compare the summary estimates obtained using adjudicated vs. non-adjudicated site designated endpoints in major cardiovascular RCTs in which both were reported. Based on these data, we lay out a framework to determine which trials may warrant EA and where it may be redundant. The value of EA is likely greater when cardiovascular trials have nuanced primary endpoints, endpoint definitions that align poorly with practice, sub-optimal data completeness, greater operator variability, and lack of blinding. EA may not be needed if the primary endpoint is all-cause death or all-cause hospitalization. In contrast, EA is likely merited for more nuanced endpoints such as myocardial infarction, bleeding, worsening heart failure as an outpatient, unstable angina, or transient ischaemic attack. A risk-based approach to adjudication can potentially allow compromise between costs and accuracy. This would involve adjudication of a small proportion of events, with further adjudication done if inconsistencies are detected.

Deep Learning Model Transfer in Forest Mapping Using Multi-Source Satellite SAR and Optical Images

Deep learning (DL) models are gaining popularity in forest variable prediction using Earth observation (EO) images. However, in practical forest inventories, reference datasets are often represented by plot- or stand-level measurements, while high-quality representative wall-to-wall reference data for end-to-end training of DL models are rarely available. Transfer learning facilitates expansion of the use of deep learning models into areas with sub-optimal training data by allowing pretraining of the model in areas where high-quality teaching data are available. In this study, we perform a “model transfer” (or domain adaptation) of a pretrained DL model into a target area using plot-level measurements and compare performance versus other machine learning models. We use an earlier developed UNet based model (SeUNet) to demonstrate the approach on two distinct taiga sites with varying forest structure and composition. The examined SeUNet model uses multi-source EO data to predict forest height. Here, EO data are represented by a combination of Copernicus Sentinel-1 C-band SAR and Sentinel-2 multispectral images, ALOS-2 PALSAR-2 SAR mosaics and TanDEM-X bistatic interferometric radar data. The training study site is located in Finnish Lapland, while the target site is located in Southern Finland. By leveraging transfer learning, the SeUNet prediction achieved root mean squared error (RMSE) of 2.70 m and R2 of 0.882, considerably more accurate than traditional benchmark methods. We expect such forest-specific DL model transfer can be suitable also for other forest variables and other EO data sources that are sensitive to forest structure.

Highly accelerated multi-shot intravoxel incoherent motion diffusion-weighted imaging in brain enabled by parametric POCS-based multiplexed sensitivity encoding.

Recently, intravoxel incoherent motion (IVIM) diffusion-weighted imaging (DWI) has also been demonstrated as an imaging tool for applications in neurological and neurovascular diseases. However, the use of single-shot diffusion-weighted echo-planar imaging for IVIM DWI acquisition leads to suboptimal data quality: for instance, geometric distortion and deteriorated image quality at high spatial resolution. Although the recently commercialized multi-shot acquisition methods, such as multiplexed sensitivity encoding (MUSE), can attain high-resolution and high-quality DWI with signal-to-noise ratio (SNR) performance superior to that of the conventional parallel imaging method, the prolonged scan time associated with multi-shot acquisition is impractical for routine IVIM DWI. This study proposes an acquisition and reconstruction framework based on parametric-POCSMUSE to accelerate the four-shot IVIM DWI with 70% reduction of total scan time (13 min 8 s versus 4 min 8 s). First, the four-shot IVIM DWI scan with 17 b values was accelerated by acquiring only one segment per b value except for b values of 0 and 600 s/mm2 . Second, an IVIM-estimation scheme was integrated into the parametric-POCSMUSE to enable joint reconstruction of multi-b images from under-sampled four-shot IVIM DWI data. In vivo experiments on both healthy subjects and patients show that the proposed framework successfully produced multi-b DW images with significantly higher SNRs and lower reconstruction errors than did the conventional acceleration method based on parallel imaging. In addition, the IVIM quantitative maps estimated from the data produced by the proposed framework showed quality comparable to that of fully sampled MUSE-reconstructed images, suggesting that the proposed framework can enable highly accelerated multi-shot IVIM DWI without sacrificing data quality. In summary, the proposed framework can make multi-shot IVIM DWI feasible in a routine MRI examination, with reasonable scan time and improved geometric fidelity.

How valid is the commercially available tennis match analysis mobile application? Is it good enough?

ABSTRACT The present study examines the validity of the SwingVision application by comparing SwingVision’s data to criterion data. Also, we investigated the difference in recording angles of SwingVision (optimal and suboptimal). Six college students played four matches, and every match was recorded from two different angles. After the data collection, recorded videos were analysed by SwingVision and human analysts (criterion). A total of 1065 strokes were analysed in the agreement of SwingVision and human analysts. Cross-tabulation with a column proportion test, Cochran’s Q test, and Kappa statistics were utilised to demonstrate the association of categorical variables (stroke, hit depth, hit zone, bounce depth, bounce zone, spin, result) between the three data (i.e. optimal, suboptimal, and criterion). Repeated measures analysis of variance (ANOVA) and Pearson correlation were used to compare speed data. SwingVision data in most variables showed high proportional similarity and percent agreement with criterion data. Additionally, the optimal angle data had much more similar results to the criterion data than the suboptimal data. Therefore, this present study documented that SwingVision is trustworthy, and users should be aware of possible errors derived from angle differences.

Model-based trajectory stitching for improved behavioural cloning and its applications

Behavioural cloning (BC) is a commonly used imitation learning method to infer a sequential decision-making policy from expert demonstrations. However, when the quality of the data is not optimal, the resulting behavioural policy also performs sub-optimally once deployed. Recently, there has been a surge in offline reinforcement learning methods that hold the promise to extract high-quality policies from sub-optimal historical data. A common approach is to perform regularisation during training, encouraging updates during policy evaluation and/or policy improvement to stay close to the underlying data. In this work, we investigate whether an offline approach to improving the quality of the existing data can lead to improved behavioural policies without any changes in the BC algorithm. The proposed data improvement approach - Model-Based Trajectory Stitching (MBTS) - generates new trajectories (sequences of states and actions) by ‘stitching’ pairs of states that were disconnected in the original data and generating their connecting new action. By construction, these new transitions are guaranteed to be highly plausible according to probabilistic models of the environment, and to improve a state-value function. We demonstrate that the iterative process of replacing old trajectories with new ones incrementally improves the underlying behavioural policy. Extensive experimental results show that significant performance gains can be achieved using MBTS over BC policies extracted from the original data. Furthermore, using the D4RL benchmarking suite, we demonstrate that state-of-the-art results are obtained by combining MBTS with two existing offline learning methodologies reliant on BC, model-based offline planning (MBOP) and policy constraint (TD3+BC).

Benchmark Crystal Structure of Defect-Free Spinel ZnFe2O4.

Accurate structural models are of paramount importance for elucidating structure-property relationships in functional materials. Spinels (AB2O4) form a highly important family of materials with complex crystal structures, and subtle structural details have a critical bearing on understanding their physical properties. In some spinels, the space group symmetry is debated, and in general, point defects such as cation inversion and interstitials add complexity. Most studies of spinels concern powder materials, and this challenges deep structural characterization. In fact, most published spinel structures have dubious atomic displacement parameters (ADPs), which is a typical sign of problematic structural description in the refinement of diffraction data. Here, we use various X-ray and neutron diffraction techniques to establish a benchmark crystal structure for the essentially defect-free spinel ferrite ZnFe2O4, which is a widely studied frustrated magnet. It is shown that the appearance of Fd3̅m forbidden reflections in the ZnFe2O4 single-crystal neutron diffraction data is an artifact of multiple scattering rather than the loss of inversion symmetry. We then provide benchmark ADPs and demonstrate how strongly these parameters affect the refined cation inversion. The ADPs reported here may be used as reference data to test the soundness of refined structural models, possibly to constrain those based on suboptimal data quality, and thereby provide a more accurate fundamental understanding of the structure-property relationship in spinel-type materials.

Lessons Learnt from Rapid Implementation of Telehealth in a Paediatric Dietetics’ Outpatient Service: Is There a Silver Lining beyond the Coronavirus Pandemic to Support Patient-Centred Care?

The aim of this paper is to report on the differences in clinical and service engagement outcomes of patients attending the paediatric dietetic outpatient service at a major metropolitan hospital before the outbreak of the coronavirus (using face-to-face care) and during the pandemic (using telehealth). This paper also reports on learning lessons from the rapid implementation of telehealth in this service. This study collected pre- and post-coronavirus pandemic data from 44 paediatric patients. Data on outcomes pre- and post-implementation were analysed. There were no statistically significant differences between pre- and post-coronavirus data for growth (p = 0.92), adherence to dietetic recommendations (p = 0.08) or attendance rate (p = 1.00). This study also found a low uptake of virtual telehealth, technical issues and suboptimal anthropometric data collection associated with this modality of care. Telehealth was not associated with a change in clinical and service engagement outcomes. Thus, telehealth service delivery is not inferior to usual face–face dietetic care and has the potential to be a useful adjunct to usual nutrition care for paediatric health service users after the coronavirus pandemic.

Artificial Intelligence for Assisted Analysis of Eddy Current Data from Heat Exchangers with Non-Ferromagnetic Tubes

Oil & Gas and petrochemical industries need regular inspection of shell and tube heat exchangers to maintain performance. NDT inspection service providers are required to complete the data collection and analysis in the shortest possible time frame, but manual analysis for defects like corrosion or pitting is time-consuming and prone to errors. This leads to safety risks and unexpected downtime, and there's a shortage of experienced ECT level II/III certified analysts globally. The industry needs tools to simplify data interpretation and increase analysis efficiency while maintaining a consistent and repeatable level of confidence. While assisted analysis for tube ECT testing has existed for several years, most solutions have been developed for the nuclear industry and employ methods that become quickly limited when the test conditions are not well controlled or when the tube bundle information is incomplete These situations are very common outside the nuclear industry, and they often contribute to reducing the efficiency and reliability of data interpretation. This communication outlines the use of AI in various tasks of tube data analysis, from landmark localization to indication detection, to improve efficiency and confidence in the analysis process. A complete analysis system has been implemented and field data results demonstrate the system's value. Robustness to suboptimal data quality is also discussed.

Visualization of surgical maneuvers using intraoperative real-time volumetric optical coherence tomography.

Ophthalmic microsurgery is traditionally performed using stereomicroscopes and requires visualization and manipulation of sub-millimeter tissue structures with limited contrast. Optical coherence tomography (OCT) is a non-invasive imaging modality that can provide high-resolution, depth-resolved cross sections, and has become a valuable tool in clinical practice in ophthalmology. While there has been substantial progress in both research and commercialization efforts to bring OCT imaging into live surgery, its use is still somewhat limited due to factors such as low imaging speed, limited scan configurations, and suboptimal data visualization. In this paper we describe, to the best of our knowledge, the translation of the fastest swept-source intraoperative OCT system with real-time volumetric imaging with stereoscopic data visualization provided via a heads-up display into the operating room. Results from a sampling of human anterior segment and retinal surgeries chosen from 93 human surgeries using the system are shown and the benefits that this mode of intrasurgical OCT imaging provides are discussed.

Offline Imitation Learning with Suboptimal Demonstrations via Relaxed Distribution Matching

Offline imitation learning (IL) promises the ability to learn performant policies from pre-collected demonstrations without interactions with the environment. However, imitating behaviors fully offline typically requires numerous expert data. To tackle this issue, we study the setting where we have limited expert data and supplementary suboptimal data. In this case, a well-known issue is the distribution shift between the learned policy and the behavior policy that collects the offline data. Prior works mitigate this issue by regularizing the KL divergence between the stationary state-action distributions of the learned policy and the behavior policy. We argue that such constraints based on exact distribution matching can be overly conservative and hamper policy learning, especially when the imperfect offline data is highly suboptimal. To resolve this issue, we present RelaxDICE, which employs an asymmetrically-relaxed f-divergence for explicit support regularization. Specifically, instead of driving the learned policy to exactly match the behavior policy, we impose little penalty whenever the density ratio between their stationary state-action distributions is upper bounded by a constant. Note that such formulation leads to a nested min-max optimization problem, which causes instability in practice. RelaxDICE addresses this challenge by supporting a closed-form solution for the inner maximization problem. Extensive empirical study shows that our method significantly outperforms the best prior offline IL method in six standard continuous control environments with over 30% performance gain on average, across 22 settings where the imperfect dataset is highly suboptimal.

What Makes Public Sector Data Valuable for Development?

Abstract Data produced by the public sector can have transformational impacts on development outcomes through better targeting of resources, improved service delivery, cost savings, increased accountability, and more. Around the world, the amount of data produced by the public sector is increasing rapidly, but we argue the full potential of data to improve development outcomes has not been realized yet. We outline 12 features needed for data to generate greater value for development and present case studies substantiating these features. We argue that a key reason why the transformational value of data has not yet been realized is that suboptimal data—data not satisfying these 12 features—are being supplied. The features are that the data should be of adequate spatial and temporal coverage (complete, frequent, and timely), should be of high quality (accurate, comparable, and granular), should be easy to use (accessible, understandable, and interoperable), and should be safe to use (impartial, confidential, and appropriate).

Pitfalls in Proteomics - Peak Capacity: when is bigger better?

Proteomic analysis (the identification and quantification of proteins and peptides) is arguably the largest discipline utilising LC-MS based characterisation methods. Despite this dependency, LC-MS is largely relegated to simply a tool by users and therefore the fundamental knowledge behind chromatographic separations is rarely understood by analysts in proteomics laboratories. This lack of knowledge can unfortunately result in sub-optimal data quality. As such, this article will address the fundamental relationship between peak capacity (a proxy for number of protein/peptide identifications (1)) and stationary phase particle diameter, specifically in regards to conventional nano-scale LC (≤75 μm I.D columns) which is the workhorse instrumentation in peptide sequencing. Ultimately, we aim to give the LC-MS practitioner insight into the best particle dimensions for their application and why.

RNAprofiling 2.0: Enhanced Cluster Analysis of Structural Ensembles

Understanding the base pairing of an RNA sequence provides insight into its molecular structure. By mining suboptimal sampling data, RNAprofiling 1.0 identifies the dominant helices in low-energy secondary structures as features, organizes them into profiles which partition the Boltzmann sample, and highlights key similarities/differences among the most informative, i.e. selected, profiles in a graphical format. Version 2.0 enhances every step of this approach. First, the featured substructures are expanded from helices to stems. Second, profile selection includes low-frequency pairings similar to featured ones. In conjunction, these updates extend the utility of the method to sequences up to length 600, as evaluated over a sizable dataset. Third, relationships are visualized in a decision tree which highlights the most important structural differences. Finally, this cluster analysis is made accessible to experimental researchers in a portable format as an interactive webpage, permitting a much greater understanding of trade-offs among different possible base pairing combinations.

Establishing language laterality: does resting-state functional MRI help?

Task-based functional MRI (tb-fMRI) is now considered the standard, noninvasive technique in establishing language laterality in children for surgical planning. The evaluation can be limited due to several factors such as age, language barriers, and developmental and cognitive delays. Resting-state functional MRI (rs-fMRI) offers a potential path to establish language dominance without active task participation. The authors sought to compare the ability of rs-fMRI for language lateralization in the pediatric population with conventional tb-fMRI used as the gold standard. The authors performed a retrospective evaluation of all pediatric patients at a dedicated quaternary pediatric hospital who underwent tb-fMRI and rs-fMRI from 2019 to 2021 as part of the surgical workup for patients with seizures and brain tumors. Task-based fMRI language laterality was based on a patient's adequate performance on one or more of the following: sentence completion, verb generation, antonym generation, or passive listening tasks. Resting-state fMRI data were postprocessed using statistical parametric mapping, FMRIB Software Library, and FreeSurfer as described in the literature. The laterality index (LI) was calculated from the independent component (IC) with the highest Jaccard Index (JI) for the language mask. Additionally, the authors visually inspected the activation maps for two ICs with the highest JIs. The rs-fMRI LI of IC1 and the authors' image-based subjective interpretation of language lateralization were compared with tb-fMRI, which was considered the gold standard for this study. A retrospective search yielded 33 patients with language fMRI data. Eight patients were excluded (5 with suboptimal tb-fMRI and 3 with suboptimal rs-fMRI data). Twenty-five patients (age range 7-19 years, male/female ratio 15:10) were included in the study. The language laterality concordance between tb-fMRI and rs-fMRI ranged from 68% to 80% for assessment based on LI of independent component analysis with highest JI and for subjective evaluation by visual inspection of activation maps, respectively. The concordance rates between tb-fMRI and rs-fMRI of 68% to 80% show the limitation of rs-fMRI in determining language dominance. Resting-state fMRI should not be used as the sole method for language lateralization in clinical practice.

Data quality - The foundation of real-world studies.

Real-world data, routinely collected from multiple sources outside typical clinical research settings, are useful in conducting real-world studies. Sub-optimal and inconsistent data quality is a challenging issue which should be addressed whilst planning and conducting real-world studies. This brief review discusses the quality aspects of data needed for RWS.

Unreported protocol deviations - The tip of the research-berg.

Failure to stay within an ethics committee (EC)-approved protocol limits is termed protocol deviation or violation (PD/PV), depending on the seriousness of the transgression and its attendant risks and/or harms. PD/PVs arise in the post-approval phase of the research and are often missed. Current guidelines expect ECs to detect, report and recommend suitable actions such that research participants' risks and harms are mitigated, to the extent possible. Yenepoya Ethics Committee-1 conducted an internal audit of ongoing postgraduate dissertations involving human participants to assess the occurrence of PD/PVs. 54 out of 80 postgraduates responded to our request for filling out a self-reported checklist. These responses were followed up with physical verification of the protocol-related documents. Protocol transgressions were classified as non-compliance (administrative issues), protocol deviations (minor transgressions with minimal or less than minimal increase in attendant risk to participants) and protocol violations (serious transgressions with more than minimal increase in attendant risk to participants). The non-compliances included non-reporting for audit and non-reporting of PDs. Protocol deviations included non-conformance to EC validity, sample size, approved methodology, informed consent process and documentation and suboptimal data storage. No protocol violations were observed. We report PD/PVs from these 54 protocols - with our assessment on the negative impact it may have on scientific validity, harm to participants, EC functioning and credibility of the institution - in the hope that our readers appreciate this important aspect of the post-approval process in the functioning of an EC.