Future Point In Time Research Articles

A Composite Control Method Based on Model Predictive Control and a Disturbance Observer for the Acquisition, Tracking, and Pointing System

Nonlinear disturbances, such as friction, backlash, and base vibration, are the main factors restricting the further improvement of the dynamic performance of the Acquisition, Tracking and Pointing (ATP) system. However, the modeling and compensation of the time-varying nonlinearities are still challenging. In this paper, a compound control algorithm based on model predictive control (MPC) and a Proportional Multiple-Integral State-Augmented Kalman Filter (PMISAKF) disturbance observer is proposed to improve the disturbance rejection performance. Specifically, a PMISAKF is used to estimate the system’s state and external disturbances in real-time. These observed disturbances are then treated as known external inputs at the current or future time points and incorporated into the predictive model of MPC. This allows the optimization process to take these external factors into consideration, enabling the calculation of an optimal control strategy and achieving high pointing and tracking accuracy for nonlinear time-varying disturbances. Experimental results show that the proposed method can effectively improve the system’s anti-interference ability and tracking accuracy.

Bidirectional relationships between chronotype and sleep hygiene in children with and without parental history of alcohol use disorder

Study objectivesChildren with evening chronotype may be at risk for insufficient sleep because their chronotype makes it difficult to sustain healthy sleep habits. We evaluated bidirectional relationships between chronotype and sleep hygiene. MethodsTwo hundred forty-six children (n = 246 at T1, n = 200 at T2, n = 147 at T3), with a mean age of 9.9 (SD=1.4) at T1, participated in a longitudinal study on sleep and substance use. Participants either had a parental history of alcohol use disorder or were matched controls. The Adolescent Sleep Hygiene Scale measured sleep hygiene. Chronotype was measured using the Morningness/Eveningness Questionnaire. We used random intercept cross-lagged panel models to examine longitudinal relations between chronotype and sleep hygiene across three time points, each approximately 1 year apart. ResultsChronotype at T1 predicted sleep hygiene at T2 (b=0.38, p < .05). Chronotype at T2 predicted sleep hygiene at T3 (b=0.38, p < .05). T1 Sleep Hygiene predicted chronotype at T2 (b=0.27, p < .05). T2 Sleep Hygiene predicted chronotype at T3 (b=0.24, p < .05). Chronotype also predicted itself over time (T1-T2: b=0.31, p < .05; T2-T3: b=0.31, p < .05). Sleep hygiene did not predict itself at future time points. Parental history of alcohol use disorder did not predict sleep hygiene or chronotype. ConclusionsThere is a bidirectional relationship between chronotype and sleep hygiene; more eveningness predicts poorer sleep hygiene at a later time point, and poorer sleep hygiene predicts more eveningness later. Sleep hygiene is not stable over time and may be a more modifiable target for public health interventions than chronotype.

ADMNet: An adaptive downsampling multi-frequency multi-channel network for long-term time series forecasting

Long-term time series forecasting finds widespread applications in various domains such as energy, finance, and transportation. Decomposing time series into sub-sequences with distinct temporal relationships (periods) for analysis and modeling is an effective approach for long-term time series prediction. However, conventional decomposition techniques often rely on fixed-size parameter kernels for sliding averages, leading to inaccurate and unreasonable captures of the underlying periodicities in the time series. Additionally, it is essential to acknowledge that time series data often exhibits multi-periodic characteristics. In many time series prediction tasks, the value at a future time point is influenced by multiple periods within the time series. To accurately and flexibly capture the periodicities of time series, we propose an enhanced adaptive cyclic feature recognizer to automatically identify the periodic lengths for sliding average parameter kernel sizing. To fully exploit the multiple periodicities inherent in time series, we merge and smooth the sequences of features corresponding to the identified cycles, obtaining two frequency-blended cyclic feature fusion terms. Furthermore, we extract high-frequency random components to preserve finer details. Finally, we individually model the three cyclic feature fusion terms. In summary, we introduce the Adaptive Multi-Frequency Multi-Channel Network (ADMNet) designed to autonomously capture multi-periodic features. Experimental results show that, compared to the current state-of-the-art seven benchmark models, the proposed model achieves an average reduction in mean squared error (MSE) ranging from 6.19% to 22.75% across eight datasets, including ETT, Traffic, and Weather. This indicates that our model delivers superior predictive performance on multiple real-world datasets.

Spatiotemporal modeling of the potential impact of climate change on shifts in bioclimatic zones in Morocco.

This study aims to contribute to the understanding of the impact of climate change on bioclimatic zones in Morocco, providing insights into potential shifts and emphasizing the need for adaptation measures to protect vulnerable species and ecosystems. To achieve this, we utilized eight general circulation models (GCMs) to simulate climate conditions under two representative concentration scenarios (RCP4.5 and RCP8.5) for two future time points (2050 and 2070). The modeling of bioclimatic zone shifts was accomplished through the implementation of the random forest (RF) algorithm. Our findings indicate that the subhumid and humid areas are expected to experience the most significant shifts, particularly toward the semi-arid zone. Shifts from subhumid to semi-arid were the most pronounced, ranging from 17.91% (RCP8.5 in 2070) to 25.68% (RCP8.5 in 2050), while shifts from humid to semi-arid ranged from 10.16% (RCP4.5 in 2050) to 22.27% (RCP8.5 in 2070). The Saharan and arid zones are expected to be the least affected, with less than 1% and 11% of their original extent expected to change, respectively. Moreover, our results suggest that forest species such as Atlas cedar and oaks are among the most vulnerable to these shifts. Overall, this study highlights the inevitability of climate change's impact on Moroccan ecosystems and provides a basis for adaptation measures, especially considering the species adapted to the bioclimatic conditions that will dominate the respective affected regions.

Modelling the age distribution of longevity leaders

Human longevity leaders with remarkably long lifespan play a crucial role in the advancement of longevity research. In this paper, we propose a stochastic model to describe the evolution of the age of the oldest person in the world by a Markov process, in which we assume that the births of the individuals follow a Poisson process with increasing intensity, lifespans of individuals are independent and can be characterized by a gamma–Gompertz distribution with time-dependent parameters. We utilize a dataset of the world’s oldest person title holders since 1955, and we compute the maximum likelihood estimate for the parameters iteratively by numerical integration. Based on our preliminary estimates, the model provides a good fit to the data and shows that the age of the oldest person alive increases over time in the future. The estimated parameters enable us to describe the distribution of the age of the record holder process at a future time point.

Lite approaches for long-range multi-step water quality prediction

Forecasting accurate water quality is very important in aquaculture, environment monitoring, and many other applications. Many internal and external factors influence water quality. Therefore, water quality parameters exhibit complex time series characteristics. Consequently, long-range accurate prediction of water quality parameters suffers from poor propagation of information from past timepoints to further future timepoints. Moreover, to synchronise the prediction model with the changes in the time series characteristics, periodic retraining of the prediction model is required and such retraining is to be done on resource-restricted computation devices. In this work, we present a low-cost training approach to improve long-range multi-step water quality prediction. We train a short-range predictor to save training effort. Then, we strive to achieve and/or improve long-range prediction using multi-step iterative ensembling during inference. Experimental results on 9 water quality datasets demonstrate that the proposed method achieves significantly lower error than the existing state-of-the-art approaches. Our approach significantly outperforms the existing approaches in several standard metrics, even in the case of future timepoints at long distances.

Collective Intelligence-Based Participatory COVID-19 Surveillance in Accra, Ghana: Pilot Mixed Methods Study.

Infectious disease surveillance is difficult in many low- and middle-income countries. Information market (IM)-based participatory surveillance is a crowdsourcing method that encourages individuals to actively report health symptoms and observed trends by trading web-based virtual "stocks" with payoffs tied to a future event. This study aims to assess the feasibility and acceptability of a tailored IM surveillance system to monitor population-level COVID-19 outcomes in Accra, Ghana. We designed and evaluated a prediction markets IM system from October to December 2021 using a mixed methods study approach. Health care workers and community volunteers aged ≥18 years living in Accra participated in the pilot trading. Participants received 10,000 virtual credits to trade on 12 questions on COVID-19-related outcomes. Payoffs were tied to the cost estimation of new and cumulative cases in the region (Greater Accra) and nationwide (Ghana) at specified future time points. Questions included the number of new COVID-19 cases, the number of people likely to get the COVID-19 vaccination, and the total number of COVID-19 cases in Ghana by the end of the year. Phone credits were awarded based on the tally of virtual credits left and the participant's percentile ranking. Data collected included age, occupation, and trading frequency. In-depth interviews explored the reasons and factors associated with participants' user journey experience, barriers to system use, and willingness to use IM systems in the future. Trading frequency was assessed using trend analysis, and ordinary least squares regression analysis was conducted to determine the factors associated with trading at least once. Of the 105 eligible participants invited, 21 (84%) traded at least once on the platform. Questions estimating the national-level number of COVID-19 cases received 13 to 19 trades, and obtaining COVID-19-related information mainly from television and radio was associated with less likelihood of trading (marginal effect: -0.184). Individuals aged <30 years traded 7.5 times more and earned GH ¢134.1 (US $11.7) more in rewards than those aged >30 years (marginal effect: 0.0135). Implementing the IM surveillance was feasible; all 21 participants who traded found using IM for COVID-19 surveillance acceptable. Active trading by friends with communal discussion and a strong onboarding process facilitated participation. The lack of bidirectional communication on social media and technical difficulties were key barriers. Using an IM system for disease surveillance is feasible and acceptable in Ghana. This approach shows promise as a cost-effective source of information on disease trends in low- and middle-income countries where surveillance is underdeveloped, but further studies are needed to optimize its use.

Development of predictive model for predicting postoperative BMI and optimize bariatric surgery: a single center pilot study

BackgroundThe pilot study addresses the challenge of predicting postoperative outcomes, particularly body mass index (BMI) trajectories, following bariatric surgery. The complexity of this task makes preoperative personalized obesity treatment challenging. ObjectivesTo develop and validate sophisticated machine learning (ML) algorithms capable of accurately forecasting BMI reductions up to 5 years following bariatric surgery aiming to enhance planning and postoperative care. The secondary goal involves the creation of an accessible web-based calculator for healthcare professionals. This is the first article that compares these methods in BMI prediction. SettingThe study was carried out from January 2012 to December 2021 at GZOAdipositas Surgery Center, Switzerland. Preoperatively, data for 1004 patients were available. Six months postoperatively, data for 1098 patients were available. For the time points 12 months, 18 months, 2 years, 3 years, 4 years, and 5 years the following number of follow-ups were available: 971, 898, 829, 693, 589, and 453. MethodsWe conducted a comprehensive retrospective review of adult patients who underwent bariatric surgery (Roux-en-Y gastric bypass or sleeve gastrectomy), focusing on individuals with preoperative and postoperative data. Patients with certain preoperative conditions and those lacking complete data sets were excluded. Additional exclusion criteria were patients with incomplete data or follow-up, pregnancy during the follow-up period, or preoperative BMI ≤30 kg/m2. ResultsThis study analyzed 1104 patients, with 883 used for model training and 221 for final evaluation, the study achieved reliable predictive capabilities, as measured by root mean square error (RMSE). The RMSE values for three tasks were 2.17 (predicting next BMI value), 1.71 (predicting BMI at any future time point), and 3.49 (predicting the 5-year postoperative BMI curve). These results were showcased through a web application, enhancing clinical accessibility and decision-making. ConclusionThis study highlights the potential of ML to significantly improve bariatric surgical outcomes and overall healthcare efficiency through precise BMI predictions and personalized intervention strategies.

Tree of life metabarcoding can serve as a biotic benchmark for shifting baselines in urbanized estuaries

Urbanization of estuaries drastically changed existing shorelines and bathymetric contours, in turn modifying habitat for marine foundational species that host critical biodiversity. And yet we lack approaches to characterize a significant fraction of the biota that inhabit these ecosystems on time scales that align with rates of urbanization. Environmental DNA (or eDNA) metabarcoding that combines multiple assays targeting a broad range of taxonomic groups can provide a solution, but we need to determine whether the biological communities it detects ally with different habitats in these changing aquatic environments. In this study, we tested whether tree of life metabarcoding (ToL-metabarcoding) data extracted from filtered seawater samples correlated with four known geomorphic habitat zones across a heavily urbanized estuary (Sydney Harbour, Australia). Using this method, we substantially expanded our knowledge on the composition and spatial distribution of marine biodiversity across the tree of life in Sydney Harbour, particularly for organisms where existing records are sparse. Excluding terrestrial DNA inputs, we identified significant effects of both distance from the mouth of Sydney Harbour and geomorphic zone on biological community structure in the ToL-metabarcoding dataset (entire community), as well as in each of the taxonomic subgroups that we considered (fish, macroinvertebrates, algae and aquatic plants, bacteria). This effect appeared to be driven by taxa as a collective versus a few individual taxa, with each taxon explaining no more than 0.62% of the variation between geomorphic zones. Similarly, taxonomic richness was significantly higher within geomorphic zones with large sample sizes, but also decreased by 1% with each additional kilometer from the estuary mouth, a result consistent with a reduction in tidal inputs and available habitat in upper catchments. Based on these results, we suggest that ToL-metabarcoding can be used to benchmark biological monitoring in other urbanized estuaries globally, and in Sydney Harbour at future time points based on detection of bioindicators across the tree of life. We also suggest that robust biotic snapshots can be archived following extensive curation of taxonomic assignments that incorporates ecological affinities, supported by records from relevant and regional biodiversity repositories.

Multi-detector fusion and Bayesian smoothing for tracking viral and chromatin structures

Automatic tracking of viral and intracellular structures displayed as spots with varying sizes in fluorescence microscopy images is an important task to quantify cellular processes. We propose a novel probabilistic tracking approach for multiple particle tracking based on multi-detector and multi-scale data fusion as well as Bayesian smoothing. The approach integrates results from multiple detectors using a novel intensity-based covariance intersection method which takes into account information about the image intensities, positions, and uncertainties. The method ensures a consistent estimate of multiple fused particle detections and does not require an optimization step. Our probabilistic tracking approach performs data fusion of detections from classical and deep learning methods as well as exploits single-scale and multi-scale detections. In addition, we use Bayesian smoothing to fuse information of predictions from both past and future time points. We evaluated our approach using image data of the Particle Tracking Challenge and achieved state-of-the-art results or outperformed previous methods. Our method was also assessed on challenging live cell fluorescence microscopy image data of viral and cellular proteins expressed in hepatitis C virus-infected cells and chromatin structures in non-infected cells, acquired at different spatial–temporal resolutions. We found that the proposed approach outperforms existing methods.

Deep learning for 4D modeling of glioblastoma multiforme with tumor treating fields (TTFields) therapy.

e14032 Background: Prediction of future tumor growth and recurrence are critical in the management of patients with glioblastoma multiforme (GBM). This study develops a deep learning method for estimating future areas of GBM spread across multiple timepoints using MRI in patients receiving tumor treating fields (TTFields) therapy. Methods: A single institutional database was queried to identify adult patients with histologically confirmed newly diagnosed and/or recurrent GBM undergoing TTFields therapy. For newly diagnosed GBM, patients received TTFields with temozolomide after maximal debulking surgery and chemoradiation therapy. For recurrent GBM, patients received TTFields as monotherapy. For each patient, all serial follow-up MRI exams were obtained, including T1 (pre-/post-contrast), T2, and T2/FLAIR sequences. On each exam, all regions of enhancing tumor core (excluding peritumoral edema, necrotic core, or resection cavity) were delineated and aligned across time points using nonlinear deformable registration. For any given pair of serial exams, a convolutional neural network (CNN) was designed to predict future GBM tumor on the follow-up exam given the precursor exam and time interval between the studies. The model is implemented as a 3D encoder-decoder architecture yielding a dense per-voxel prediction of future tumor probability optimized using a binary cross-entropy loss function. Class weights were used to develop high sensitivity and high positive predictive value (PPV) model variants. To generate final logit scores, time information is concatenated to the penultimate feature map as an additional feature channel, allowing the model to calibrate each estimate based on elapsed time between any pair of exams. Upon convergence, a 4D learned representation allows for prediction of spatial distribution of GBM tumor at any future time point. Results: A total of 123 patients (1112 total MR exams) were identified. For any given single patient, a median of 6 follow-up exams (IQR 2.5-12.5) at a median interval follow-up time of 46 days (IQR 15.75-63 days) between exams was observed. Upon five-fold cross-validation, the model demonstrated a 0.44 Dice score overlap between predicted and true areas of future GBM tumor growth. The high sensitivity model yielded a per-voxel sensitivity of 0.91 (IQR 0.77-0.99) and PPV of 0.26 (IQR 0.17 to 0.32), while the high PPV model yielded a per-voxel sensitivity of 0.14 (IQR 0.00 to 0.46) and PPV of 0.75 (IQR 0.56 to 0.94). Upon visual confirmation, model predictions across incremental time values for any given exam yielded expected gradual growth of tumor over time. Conclusions: A deep learning model can accurately predict future areas of GBM tumor growth in patients receiving TTFields therapy, with optimal performance that may be calibrated for high sensitivity or high PPV based on clinical use case.

New joint model for the dynamic prediction of heart failure in phospholamban (PLN) p.(Arg14del) positive individuals

Abstract Background PLN p.(Arg14del) positive individuals are at risk for developing arrhythmogenic cardiomyopathy which can cause severe heart failure. Future gene therapy could be lifesaving and is rapidly advancing. It is therefore of importance to identify eligible patients for this evolving treatment, with individuals risk prediction playing an imperative role in this process. Purpose To develop a dynamic prediction model for the individual risk prediction of heart failure in PLN p.(Arg14del) positive individuals. Method Data were collected of 564 PLN p.(Arg14del) positive individuals with no baseline history of heart failure or myocardial infarction. During a median follow-up of 6.3 years (interquartile range 2.9-10.0), 74 individuals experienced heart failure, defined as a composite endpoint of heart failure hospitalization, left ventricle assist device implantation, heart transplantation and heart failure-related death. For the individual prediction of time to heart failure we used a joint model which simultaneously estimates a longitudinal model for repeatedly measured left ventricle ejection fraction (LVEF) during follow-up and a Cox regression model for the hazards of the heart failure endpoint. Using the longitudinal model we predicted the LVEF-values of all patients at future time points using the estimated random-effects and these predicted LVEF values were used as a time-dependent covariate in the Cox model which also contained age at start of follow-up as predictor variable. Results The joint model trained in this study had a very good performance in discriminating the individuals at risk for heart failure, with the 5-year mean AUC of 0.86 and 10-year mean AUC of 0.84. The LVEF had a hazard ratio of 0.89 (95% confidence interval (CI) 0.86-0.93, p&amp;lt;0.001), and age at baseline 1.04 (95% CI 1.03-1.06], p&amp;lt;0.001), meaning a lower LVEF and higher age at presentation were associated with heart failure. Figure 1 shows an example of an individual risk prediction with on the x-axis the follow-up time in years, on the y-axis (left) the LVEF in % and on the y-axis (right) the predicted survival probability. Conclusion Individual risk prediction with a simple model including only LVEF and age is of great importance for identifying PLN p.(Arg14del) positive individuals who are at risk of developing heart failure. The introduction of this novel joint prediction model is the first step in identifying individuals at risk for heart failure and may aid patient selection for future gene therapy interventions.

Abstract PO5-03-06: Optimizing therapeutic regimens via digital twins to improve triple negative breast cancer response to neoadjuvant therapy

Abstract Introduction: Neoadjuvant systemic therapy (NAT) has been the standard-of-care of stage II-III, locally advanced triple-negative breast cancer (TNBC). However, about 50% of TNBC patients achieve a pathological complete response (pCR) to conventional neoadjuvant chemotherapy (NAC)1. Recently approved combination of NAC with the immunotherapy pembrolizumab, has improved the pCR rate by 7.5%2, although with a 44% risk of immune-related adverse reaction3. Aside from the need to develop new therapies with higher efficacy and lower toxicity, a critical barrier to improving TNBC response is the lack of rigorous ways to personally tailor therapeutic regimens. We seek to address this challenge by employing digital twins (i.e., mathematical models that provide virtual representation of individual patients and predict the changes at future time points) to systematically evaluate individual TNBC patient’s response to different NAT regimens, thereby patient-specifically optimizing treatments. Methods: A TNBC cohort (n = 139) from the ARTEMIS trial (NCT02276433)4 was used for this study. All patients received 4 cycles of Adriamycin/Cytoxan (A/C) every 2 weeks, followed by 12 cycles of weekly Taxol or experimental therapy in Phase II trials. All patients had surgery after NAT and post-surgical pathology to assess response status. For each patient, longitudinal MRIs were collected before, during, and after A/C. We have developed digital twins to integrate the longitudinal MRIs with a mechanism-based model to accurately predict TNBC response5. The model was based on a reaction-diffusion equation that describes the change in tumor cellularity due to migration, proliferation, and drug-induced death. With parameters personalized using MRIs, the patient-specific model (i.e., digital twin) significantly improved the accuracy to predict pCR5. We investigated patient-specific treatment optimization on 37 (19 pCR, 18 non-pCR) chemo-sensitive patients (≥ 70% volume decrease after A/C) who received only NAC. We evaluated the effect of altering the A/C/Taxol schedules on patient response. Specifically, each patient’s digital twin was used to predict the patient’s response to 128 clinically reasonable schedules of A/C/Taxol; i.e., 8 candidate A/C schedules combining with 16 candidate Taxol schedules (Table 1). The predicted response (pCR or non-pCR) from each alternative schedule was compared to the patient response from the actual treatment. Results: Without changing the total dose, shortening the duration of A/C/Taxol administration increased the treatment efficacy. The effectiveness of altering the schedules varied substantially in different patients. In particular, 8 patients who had non-pCR responses to their actual treatment were predicted to achieve pCR with the dense-dose Taxol (i.e., 4 cycles Taxol, 2 weeks per cycle), indicating a 21.62% improvement of pCR rate in the cohort. Discussion and conclusion: The preliminary results with our digital twin approach provided a unique opportunity of improving TNBC response to NAT through patient-specific optimization of therapeutic schedules. The ongoing effort focuses on accounting for toxicity and investigating the effects of altering therapy types and doses in combination with the schedules on the patient response. [1] Spring et al., Clin Cancer Res, 2020. [2] Schmid et al., NEJM, 2022. [3] Shah et al., Clin Cancer Res, 2022. [4] Yam et al., Clin Cancer Res, 2021. [5] Wu et al., Cancer Res, 2022. Table 1. Candidate therapeutic schedules Citation Format: Chengyue Wu, Ernesto Lima, Casey Stowers, Zhan Xu, Clinton Yam, Jong Bum Son, Jingfei Ma, Gaiane Rauch, Thomas Yankeelov. Optimizing therapeutic regimens via digital twins to improve triple negative breast cancer response to neoadjuvant therapy [abstract]. In: Proceedings of the 2023 San Antonio Breast Cancer Symposium; 2023 Dec 5-9; San Antonio, TX. Philadelphia (PA): AACR; Cancer Res 2024;84(9 Suppl):Abstract nr PO5-03-06.

Infants’ brain responses to social interaction predict future language growth

In face-to-face interactions with infants, human adults exhibit a species-specific communicative signal. Adults present a distinctive "social ensemble": they use infant-directed speech (parentese), respond contingently to infants' actions and vocalizations, and react positively through mutual eye-gaze and smiling. Studies suggest that this social ensemble is essential for initial language learning. Our hypothesis is that the social ensemble attracts attentional systems to speech and that sensorimotor systems prepare infants to respond vocally, both of which advance language learning. Using infant magnetoencephalography (MEG), we measure 5-month-old infants' neural responses during live verbal face-to-face (F2F) interaction with an adult (social condition) and during a control (nonsocial condition) in which the adult turns away from the infant to speak to another person. Using a longitudinal design, we tested whether infants' brain responses to these conditions at 5months of age predicted their language growth at five future time points. Brain areas involved in attention (right hemisphere inferior frontal, right hemisphere superior temporal, and right hemisphere inferior parietal) show significantly higher theta activity in the social versus nonsocial condition. Critical to theory, we found that infants' neural activity in response to F2F interaction in attentional and sensorimotor regions significantly predicted future language development into the third year of life, more than 2 years after the initial measurements. We develop a view of early language acquisition that underscores the centrality of the social ensemble, and we offer new insight into the neurobiological components that link infants' language learning to their early brain functioning during social interaction.

Learning Spatio-Temporal Model of Disease Progression With NeuralODEs From Longitudinal Volumetric Data.

Robust forecasting of the future anatomical changes inflicted by an ongoing disease is an extremely challenging task that is out of grasp even for experienced healthcare professionals. Such a capability, however, is of great importance since it can improve patient management by providing information on the speed of disease progression already at the admission stage, or it can enrich the clinical trials with fast progressors and avoid the need for control arms by the means of digital twins. In this work, we develop a deep learning method that models the evolution of age-related disease by processing a single medical scan and providing a segmentation of the target anatomy at a requested future point in time. Our method represents a time-invariant physical process and solves a large-scale problem of modeling temporal pixel-level changes utilizing NeuralODEs. In addition, we demonstrate the approaches to incorporate the prior domain-specific constraints into our method and define temporal Dice loss for learning temporal objectives. To evaluate the applicability of our approach across different age-related diseases and imaging modalities, we developed and tested the proposed method on the datasets with 967 retinal OCT volumes of 100 patients with Geographic Atrophy and 2823 brain MRI volumes of 633 patients with Alzheimer's Disease. For Geographic Atrophy, the proposed method outperformed the related baseline models in the atrophy growth prediction. For Alzheimer's Disease, the proposed method demonstrated remarkable performance in predicting the brain ventricle changes induced by the disease, achieving the state-of-the-art result on TADPOLE cross-sectional prediction challenge dataset.

Utility in Time Description in Priority Best–Worst Discrete Choice Models: An Empirical Evaluation Using Flynn’s Data

Discrete choice models (DCMs) are applied in many fields and in the statistical modelling of consumer behavior. This paper focuses on a form of choice experiment, best–worst scaling in discrete choice experiments (DCEs), and the transition probability of a choice of a consumer over time. The analysis was conducted by using simulated data (choice pairs) based on data from Flynn’s (2007) ‘Quality of Life Experiment’. Most of the traditional approaches assume the choice alternatives are mutually exclusive over time, which is a questionable assumption. We introduced a new copula-based model (CO-CUB) for the transition probability, which can handle the dependent structure of best–worst choices while applying a very practical constraint. We used a conditional logit model to calculate the utility at consecutive time points and spread it to future time points under dynamic programming. We suggest that the CO-CUB transition probability algorithm is a novel way to analyze and predict choices in future time points by expressing human choice behavior. The numerical results inform decision making, help formulate strategy and learning algorithms under dynamic utility in time for best–worst DCEs.

Intraday residual transfer learning in minimally observed power distribution networks dynamic state estimation

Abstract Traditionally, electricity distribution networks were designed for unidirectional power flow without the need to accommodate generation installed at the point of use. However, with the increase in Distributed Energy Resources and other Low Carbon Technologies, the role of distribution networks is changing. This shift brings challenges, including the need for intensive metering and more frequent reconfiguration to identify threats from voltage and thermal violations. Mitigating action through reconfiguration is informed by State Estimation, which is especially challenging for low voltage distribution networks where the constraints of low observability, non-linear load relationships, and highly unbalanced systems all contribute to the difficulty of producing accurate state estimates. To counter low observability, this paper proposes the application of a novel transfer learning methodology, based upon the concept of conditional online Bayesian transfer, to make forward predictions of bus pseudo-measurements. Day ahead load forecasts at a fully observed point on the network are adjusted using the intraday residuals at other points in the network to provide them with load forecasts without the need for a complete set of forecast models at all substations. These form pseudo-measurements that then inform the state estimates at future time points. This methodology is demonstrated on both a representative IEEE Test network and on an actual GB 11 kV feeder network.

Using Bayesian Neural Networks to Select Features and Compute Credible Intervals for Personalized Survival Prediction.

An Individual Survival Distribution (ISD) models a patient's personalized survival probability at all future time points. Previously, ISD models have been shown to produce accurate and personalized survival estimates (for example, time to relapse or to death) in several clinical applications. However, off-the-shelf neural-network-based ISD models are usually opaque models due to their limited support for meaningful feature selection and uncertainty estimation, which hinders their wide clinical adoption. Here, we introduce a Bayesian-neural-network-based ISD (BNN-ISD) model that produces accurate survival estimates but also quantifies the uncertainty in model's parameter estimation, which can be used to (1) rank the importance of the input features to support feature selection and (2) compute credible intervals around ISDs for clinicians to assess the model's confidence in its prediction. Our BNN-ISD model utilized sparsity-inducing priors to learn a sparse set of weights to enable feature selection. We provide empirical evidence, on 2 synthetic and 3 real-world clinical datasets, that BNN-ISD system can effectively select meaningful features and compute trustworthy credible intervals of the survival distribution for each patient. We observed that our approach accurately recovers feature importance in the synthetic datasets and selects meaningful features for the real-world clinical data as well, while also achieving state-of-the-art survival prediction performance. We also show that these credible regions can aid in clinical decision-making by providing a gauge of the uncertainty of the estimated ISD curves.

Accurate personalized survival prediction for amyotrophic lateral sclerosis patients

Amyotrophic Lateral Sclerosis (ALS) is a rapidly progressive neurodegenerative disease. Accurately predicting the survival time for ALS patients can help patients and clinicians to plan for future treatment and care. We describe the application of a machine-learned tool that incorporates clinical features and cortical thickness from brain magnetic resonance (MR) images to estimate the time until a composite respiratory failure event for ALS patients, and presents the prediction as individual survival distributions (ISDs). These ISDs provide the probability of survival (none of the respiratory failures) at multiple future time points, for each individual patient. Our learner considers several survival prediction models, and selects the best model to provide predictions. We evaluate our learned model using the mean absolute error margin (MAE-margin), a modified version of mean absolute error that handles data with censored outcomes. We show that our tool can provide helpful information for patients and clinicians in planning future treatment.

Two-Stage Model-Based Predicting PV Generation with the Conjugation of IoT Sensor Data.

This paper proposes a novel short-term photovoltaic voltage (PV) prediction scheme using IoT sensor data with the two-stage neural network model. It is efficient to use environmental data provided by the meteorological agency to predict future PV generation. However, such environmental data represent the average value of the wide area, and there is a limitation in detecting environmental changes in the specific area where the solar panel is installed. In order to solve such issues, it is essential to establish IoT sensor data to detect environmental changes in the specific area. However, most conventional research focuses only on the efficiency of IoT sensor data without taking into account the timing of data acquisition from the sensors. In real-world scenarios, IoT sensor data is not available precisely when needed for predictions. Therefore, it is necessary to predict the IoT data first and then use it to forecast PV generation. In this paper, we propose a two-stage model to achieve high-accuracy prediction results. In the first stage, we use predicted environmental data to access IoT sensor data in the desired future time point. In the second stage, the predicted IoT sensors and environmental data are used to predict PV generation. Here, we determine the appropriate prediction scheme at each stage by analyzing the model characteristics to increase prediction accuracy. In addition, we show that the proposed prediction scheme could increase prediction accuracy by more than 12% compared to the baseline scheme that only uses a meteorological agency to predict PV generation.