Prediction Of Distribution Research Articles

Economic policy uncertainty and capital flows' tail risk in China

Since the financial crisis, uncertainty has been a focus of attention in the financial sector. As the two-way opening-up of China's financial market continues to accelerate, special attention must be paid to the risks and challenges arising from cross-border capital flows to maintain financial market stability in China. This study uses predictive quantile regression and skewed t-distribution fitting method to examine the impact of Economic Policy Uncertainty (EPU) on the full probability distribution and tail risks of capital flows. Additionally, we apply the “Capital Flows at Risk (CFaR)” analysis framework from the new paradigm of macroeconomic research to the situation in China, to study the impact of EPU on CFaR. We find that EPU significantly affects the predictive distribution of capital flows. When EPU is rising, the downside risk of capital flows, that is, the left-tail risk of capital flows, will increase; however, when EPU is declining, the upside risk of capital flows, that is, the right-tail risk of capital flows, will increase. In recent years, the scale of China's CFaR has shown a fluctuating upward trend, and EPU has increased the scale of CFaR in the country. This study is significant for stabilizing international capital flows and preventing and defusing the external financial shocks of cross-border capital inflows and outflows.

Constructing density forecasts from quantile regressions: Multimodality in macrofinancial dynamics

SummaryQuantile regression methods are increasingly used to forecast tail risks and uncertainties in macroeconomic outcomes. This paper reconsiders how to construct predictive densities from quantile regressions. We compare a popular two‐step approach that fits a specific parametric density to the quantile forecasts with a nonparametric alternative that lets the “data speak.” Simulation evidence and an application revisiting GDP growth uncertainties in the United States demonstrate the flexibility of the nonparametric approach when constructing density forecasts from both frequentist and Bayesian quantile regressions. They identify its ability to unmask deviations from symmetrical and unimodal densities. The dominant macroeconomic narrative becomes one of the evolution, over the business cycle, of multimodalities rather than asymmetries in the predictive distribution of GDP growth when conditioned on financial conditions.

On time-horizons based post-processing of flow forecasts

Post-processing raw stream flow forecasts are generally understood as estimating the univariate predictive density of stage or discharge values at selected future time steps, which is conditional on a single or multiple streamflow forecasts and observations up to the forecast start time to. The predictive density indicates to a forecaster in the most comprehensive way which flood level is likely to be expected. To this end, a variety of post-processing methods were proposed, which have respective strengths and weaknesses. These methods focus near-exclusively on the probabilistic forecast of the predictand at a single set future time ti, without addressing the predictive capability over the sequence of temporal sub-horizons (to, t1] ⊂ (to, t2] ⊂ … ⊂ (to, tk] nested into the overall forecast horizon. Here, we demonstrate the advantages of time-horizon dependent processing of streamflow forecasts, which evaluates the evolution of the predictive density over the sub-horizons by considering the temporal correlation among forecast ensemble members in addition to their cross-correlation with observations. The resulting probabilistic forecast consists of a multivariate distribution of stages and/or discharges at lagged forecasting times. These multivariate predictive distributions have the advantage of providing the likelihood of exceeding a critical threshold during the forecasting horizon while simultaneously offering valuable insights into the expected time of such exceedance. This approach supports not only decisions on issuing timely flood warnings but also the planing and roll-out of mitigating actions.

Summarizing quantile estimates of weight at length to compare condition: Geographic and temporal variation in Arctic Grayling

AbstractObjectiveWe developed a novel, simple approach to summarize variation in quantile regression estimates of weight at length across multiple levels of grouping factors (locations or years) and demonstrate its utility for comparing fish condition using Arctic Grayling Thymallus arcticus data. We compared condition among 28 geographic locations of Arctic Grayling populations and among 12 and 25 years for two populations.MethodsWe summarized variation among locations and among years within locations using percentiles (maximum, 75th, 50th, 25th, and minimum) of predicted quantiles of weight at length, allowing us to simplify the distributions of predicted quantiles of weight into highest to lowest fourths (quartiles) across locations or years. One part of this summary information, 75th percentiles of median (0.50 quantile) predicted weights at length across populations, is identical to summary statistics used to develop standard weight equations for computing relative weights. However, our approach avoids biases inherent in fitting a single standard weight equation and can accommodate among‐fish variation in weight at length via multiple quantiles when developing reference standards for comparative or diagnostic purposes. A selected new weight at length can be compared with the distribution of quantile predictions among locations in terms of weights or as proportionate weights similar to relative weight.ResultArctic Grayling in 28 geographic locations, from the southern limit of their distribution in Montana and Wyoming north through Canada and Alaska to the Arctic coast, had considerable variation in allometric exponents in weight–length relationships and in predicted weights at length. Predicted weights of Arctic Grayling for heavier fish (0.90 quantiles) in populations with less (lower quartiles) to greater (upper quartiles) weights were 127–194 g and 336–510 g at total lengths = 250 and 350 mm, respectively. Lighter fish (0.10 quantiles) in populations with less to greater weights were 100–133 g and 259–364 g, at the same total lengths, respectively. We provide statistical code and tabled values that can be used as reference standards as a replacement for standard weight equations for comparing new samples of fishes. We also quantified annual variation in condition within two intensively sampled populations, the Kuparuk River in Alaska (25 years, 1985–2012) and Red Rock Creek, Montana (12 years, 1994–2016). Annual distributions of weights at length were greater for Arctic Grayling in Red Rock Creek than those in the Kuparuk River. Variation in predicted distributions of weights at length among years was also greater for Red Rock Creek, although fish from both locations exhibited substantial among‐year variation that exceeded half the variation estimated among geographic locations.ConclusionOur estimates indicated that there was substantial spatial and temporal variation in condition of Arctic Grayling that might be investigated further to help understand constraints on populations to aid conservation and management of the species. Our summary approach can be easily extended to other investigations of fish weight at length relationships where multiple quantile estimates are made across many levels of grouping factors.

A predicted distribution for Galois groups of maximal unramified extensions

A predicted distribution for Galois groups of maximal unramified extensions

Bayesian typhoon precipitation prediction with a mixture of ensemble forecast-based and historical event-based prediction functions

Statistical post-processing has been widely adopted in probabilistic forecasting, especially when a predictive distribution of weather variables is of interest. Among the available post-processing methods, a common tool is the ensemble Bayesian model averaging. Challenges arise when it is applied to heavy rain prediction, including the precipitation produced by tropical cyclones. The issues are related to the rare extreme event and limited available data for model training. To address this, we propose in this research a Bayesian mixture model, Bmix, based on information from forecast members and historical tropical cyclones, rather than solely on immediately observed precipitation data. This approach provides a grid-specific Bayesian predictive distribution and an easy-to-derive probabilistic forecast with posterior samples generated from the Markov chain Monte Carlo algorithm. This post-processing procedure can be executed on each grid simultaneously, reducing the computational time significantly. In addition, a categorized Bmix (C.Bmix) is implemented to accommodate different scenarios such as heavy or torrential rain in applications. The proposed approaches are demonstrated on the 24-hour accumulated precipitation forecast of a category 4-equivalent super typhoon Dujuan landing in Taiwan in 2015; while two others, Matmo and Soudelor, are adopted to provide historical information. Over the Taiwan main island, a total of 1282 grids are considered, each with a resolution of 5 km and 20 ensemble member forecasts from the Weather Research and Forecasting Ensemble Prediction System (WEPS) at a 6-hour interval during the typhoon period. The analyses show that the proposed mixture approaches present larger percentages of positive CRPSS values than traditional BMA. For instance, at the time when Dujuan made landfall, the percentages are 92.2 % for Bmix, 95.2 % for C.Bmix, and 15.4 % for BMA when the true precipitation amount exceeds 200 mm/24 h. The proposed methods outperform in the probability prediction and the overall pattern of precipitation.

Coordinate Attention Guided Dual-Teacher Adaptive Knowledge Distillation for image classification

Knowledge distillation (KD) refers to transferring the knowledge learned from a teacher network with complex architecture and strong learning ability to another student network with light-weight and weak learning ability through a specific distillation strategy. However, most existing KD approaches to image classification often employ a single teacher network to guide the training of the student network. When the teacher network makes an erroneous prediction, the transferred knowledge will deteriorate the performance of the student network. To address or mitigate the above issue, we develop a novel KD approach called Coordinate Attention Guided Dual-Teacher Adaptive Knowledge Distillation (CAG-DAKD), to deliver more discriminative and comprehensive knowledge obtained from two teacher networks to a compact student network. Specifically, we integrate the positive prediction distribution of two teacher networks according to whether the two teacher networks predict correctly and the magnitude of the cross-entropy to deliver better output distribution to guide the student network. Furthermore, to distill the most valuable knowledge from the first teacher network that has a similar architecture to the student network, a coordinate attention mechanism is introduced into different layers of the first teacher network so that the student network can effectively learn more discriminative feature representations. We conduct extensive experiments on three standard image classification datasets: CIFAR10, CIFAR100, and ImageNet to verify the superiority of the proposed method over other state-of-the-art competitors. Code will be available at https://github.com/mdt1219/CAG-DAKD.git/

Connecting spaceborne lidar with NFI networks: A method for improved estimation of forest structure and biomass

Spaceborne lidar provides a unique opportunity to supplement the field plot measurements of national forest inventories (NFIs) by providing dense measurements of vertical canopy structure. For full waveform instruments such as the Global Ecosystem Dynamics Investigation (GEDI), measurements take the form of reflected energy as a function of height within an observed footprint. Many forest attributes cannot be directly computed from the waveforms, and thus statistical models relating the measurements to target attributes must be trained using field plot data. Because the discrete footprint samples produced by spaceborne lidar instruments have little chance of collocation with pre-existing field plots, model calibration remains a primary challenge. We leveraged the density and spatial correlation of GEDI observations to make predictions of the cumulative waveform over 326,787 NFI plots distributed across the contiguous United States. The predictions yield probability distributions, giving not only a predicted waveform but quantification of prediction uncertainty. The product of this work is a data set comprised of the NFI plots paired with the predictive waveform distributions. The predicted waveforms and their uncertainties can be used to train models with “error-in-variables” techniques that account for and filter the uncertainty in the predicted waveforms. This data set and the corresponding techniques allow statistically rigorous model training without requirements of collocation, yielding downstream forest attribute predictions that are consistent with NFI measurements and estimates. We demonstrate the data set by training a model for forest above ground biomass density (AGBD) and then use the model to produce a spatially complete 1 km map of AGBD for the contiguous United States. We further compare our AGBD predictions to US NFI estimates at a 64,000 hectare scale, showing the increase in precision to be proportional to the collection of almost 500,000 additional NFI plots across forested regions of CONUS.

Characteristics and phylogenetic distribution of megaplasmids and prediction of a putative chromid in Pseudomonas aeruginosa

Research on megaplasmids that contribute to the spread of antimicrobial resistance (AMR) in Pseudomonas aeruginosa strains has grown in recent years due to the now widely used technologies allowing long-read sequencing. Here, we systematically analyzed distinct and consistent genetic characteristics of megaplasmids found in P. aeruginosa. Our data provide information on their phylogenetic distribution and hypotheses tracing the potential evolutionary paths of megaplasmids. Most of the megaplasmids we found belong to the IncP-2-type, with conserved and syntenic genetic backbones carrying modules of genes associated with chemotaxis apparatus, tellurite resistance and plasmid replication, segregation, and transmission. Extensively variable regions harbor abundant AMR genes, especially those encoding β-lactamases such as VIM-2, IMP-45, and KPC variants, which are high-risk elements in nosocomial infection. IncP-2 megaplasmids act as effective vehicles transmitting AMR genes to diverse regions. One evolutionary model of the origin of megaplasmids claims that chromids can develop from megaplasmids. These chromids have been characterized as an intermediate between a megaplasmid and a chromosome, also containing core genes that can be found on the chromosome but not on the megaplasmid. Using in silico prediction, we identified the “PABCH45 unnamed replicon” as a putative chromid in P. aeruginosa, which shows a much higher similarity and closer phylogenetic relationship to chromosomes than to megaplasmids while also encoding plasmid-like partition genes. We propose that such a chromid could facilitate genome expansion, allowing for more rapid adaptations to novel ecological niches or selective conditions, in comparison to megaplasmids.

Structure-Aware Prototypical Neural Process for Few-Shot Graph Classification.

Graph classification plays an important role in a wide range of applications from biological prediction to social analysis. Traditional graph classification models built on graph kernels are hampered by the challenge of poor generalization as they are heavily dependent on the dedicated design of handcrafted features. Recently, graph neural networks (GNNs) become a new class of tools for analyzing graph data and have achieved promising performance. However, it is necessary to collect a large number of labeled graph data for training an accurate GNN, which is often unaffordable in real-world applications. Therefore, it is an open question to build GNNs under the condition of few-shot learning where only a few labeled graphs are available. In this article, we introduce a new Structure-aware Prototypical Neural Process (SPNP for short) for a few-shot graph classification. Specifically, at the encoding stage, SPNP first employs GNNs to capture graph structure information. Then, SPNP incorporates such structural priors into the latent path and the deterministic path for representing stochastic processes. At the decoding stage, SPNP uses a new prototypical decoder to define a metric space where unseen graphs can be predicted effectively. The proposed decoder, which contains a self-attention mechanism to learn the intraclass dependence between graphs, can enhance the class-level representations, especially for new classes. Furthermore, benefited from such a flexible encoding-decoding architecture, SPNP can directly map the context samples to a predictive distribution without any complicated operations used in previous methods. Extensive experiments demonstrate that SPNP achieves consistent and significant improvements over state-of-the-art methods. Further discussions are provided toward model efficiency and more detailed analysis.

Bayesian extreme value models: Asymptotic behavior, hierarchical convergence, and predictive robustness

The efficacy of Bayesian extreme value models is examined, with a focus on their ability to analyze tail behavior and their predictive accuracy. The hierarchical structure in Bayesian extreme value models helps model heterogeneity by borrowing strength across groups. The theoretical results indicate that the Kullback-Leibler divergence between posteriors drawn from various priors is bounded, reinforcing the stability of Bayesian extreme value models against prior selections. As the sample size grows, the influence of priors diminishes. In addition, the results establish that the hyperposterior distribution converges to a specific distribution as group size increases. This robustness is restricted to prior assumptions and integrates into its hierarchical structure. Another finding confirms the convergence of predictive distributions, especially for big data analyses. The advantage of hierarchical modeling over non-hierarchical counterparts is underscored as the variance component of the predictive loss function diminishes in hierarchical Bayesian extreme value models.

Permeability Prediction and Facies Distribution for Yamama Reservoir in Faihaa Oil Field: Role of Machine Learning and Cluster Analysis Approach

Empirical and statistical methodologies have been established to acquire accurate permeability identification and reservoir characterization, based on the rock type and reservoir performance. The identification of rock facies is usually done by either using core analysis to visually interpret lithofacies or indirectly based on well-log data. The use of well-log data for traditional facies prediction is characterized by uncertainties and can be time-consuming, particularly when working with large datasets. Thus, Machine Learning can be used to predict patterns more efficiently when applied to large data. Taking into account the electrofacies distribution, this work was conducted to predict permeability for the four wells, FH1, FH2, FH3, and FH19 from the Yamama reservoir in the Faihaa Oil Field, southern Iraq. The framework includes: calculating permeability for uncored wells using the classical method and FZI method. Topological mapping of input space into clusters is achieved using the self-organizing map (SOM), as an unsupervised machine-learning technique. By leveraging data obtained from the four wells, the SOM is effectively employed to forecast the count of electrofacies present within the reservoir. According to the findings, the permeability calculated using the classical method that relies exclusively on porosity is not close enough to the actual values because of the heterogeneity of carbonate reservoirs. Using the FZI method, in contrast, displays more real values and offers the best correlation coefficient. Then, the SOM model and cluster analysis reveal the existence of five distinct groups.

Semi-supervised semantic labeling of remote sensing images with improved image-level selection retraining

In recent years, image semantic segmentation technology has developed rapidly, but image annotation usually requires a significant amount of human and financial resources, especially for remote sensing image annotation, which can be expensive and sometimes even unaffordable. To address this issue, this paper integrates the idea of curriculum learning into the self-training method and screens reliable pseudo-labels through computing image-level confidence, significantly reducing the confirmation error problem. Furthermore, the semi-supervised model in this paper combines implicit semantic enhancement with strong data augmentation, which can reduce the coupling between the teacher model and the student model’s prediction distribution and enhance the model’s robustness. Finally, the proposed semi-supervised method is experimentally verified using the ISPRS competition dataset and compared with existing state-of-the-art (SOTA) methods. Experimental results show that the proposed semi-supervised segmentation method achieves higher segmentation accuracy compared to self-training methods. Moreover, despite not using iterative training to simplify the training process, the proposed method still yields satisfactory segmentation results.

Case weighted power priors for hybrid control analyses with time-to-event data.

We develop a method for hybrid analyses that uses external controls to augment internal control arms in randomized controlled trials (RCTs) where the degree of borrowing is determined based on similarity between RCT and external control patients to account for systematic differences (e.g., unmeasured confounders). The method represents a novel extension of the power prior where discounting weights are computed separately for each external control based on compatibility with the randomized control data. The discounting weights are determined using the predictive distribution for the external controls derived via the posterior distribution for time-to-event parameters estimated from the RCT. This method is applied using a proportional hazards regression model with piecewise constant baseline hazard. A simulation study and a real-data example are presented based on a completed trial in non-small cell lung cancer. It is shown that the case weighted power prior provides robust inference under various forms of incompatibility between the external controls and RCT population.

Calibrated EMOS: applications to temperature and wind speed forecasting

Ensembles of meteorological quantities obtained from numerical models can be used for forecasting weather variables. Unfortunately, such ensembles are often biased and under-dispersed and therefore need to be post-processed. Ensemble model output statistics (EMOS) is a widely used post-processing technique to reduce bias and dispersion errors of numerical ensembles. In the EMOS approach, a full probabilistic prediction is given in the form of a predictive distribution with parameters depending on the ensemble forecast members. Parameters are then estimated and substituted, thus obtaining a so-called estimative predictive distribution. Nonetheless, estimative distributions may perform poorly in terms of the coverage probability of the corresponding quantiles. This work proposes the use of predictive distributions based on a bootstrap adjustment of estimative predictive distributions, in the context of EMOS models. These distributions are calibrated, which means that the corresponding quantiles provide exact coverage probabilities, in contrast to the estimative distributions. The introduction of the bootstrap calibrated procedure for EMOS is the innovative aspect of this study. The performance of the suggested calibrated EMOS is evaluated in two simulation studies, comparing the different predictive distributions by means of the log-score, the continuous ranked probability score, and the coverage of the corresponding predictive quantiles. The results of these simulation studies show that the proposed calibrated predictive distributions improve estimative solutions, both reducing the mean scores and producing quantiles with exact coverage levels. The good performance of the new calibrated EMOS is further stressed in two real data applications, one about maximum daily temperatures at sites located in the Veneto region (Italy) and the other one about wind speed forecasts at weather stations over Germany.

Noisy Correspondence Learning with Self-Reinforcing Errors Mitigation

Cross-modal retrieval relies on well-matched large-scale datasets that are laborious in practice. Recently, to alleviate expensive data collection, co-occurring pairs from the Internet are automatically harvested for training. However, it inevitably includes mismatched pairs, i.e., noisy correspondences, undermining supervision reliability and degrading performance. Current methods leverage deep neural networks' memorization effect to address noisy correspondences, which overconfidently focus on similarity-guided training with hard negatives and suffer from self-reinforcing errors. In light of above, we introduce a novel noisy correspondence learning framework, namely Self-Reinforcing Errors Mitigation (SREM). Specifically, by viewing sample matching as classification tasks within the batch, we generate classification logits for the given sample. Instead of a single similarity score, we refine sample filtration through energy uncertainty and estimate model's sensitivity of selected clean samples using swapped classification entropy, in view of the overall prediction distribution. Additionally, we propose cross-modal biased complementary learning to leverage negative matches overlooked in hard-negative training, further improving model optimization stability and curbing self-reinforcing errors. Extensive experiments on challenging benchmarks affirm the efficacy and efficiency of SREM.

HACDR-Net: Heterogeneous-Aware Convolutional Network for Diabetic Retinopathy Multi-Lesion Segmentation

Diabetic Retinopathy (DR), the leading cause of blindness in diabetic patients, is diagnosed by the condition of retinal multiple lesions. As a difficult task in medical image segmentation, DR multi-lesion segmentation faces the main concerns as follows. On the one hand, retinal lesions vary in location, shape, and size. On the other hand, because some lesions occupy only a very small part of the entire fundus image, the high proportion of background leads to difficulties in lesion segmentation. To solve the above problems, we propose a heterogeneous-aware convolutional network (HACDR-Net) that composes heterogeneous cross-convolution, heterogeneous modulated deformable convolution, and optional near-far-aware convolution. Our network introduces an adaptive aggregation module to summarize the heterogeneous feature maps and get diverse lesion areas in the heterogeneous receptive field along the channels and space. In addition, to solve the problem of the highly imbalanced proportion of focal areas, we design a new medical image segmentation loss function, Noise Adjusted Loss (NALoss). NALoss balances the predictive feature distribution of background and lesion by jointing Gaussian noise and hard example mining, thus enhancing awareness of lesions. We conduct the experiments on the public datasets IDRiD and DDR, and the experimental results show that the proposed method achieves better performance than other state-of-the-art methods. The code is open-sourced on github.com/xqh180110910537/HACDR-Net.

AI-KD: Adversarial learning and Implicit regularization for self-Knowledge Distillation

We present a novel adversarial penalized self-knowledge distillation method, named adversarial learning and implicit regularization for self-knowledge distillation (AI-KD), which regularizes the training procedure by adversarial learning and implicit distillations. Our model not only distills the deterministic and progressive knowledge which are from the pre-trained and previous epoch predictive probabilities but also transfers the knowledge of the deterministic predictive distributions using adversarial learning. The motivation is that the self-knowledge distillation methods regularize the predictive probabilities with soft targets, but the exact distributions may be hard to predict. Our proposed method deploys a discriminator to distinguish the distributions between the pre-trained and student models while the student model is trained to fool the discriminator in the trained procedure. Thus, the student model not only can learn the pre-trained model’s predictive probabilities but also align the distributions between the pre-trained and student models. We demonstrate the effectiveness of the proposed method with network architectures on multiple datasets and show the proposed method achieves better performance than existing approaches.

Abstract 4797: Impact of COVID-19 pandemic on breast cancer screening and diagnosis stage in a large midwestern United States academic medical center

Abstract Background: Access to breast cancer screening mammogram services decreased in association with the COVID-19 pandemic. There is also evidence for increases in late-stage breast cancer diagnoses. Our objectives were to determine: 1) the COVID-19-affected period on mammogram screening, 2) the proportion of pandemic-associated missed or delayed mammogram screening visits overall and by race/ethnicity and age group, and 3) evidence for pandemic-associated shifts in diagnosis stage. Methods: Screening mammogram encounter data between 1-1-2019 and 12-31-2022 were extracted from EPIC for females ≥ 40 years old for the screening analysis. We used Bayesian state space models to describe weekly screening mammogram counts, modeling an interruption that phased in and out from 3-1-2020 to 9-1-2020. We used the posterior predictive distribution to simulate differences between a predicted, uninterrupted process and the observed screening mammogram counts. Breast cancer diagnoses at ≥ 21 years from the tumor registry between 12-1-2018 and 11-30-2021 were included in the stage analysis. We used logistic regression models to estimate late-stage diagnosis odds comparing matched three-month periods during the pandemic to before the pandemic. Results: A total of 319,492 encounters among 146,644 women were included. Model-estimated screening mammograms dropped by 98.8% (95% CI 95.1 to 100) between 3-15-2020 and 5-24-2020, returning to pre-pandemic levels or higher after this period. Drops in screening mammogram encounters did not vary significantly by race/ethnicity or age group (p > .75). Among 4,669 breast cancer diagnoses, we found no significant differences in the odds of late-stage diagnoses for any period vs. the same period before the pandemic (Table). Conclusions: These data suggest a short-term pandemic effect on screening mammograms. Evidence for increases in late-stage diagnoses is limited. These results may inform future pandemic planning. OR (95% CI) Period comparison 1.19 (0.74 to 1.91) 3/2020 to 5/2020 vs. 3/2019 to 5/2019 1.05 (0.69 to 1.60) 6/2020 to 8/2020 vs. 6/2019 to 8/2019 1.14 (0.73 to 1.78) 9/2020 to 11/2020 vs. 9/2019 to 11/2019 0.92 (0.60 to 1.41) 12/2020 to 2/2021 vs. 12/2019 to 2/2020 1.08 (0.71 to 1.65) 3/2021 to 5/2021 vs. 3/2019 to 5/2019 0.89 (0.59 to 1.33) 6/2021 to 8/2021 vs. 6/2019 to 8/2019 1.31 (0.84 to 2.04) 9/2021 to 11/2021 vs. 9/2019 to 11/2019 Citation Format: Kimberly J. Johnson, RJ Waken, Caitlin P. O'Connell, Derek Brown. Impact of COVID-19 pandemic on breast cancer screening and diagnosis stage in a large midwestern United States academic medical center [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2024; Part 1 (Regular Abstracts); 2024 Apr 5-10; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2024;84(6_Suppl):Abstract nr 4797.

Human judgement forecasting of COVID-19 in the UK.

In the past, two studies found ensembles of human judgement forecasts of COVID-19 to show predictive performance comparable to ensembles of computational models, at least when predicting case incidences. We present a follow-up to a study conducted in Germany and Poland and investigate a novel joint approach to combine human judgement and epidemiological modelling. From May 24th to August 16th 2021, we elicited weekly one to four week ahead forecasts of cases and deaths from COVID-19 in the UK from a crowd of human forecasters. A median ensemble of all forecasts was submitted to the European Forecast Hub. Participants could use two distinct interfaces: in one, forecasters submitted a predictive distribution directly, in the other forecasters instead submitted a forecast of the effective reproduction number R t. This was then used to forecast cases and deaths using simulation methods from the EpiNow2 R package. Forecasts were scored using the weighted interval score on the original forecasts, as well as after applying the natural logarithm to both forecasts and observations. The ensemble of human forecasters overall performed comparably to the official European Forecast Hub ensemble on both cases and deaths, although results were sensitive to changes in details of the evaluation. R t forecasts performed comparably to direct forecasts on cases, but worse on deaths. Self-identified "experts" tended to be better calibrated than "non-experts" for cases, but not for deaths. Human judgement forecasts and computational models can produce forecasts of similar quality for infectious disease such as COVID-19. The results of forecast evaluations can change depending on what metrics are chosen and judgement on what does or doesn't constitute a "good" forecast is dependent on the forecast consumer. Combinations of human and computational forecasts hold potential but present real-world challenges that need to be solved.