text_with_holes stringlengths 113 4.67k | text_candidates stringlengths 40 1.73k | A stringclasses 6
values | B stringclasses 6
values | C stringclasses 6
values | D stringclasses 6
values | label stringclasses 4
values |
|---|---|---|---|---|---|---|
Another recent direction of relevance, grouped together under the term spectral deconfounding, includes work by Ćevid et al. <|MaskedSetence|> (2022); Fan et al. (2024); Wang and Shah (2025); Sun et al. <|MaskedSetence|> For example, Ouyang et al. <|MaskedSetence|> In contrast, we show that with multiple response ... | **A**: (2024) among others, and considers estimation and inference in high-dimensional regression models with unmeasured confounders.
**B**: (2023) focused on inference in high-dimensional GLMs with unmeasured confounders by generalizing the decorrelated score approach from Ning and Liu (2017) to account for the effec... | CAB | CAB | CAB | ACB | Selection 3 |
<|MaskedSetence|> We focus on the setting that incorporates the transaction costs. <|MaskedSetence|> <|MaskedSetence|> Consistent with the results in Figure 3, the majority of portfolio gains are realized from the long leg rather than the short leg. Actually, the returns gained from the short legs are negative, and ... | **A**:
Table 1 reports the detailed performance of the combined long-short portfolio, as well as its long and short components.
**B**: It shows that scores estimated from augmented features yield higher annualized percentage returns (APR) and Sharpe ratios compared to the baseline.
**C**: Furthermore, factors ext... | ABC | ABC | ABC | CAB | Selection 3 |
1.
Sample-size determination under delayed effects. To our knowledge, we provide the first closed-form algorithm for calibrating interim and final sample sizes in a two-stage Bayesian design under non-proportional hazards. <|MaskedSetence|> <|MaskedSetence|> (2020)) left sample size to ad-hoc rules or relied on pr... | **A**: By optimising a weighted expected-sample-size criterion, DTE–BOP2 achieves near-minimal patient exposure under the null while maintaining high power under the alternative, and—importantly—retains these guarantees when the same boundary is embedded in designs with three or more looks..
**B**: Zhou et al.
**C**:... | CBA | CBA | CBA | BCA | Selection 1 |
Here, we consider methods for componentwise inference for variance components in the presence of nuisance parameters. The methods are based on universal inference, and in particular split likelihood ratio tests (Wasserman et al.,, 2020). <|MaskedSetence|> Thus, even when there are no nuisance parameters, the methods ... | **A**: For example, in a setting with crossed random effects, which are known to complicate both computation and theory, we decrease the required time by several orders of magnitude compared to a naive implementation..
**B**: Consequently, the proposed tests and confidence intervals are uniformly valid in finite sampl... | BCA | BCA | BCA | CBA | Selection 2 |
<|MaskedSetence|> <|MaskedSetence|> We first review the original, probit, and monotone BART models from the literature. <|MaskedSetence|> In Section 4 we demonstrate the proposed method with a simulation study. Section 5 concludes.. | **A**:
In this paper, we introduce an implementation of monotone BART for binary outcomes.
**B**: In Section 3 we propose probit monotone BART, including the model set up and some details on the code implementation.
**C**: This is done with a probit link, implemented using the ideas of normal latent variables/data a... | BCA | ACB | ACB | ACB | Selection 4 |
<|MaskedSetence|> Based on posterior samples from the LFCM, we estimate the jump length distribution, the activity type distribution, the mean and variance of a Brownian motion describing the activities’ movement. Areas of frequent return are a function of these quantities.
Using one week of data from the individual... | **A**: Figure 18 displays the trajectory of the device in 500 points simulated using the MAP estimates, where Brownian motion movement is shown in red, Lévy jumps are shown in gray, and the identified activity regions are shaded in blue.
**B**: This simulation can be compared with Figure 17, which is the true mobility... | CAB | ABC | CAB | CAB | Selection 4 |
<|MaskedSetence|> <|MaskedSetence|> (2015); Shi et al. (2021)). <|MaskedSetence|> (2023), which has form of
. | **A**: (2011); Han et al.
**B**: Another more efficient measure in this case to study the estimated accuracy when unclassified set remain is the adjusted MCC(MCCa)MCC(MCCa) defined by Shi et al.
**C**:
To considered the effects of un-classification in Table 1, the success
rates for the detection of existing l... | CAB | CAB | CAB | CAB | Selection 3 |
<|MaskedSetence|> This was probably not realized in [6] since it was not stated in the description/manual of [10]. On the other hand, [6] gives an interpretation of the fixed point iteration for ψα\psi_{\alpha} as a Newton-Raphson method under additional assumptions.
To the best of our knowledge, there is no general... | **A**: 2.2, Cor.
**B**:
(a version of the fixed point iteration for ψα\psi_{\alpha}) is also used by the codes in [10] to compute the empirical expectiles.
**C**: 2.3] provides quadratic convergence as expected from a Newton-type method..
| CAB | BAC | BAC | BAC | Selection 2 |
<|MaskedSetence|> In section 3 it is shown that nm(n+m)Tnm\sqrt{nm(n+m)}T_{nm} converges in distribution to a random variable ZZ, which follows the Maxwell-Boltzmann law. Moreover, we report on the results of a small Monte Carlo simulation study. <|MaskedSetence|> In the last section, the basic idea is carried ov... | **A**: Here, the VV-test proves to be superior to the Smirnov-test (SS-test).
**B**: By Remark 1 the event {Rnm=n+m}\{R_{nm}=n+m\} has positive probability m(n+m)(n+m−1)\frac{m}{(n+m)(n+m-1)} and therefore cannot be ignored.
The paper is structured as follows: In section 2 we motivate the upper TnmT_{nm}-test (c... | BAC | BAC | BAC | ABC | Selection 2 |
<|MaskedSetence|> Informally, GRDPG assigns each vertex a low-dimensional vector called the latent position, and the connection probability between any pair of vertices is given by the generalized inner product of the associated latent positions. We defer the formal definition to Section 2.1. <|MaskedSetence|> GRDPG ... | **A**: GRDPG has been attracting attention because it not only has a simple low-rank structure but also is versatile as it encompasses several popular network models, such as stochastic block models (Holland et al.,, 1983), degree-corrected stochastic block models (Karrer and Newman,, 2011), mixed membership stochastic... | BAC | CAB | CAB | CAB | Selection 3 |
The response of interest is the average systolic blood pressure, averaging over the four available measurements. <|MaskedSetence|> <|MaskedSetence|> Here it is reasonable to implement the exponential density weight, as the majority of the physical intensity values are small. <|MaskedSetence|> By cross-validation we... | **A**: The potential for large values of physical activity intensity ss, which serves as argument of the XiX_{i}, means that the domain has no clear upper bound, motivating to consider an infinite domain (0,∞)(0,\infty) for the functional predictor.
**B**: We investigate the performance of the proposed wFLM with a wei... | ABC | ABC | ABC | CBA | Selection 2 |
Another approach is to fit a separate generative model to real data to generate counterfactuals to serve as ground truth. This aims to create more realistic DGPs than purely synthetic setups, while maintaining experimental control by making the generative model adhere to specified requirements. Like CI estimators used... | **A**: Other model-driven evaluation research focuses more on matching user-defined functions than on realism, and on the ATE rather than the CATE.
**B**: An interesting variant is RealCause [14], which enables this experimental control due to its easy manipulation of the DGP with ‘knobs’ for confounding, effect heter... | BAC | BAC | CAB | BAC | Selection 1 |
<|MaskedSetence|> Specifically, our approach integrates causal relationships into SHAP calculations by employing the Peter-Clark (PC) algorithm [6] for causal edge discovery and the Intervention Calculus when the DAG is Absent (IDA) algorithm [7] for causal strength quantification. The PC algorithm is a constraint-bas... | **A**: Extensive experiments were conducted to highlight the advantages of our method.
.
**B**: In this paper, we propose a novel Causal SHAP framework that explains how each feature in the dataset contributes to the model’s prediction while respecting the causal relationships within the data.
**C**: IDA builds upon ... | BCA | BCA | BCA | ACB | Selection 3 |
<|MaskedSetence|> However, they offer no guidance on how to identify such a model from data, nor do they specify its size or neuronal structure.
Guarantees based on constructive approximation, especially [113, 88, 16, 50, 101, 96], improve upon this by providing estimates for the number of neurons and their arrangem... | **A**: Classical existence theorems, i.e., the classical UAT [54, 28, 60], guarantee the existence of a sufficiently large neural network that can approximate any given target function to arbitrary accuracy.
**B**: Furthermore, although these results do provide a (semi-)explicit architectural construction, their archi... | ABC | ABC | ABC | CAB | Selection 2 |
We begin by outlining the architecture of our hybrid approach and its constituent algorithms. <|MaskedSetence|> <|MaskedSetence|> Special attention is given to how the hybrid method produces distinct clusters in the visualization. We also discuss how a bipartite document–topic graph is leveraged to enhance cluster co... | **A**: For interpretability and human-in-the-loop validation, we include analysis of a topic distribution bar chart and a bipartite network diagram, illustrating how domain experts can engage with the results.
**B**: We then provide a rigorous exposition of UMAP’s statistical underpinnings, focusing on its objective f... | BCA | BCA | BCA | ABC | Selection 3 |
This mirrors expression (II.1), but with one crucial difference: in order for SS to be an e-variable, the denominator P0w0P_{0}^{w_{0}} cannot be chosen freely, as it must be the prior w0∗w_{0}^{*} that ensures that SS qualifies as a GRO e-variable (i.e., satisfies the e-variable condition).
This formulation, however... | **A**: In particular, smaller values of −logS(𝐱)-\log S(\mathbf{x}) correspond to larger e-values and hence stronger evidence against the null model ℳ0\mathcal{M}_{0}.
This observation addresses a longstanding issue in MDL: although it provides a principled model comparison method, it lacks explicit statistical guar... | ACB | ACB | ACB | BAC | Selection 2 |
<|MaskedSetence|> <|MaskedSetence|> Under noise-free conditions (i.e., noise std=0), both AUC-ROC and VUS-ROC maintain their ranking capabilities without degradation. However, when noise is present, AUC-ROC exhibits ranking errors at low false positive rates, while VUS-ROC shows ranking errors at high false positive ... | **A**:
A-D Impact of Noise on Different Metrics
Figs.
**B**: This phenomenon occurs because Aff-F1 is based on interval membership, where increased false positives may be interpreted as valid warnings, leading to inflated Aff-F1 scores and consequently introducing experimental bias..
**C**: (S11a-S11d) demonstrate... | ACB | ACB | BCA | ACB | Selection 1 |
Tables 2 and 2 present the accuracy and model consistency, respectively, of all methods on the linear synthetic dataset. Our method, NP2M2, achieves the highest accuracy in 4 out of 6 settings and ranks the second in the remaining two, as shown in Tables 2. <|MaskedSetence|> <|MaskedSetence|> Conversely, when 𝒟(θ)... | **A**: Our method consistently ranks first across all settings in terms of both accuracy and model consistency, demonstrating its superior performance on the complex nonlinear dataset.
**B**: In the two settings where RRM NN achieves the highest accuracy and NP2M2 ranks second, the data distribution map 𝒟(θ)\mathcal... | CBA | CBA | ABC | CBA | Selection 4 |
To identify the optimal multi-task training configuration, we evaluate how much of a performance drop we incur at each layer compared to the single-task setting in GNNs. <|MaskedSetence|> Our findings indicate that ADMP-GNN ST outperforms ADMP-GNN ALM. <|MaskedSetence|> Furthermore, ADMP-GNN ST exhibits a smaller st... | **A**: Notably, the performance of ADMP-GNN ST is comparable to, or even exceeds, that of GNN when trained under the single-task setting.
**B**: The comparative analysis of the three strategies for both GCN is detailed in Tables 1 and 8.
**C**: The next step, outlined in Section 3.5, is to learn a policy that selects... | CAB | BAC | BAC | BAC | Selection 4 |
<|MaskedSetence|> It comprises comprehensive point-by-point records from the 2023 Wimbledon Championships Men’s Singles tournament, covering 31 matches after the first two rounds. This dataset includes the final between Carlos Alcaraz and Novak Djokovic (identified as match-id: 2023-wimbledon-1701). The complete data ... | **A**: The scoring rule in a game is as follows.
**B**:
3 Empirical analysis
The dataset employed in this study is sourced from the 2024 Mathematical Contest in Modeling (MCM) (Problem C: Momentum in Tennis).
**C**: If both players reach 3 points each, the score is called Deuce.
| BAC | BAC | BAC | BAC | Selection 1 |
In Figure.1, we provide the recovery performance as a function of the sparsity level KK. A higher level of critical sparsity indicates improved empirical reconstruction performance. The simulation results reveal that the critical sparsity of JS-gOMP algorithms is much larger compared to OMP and gOMP algorithms. As the ... | **A**: As more unnecessary components start to appear in the output making the signal less sparse, the relative contribution of the actual atoms in the signal tends to reduce in both cases (OMP, gOMP), whereas the JS-gOMP gives considerably better results in reducing the noise (as per Figure.
**B**: 2).
Figure 2: co... | CAB | CAB | CBA | CAB | Selection 4 |
<|MaskedSetence|> <|MaskedSetence|> For example, Cohn et al. (2022) demonstrate that log-linear models do not always produce estimates with economically meaningful interpretations, and can even result in the wrong sign in expectation. <|MaskedSetence|> More importantly, as discussed in previous works, log-linear reg... | **A**: However, previous works have raised some important drawbacks of the log-linear regression approach.
**B**:
Prominent applications of log-linear regression include the modeling of trade data and panel data with non-negative outcomes, such as earnings (Card, 2001), and count and count-like data in finance, suc... | BAC | BAC | CBA | BAC | Selection 4 |
Cameroon faces recurrent food insecurity driven by chronic poverty, conflict-induced displacement, volatile markets, and climate shocks (World Food Programme, 2024b; d’Errico et al., 2023). <|MaskedSetence|> <|MaskedSetence|> (2024), have underscored the complexity of these dynamics and the value of spatio-temporal p... | **A**: These challenges show substantial spatial and temporal variability, as different regions experience varying levels and timings of food access constraints.
**B**: Exploratory and descriptive analyses, such as those by Ayalew et al.
**C**: As a result, such models may struggle to reflect local deviations or abru... | ABC | ABC | CBA | ABC | Selection 4 |
We define Bias 1 as the difference in entropy between marginal distributions and show how it can be controlled, in the bivariate categorical setup, by parameterized deviations from a Bayesian Dirichlet equivalence (BDe) prior. Likewise, we define Bias 2 as the difference of Kullback-Leibler Divergences between distribu... | **A**: We choose the common bivariate causal discovery problem with categorical data; a simple setting that serves as a clear and illustrative example.
**B**: Specifically, our contributions are:
.
**C**: To the best of our knowledge, these biases – in particular for the setup of continuing interventions – have not b... | CAB | CAB | BAC | CAB | Selection 2 |
<|MaskedSetence|> <|MaskedSetence|> While this design demonstrates improved predictive accuracy across a range of PDE benchmarks, it typically requires two to three times longer training time. The original “vanilla” DeepONet, on the other hand, remains attractive for its computational simplicity. <|MaskedSetence|> M... | **A**: In response to this architectural limitation, various extensions have been proposed to improve representational capacity, including SVD-DeepONet [31], the R-adaptive DeepONet designed for discontinuities [41], multi-input EDeepONet [29], and the Separable Operator Networks [40].
**B**: These models reflect an a... | BAC | ABC | ABC | ABC | Selection 3 |
We observe a different compression phenomenon with the high-variance initialization, when the samples are initially distributed around the modes instead of between the modes. Figure 5 demonstrates that in this case, the preconditioning has a significant acceleration effect, where the KL divergence decreases faster tha... | **A**: .
**B**: In this case, the preconditioning accelerates the convergence.
**C**: Moreover, Figure 6 demonstrates that PBRWP particles are less noisy than BRWP particles at low iterations.
Figure 5: Evolution of the KL divergence between baselines and BRWP-based methods for the bimodal distribution, initialized... | CBA | BAC | CBA | CBA | Selection 4 |
In the current application, gNg_{N} is an estimated regression function that indexes the target parameter. <|MaskedSetence|> <|MaskedSetence|> In the non-parametric literature, cross-fitting is a popular approach to avoid Donsker conditions [[, see, e.g.,]]chernozhukov2018double. <|MaskedSetence|> We, therefore, d... | **A**: This limits the complexity of the allowed surrogate index estimators for valid inference, even though many machine learning methods can still be used.
**B**: We will impose Donsker conditions on this estimator.
**C**: Methods for inference about data-adaptive target parameters that use cross-fitting exist (see... | BAC | BAC | BAC | BAC | Selection 4 |
<|MaskedSetence|> <|MaskedSetence|> Given these properties, we would like confidence intervals that are as narrow as possible, and return many true positives.
Data-Generating Process. In both simulations, we simulate 250 datasets according to data-generating processes described in detail in Section˜F.2 and illustra... | **A**: Coverage should be above the nominal level (dashed line in first column), and the proportion of false positives should be below 0.050.05.
**B**: In the latter case, we anticipate wider confidence intervals, reflecting the inherent difficulty of the task.
**C**:
Figure 2: From left to right, coverage average c... | CAB | CAB | BAC | CAB | Selection 2 |
The definition of “recovery” is not the same. The necessity statement (i) is about the impossibility of exact recovery (i.e. <|MaskedSetence|> vanishing rescaled error) that the sufficiency statement (ii) ensures. <|MaskedSetence|> To make the phase transition clean, one must show that the threshold is sufficient for... | **A**: In particular, it is possible in theory that both statements hold simultaneously (possibility of approximate recovery but impossibility of exact recovery).
**B**: We believe the same holds in our sparse setting and leave the proof for future work.
.
**C**: equality of supports), which is stronger than appro... | CAB | CAB | CAB | BAC | Selection 1 |
Here we are also interested in the connection between network properties and disease dynamics, but we shift our focus to the role played by data available on the spreading of the disease. <|MaskedSetence|> These epidemic networks are not given or taken as an assumption of social-interactions, as in the previous approa... | **A**: This approach is in line with the broader tendency to employ inferential approaches in network science [3, 4, 5, 6, 7] .
The networks we investigate here are transmission trees, with nodes representing infected individuals and directed links representing who infected who.
**B**: Instead, they are inferred fro... | ABC | ABC | ABC | CAB | Selection 1 |
<|MaskedSetence|> The remainder of the paper is organized as follows. Section 2 reviews related work. In Section 3, we introduce the necessary preliminaries, including notation, basic assumptions, and a formal statement of the problem. <|MaskedSetence|> In Section 5, we propose our algorithm for learning the structur... | **A**: Finally, Section 7 concludes the paper and outlines directions for future research.
.
**B**:
The main contributions of this paper are summarized in Table 1.
**C**: Section 4 presents worst-case lower bounds on the maximum experiment size and the total number of experiments required to identify directed edge... | BCA | BCA | BCA | ABC | Selection 3 |
We analyze data from a retrospective cohort of liver transplant candidates at Johns Hopkins Hospital. The target population consists of patients referred for liver transplantation from 1/1/2016-12/31/2017. <|MaskedSetence|> Baseline SDOH variables include age, sex, race/ethnicity, and neighborhood ADI. <|MaskedSeten... | **A**: Other social determinants such as educational attainment and native language were considered but ultimately not included in the analysis, since either reliable information was not recorded in available data or there was insufficient variation across these variables in the analysis.
**B**: We use insurance statu... | CAB | CAB | CBA | CAB | Selection 4 |
5. <|MaskedSetence|> In an ordinal GLM with the complementary log-log link function of Eq. <|MaskedSetence|> <|MaskedSetence|> All cumulative models have the property of “collapsibility” (Tutz,, 2012), which is a preferred property of ordinal models that holds if the interpretation of a parameter is unchanged after ... | **A**: Proportional Hazards Equivalence and Collapsibility
For certain model structures there is an equivalence between the cumulative and sequential model.
**B**: These latter models are examples of cumulative models.
**C**: (13) and covariates with global effects, the sequential model is equivalent to the popular ... | ACB | ACB | ACB | ACB | Selection 2 |
RSSampling provides sampling functions for both classical RSS and several modified RSS variants, such as Median RSS (MRSS), Percentile RSS (PRSS), Extreme RSS (ERSS), and Double RSS (DRSS). <|MaskedSetence|> NSM3 includes only the classical RSS procedure as a sampling tool, along with critical value calculations for a... | **A**: RSStest primarily focuses on mean testing for RSS and MRSS and generating RSS data under a normal distribution.
**B**: Unlike other packages, it provides methods to calculate efficient sample allocations for URSS, improving estimation efficiency for both continuous and binary data.
.
**C**: Additionally, it in... | CAB | BAC | CAB | CAB | Selection 4 |
Figure 5 illustrates the prevalence of identity slippage and distribution of claim types across sections and disciplines. We found identity slippage to be widespread across disciplines. Within economics, identity slippage occurred at least once in 60 studies (68%). <|MaskedSetence|> However, the nature of interpretat... | **A**: A more detailed cross-discipline analysis is provided in the Appendix.
.
**B**: For this reason, we refrain from comparing the overall severity of identity slippage across disciplines solely on the basis of the presented numbers.
**C**: Quantitatively, prevalence was comparable in political science science (... | CBA | CBA | CAB | CBA | Selection 1 |
Approaches targeting ROC-related metrics can be broadly categorized into two main types. <|MaskedSetence|> For example, Sun et al. (2017) proposed maximizing AUC through feature complementarity, while Hsu et al. (2014) optimized pAUC using stepwise selection. Although practical and easy to implement, these methods oft... | **A**: However, these methods often rely on restrictive assumptions, such as multivariate normality or single-index models, which may not hold in real-world applications.
**B**: The first category comprises empirical performance-based methods, which typically follow a two-step framework: biomarkers are first selected ... | BCA | CAB | BCA | BCA | Selection 4 |
<|MaskedSetence|> <|MaskedSetence|> (2017)] or to search for latent clusters in [Li
et al. <|MaskedSetence|> Rare examples of pattern discovery on categorical data are cited in the overview [Subramanian et al. (2020)]
for the integration of genomics or metabolomics data.
It is therefore natural to assume that effect... | **A**: In current machine learning, pattern discovery or association rule learning represents an important task [Golden (2020)].
**B**: So far, individual applications exploiting regularized approaches considered mainly continuous data.
For example, the learned patterns were used to construct classification rules in [... | CAB | ABC | ABC | ABC | Selection 2 |
We then show that if the algebraic boundary of a positive geometry defines a hyperbolic hypersurface, then the positive geometry admits a so called dual volume representation. We call such positive geometries hyperbolic. Hence, every completely monotone positive geometry is hyperbolic. For a positive geometry PP, being... | **A**: From these known results, we deduce the following in the context of positive geometries.
.
**B**: These are the polynomials admitting a symmetric determinantal representation [32, Section 4].
**C**: In summary, the fundamental solution EE to a PDE associated to a homogeneous polynomial pp, yields its Riesz mea... | CBA | CBA | CBA | CBA | Selection 1 |
Wav2Vec2 [16] and HuBERT [17] are both pre-trained using self-supervised learning. Wav2Vec2 relies on a contrastive task trained on 53k hours of Librivox data, while HuBERT employs a predictive task trained on 60k hours of Libri-light data. <|MaskedSetence|> <|MaskedSetence|> <|MaskedSetence|> Its pre-training lever... | **A**: Its training data is significantly larger, albeit potentially noisier, with over 5 million weakly labeled hours.
**B**: Comparing their results will allow us to evaluate the capacity of self-supervised speech representations to extract meaningful information from infant cries.
Whisper [18], on the other hand, i... | BAC | BAC | BCA | BAC | Selection 2 |
Organization of this survey
In Section 2, we consider expected calibration error (ECE) and explore its weaknesses. In Section 3, we introduce weighted calibration measures which capture the notion of indistinguishability to limited classes of distinguishers. <|MaskedSetence|> In Section 4, we describe Calibration d... | **A**: In Section 6, we define the distance to calibration, which proposes a ground-truth notion of what approximate calibration ought to mean, and show how smooth calibration shows up naturally in this setting..
**B**: This unifies several different notions of approximate calibration in the literature.
**C**: We rev... | BCA | BCA | BCA | CAB | Selection 2 |
<|MaskedSetence|> We now provide additional details regarding the synthetic data generation and hyperparameter choices when running the different algorithms.
The synthetic datasets are generated to mimic some characteristics observed in MALDI-MSI data. Specifically, we consider a grid of 20×2020\times 20 pixels, wit... | **A**: In Section 4.1 of the manuscript, we present a simulation study to analyze the biclustering recovery performance of Pose, compared to other biclustering methods.
**B**: The remaining rows are classified as “noise”.
**C**: To generate the m/z intensity level and create a biclustering structure, for each column ... | CBA | ACB | ACB | ACB | Selection 3 |
Root Cause Analysis (RCA) range from manual, diagram-based inspections to advanced automated systems leveraging statistical or causal inference techniques [13]. These approaches are broadly employed in domains like epidemiology, medical diagnostics and, increasingly, system observability, particularly for microservice... | **A**: Several scoring metrics have been proposed, including Shapley-style attributions derived from soft interventions, as implemented in libraries such as DoWhy.
**B**: The complexity and opacity of modern systems, due to hidden components or limited monitoring, pose significant challenges for traditional RCA method... | CAB | BCA | BCA | BCA | Selection 3 |
<|MaskedSetence|> The concept of Latin square comes from combinatorial mathematics — an N×NN\times N square with NN different symbols appearing only once per column and row. <|MaskedSetence|> LHS distributes points to tile the space while preserving the one-dimensional projection property—i.e., ensuring that each mar... | **A**: A Latin hypercube generalizes this property to a PP-dimensional hypercube, where each dimension is binned into NN disjoint intervals [i/N,(i+1)/N)[i/N,(i+1)/N) where i=0,1,…,N−1i=0,1,...,N-1 with marginal probability 1/N1/N.
**B**: This implies that the number of targeted samples must be known a priori, and onc... | CAB | CAB | CAB | BCA | Selection 1 |
<|MaskedSetence|> <|MaskedSetence|> One approach is to fix the monomials corresponding to non‐interaction terms and perform random sampling on the others. Another possibility is to randomly select parameters rather than monomials to construct a smaller ZVCV design matrix, analogous to the apriori‐ZVCV method in South... | **A**:
Our proposed methods do not scale well computationally as dd increases.
**B**: We do not investigate these alternatives here.
.
**C**: Various more computationally efficient selection operators exist, albeit at the expense of statistical efficiency.
| ACB | ACB | ACB | ACB | Selection 2 |
Beyond telematics and driver-specific information, external contextual variables such as weather conditions have also been increasingly incorporated into risk assessments. Mornet et al. [24] constructed an economic index for insurance risk management based on historical wind speed records in France. Similarly, Gao and ... | **A**: [26] developed claim frequency models by integrating telematics and detailed weather data into frameworks.
Most studies take a feature-based approach, extracting driving behavior features from basic signals.
**B**: Then, they used negative binomial (NB) regression to model the number of near-miss events [10].... | ABC | ABC | BAC | ABC | Selection 1 |
<|MaskedSetence|> To quantify the agreement, we interpolated the posterior band onto the EM grid and computed the fraction of EM step heights lying within the band. <|MaskedSetence|> This echoes the simulation findings: the EM estimator provides a faithful nonparametric baseline, whereas the Bayesian Weibull AFT yiel... | **A**: The pointwise coverage was 0.778, indicating that the Bayesian posterior adequately encompassed the nonparametric shape while smoothing the jagged step function.
**B**:
Figure 7 overlays the EM (Turnbull) step function with the Bayesian posterior median survival curve and its 95% credible band.
**C**: See App... | BAC | CBA | BAC | BAC | Selection 1 |
Finding new probability models to handle over-dispersed data is crucial in addressing challenges posed by over-dispersion. <|MaskedSetence|> In this context, we introduced a novel count data model, the Poisson-Copoun distribution, along with its regression framework and a three inflated version. The model combines the... | **A**: Furthermore, residual analysis using randomized quantile residuals validated the robustness of the fitted model.
.
**B**: Appropriately selecting such models for various scenarios leads to better data fits and more reliable interpretations.
**C**: A regression model based on the PCD was also developed, and its... | BCA | BCA | BCA | CAB | Selection 2 |
3 Calibration PI with Kernel Estimators
Besides the cPI with DNN estimators, we can consider a cPI relying on the estimated value of CDF evaluated at qjq_{j} given XfX_{f} based on a standard kernel estimator. <|MaskedSetence|> Then, we can build a cPI according to the same procedure that was applied before. We sho... | **A**: Let’s assume that we have built the appropriate kernel estimator F^Y|Xf\widehat{F}_{Y|X_{f}}, and we can evaluate it at the same grid points we used to determine the cPI with DNN.
**B**: Also, we develop the theory to show the asymptotic validity of the corresponding cPI.
**C**: (2021).
| ACB | ACB | ABC | ACB | Selection 2 |
1.3.2. <|MaskedSetence|> Developing generative models directly in this infinite-dimensional setting hold the promise of resolution-independent behavior, a perspective that has proven successful in Bayesian inverse problems [55] and operator learning [33, 38].
Numerous studies have investigated generative modeling in ... | **A**: We refer the reader to [14] for a recent comprehensive survey of infinite-dimensional diffusion models..
**B**: Generative models in function space
As spatial resolution increases, data distributions can be conceptualized as measures over function spaces.
**C**: In [30] similar techniques are extended to flow... | BCA | CBA | BCA | BCA | Selection 3 |
<|MaskedSetence|> The gradient gtyg_{t_{y}} is obtained directly from the aggregation of the asynchronously generated samples from all workers. <|MaskedSetence|> <|MaskedSetence|> After accumulating nτn\tau iterations across all workers, synchronization occurs to compute the LSAM score according to Equation˜5 and d... | **A**:
We use Nesterov momentum (Nesterov, 1983) in both the inner sampling loop and the outer optimization step.
**B**: Each worker independently generates samples in parallel, adhering to the conditional distribution in Equation˜5.
**C**: This setup constitutes a two-time-scale stochastic approximation (Doan, 2023... | BAC | ACB | ACB | ACB | Selection 2 |
Figure 7.11. <|MaskedSetence|> The dashed lines corresponding to noise-to-signal ratios under different noise levels are computed by (7.4).
We observe that the random Fourier feature model trained with resampling effectively performs denoising across all tested noise-to-signal ratios—0.25%0.25\%, 1.0%1.0\%, and 4.... | **A**: The model achieves a relative generalization error approximately one order of magnitude smaller than the corresponding NSR values.
**B**: Relative generalization error of the trained random Fourier feature model β(x)\beta(x) with varied noise level parameter ss and increased network size KK.
**C**: Moreover, ... | BAC | BAC | ABC | BAC | Selection 1 |
<|MaskedSetence|> <|MaskedSetence|> BNs offer a flexible data-driven approach, which allows to more accurately reflect the underlying causal structure of the data, thereby reducing the risk of misspecification errors that can lead to biased estimates and inefficiencies in the causal inference process.
In addition to ... | **A**: BNs, with their graphical representation of probabilistic relationships, inherently address these challenges by allowing for the inclusion of multiple covariate interactions and dependencies without pre-imposed constraints.
BNs significantly improve the precision and reliability of Average Treatment Effect (ATE)... | BCA | BCA | BCA | ABC | Selection 3 |
<|MaskedSetence|> Each session followed a fixed route with both high-stress urban and low-stress highway segments. Participants wore E4 devices on both wrists, of which only the left wrist data is used. <|MaskedSetence|> Stress events were defined using a threshold of 0.75 (Bustos et al., 2021). <|MaskedSetence|> | **A**: The dataset contains 10.7 hours of data.
.
**B**: An observer continuously rated driver stress using a slider (0 = no stress, 1 = extreme stress), validated by participants.
**C**:
ROAD: The ROAD dataset (Haouij et al., 2018) was collected from 14 driving sessions with 10 adults.
| BAC | CBA | CBA | CBA | Selection 3 |
<|MaskedSetence|> As explained in [5], t-SNE can be viewed as a method that first clusters the data while embedding it and then aligns the clustered data. We propose an alternate method that consists of first clustering the data, then embedding each cluster individually, and finally finding rigid transformations to al... | **A**: Additionally, the method includes a tuning parameter to increase separation between clusters in a way that preserves relative distances between clusters.
**B**: Our approach is modular in that the clustering and embedding methods may be chosen by the user based on what information they aim to obtain from the vi... | CAB | ACB | CAB | CAB | Selection 1 |
In this section we perform simulation studies to investigate the performance of the MMG. We first generate data from a Gaussian graphical model and examine the performance of G-MMG. Next, we generate data from a mixture of Gaussian graphical models and apply MP-MMG. <|MaskedSetence|> Note that MMG identifies the true ... | **A**: We compare four imputation strategies, including complete-case (CC) analysis, MICE (Van Buuren, 2018), missForest (Stekhoven and Bühlmann, 2012), and the proposed G-MMG/MP-MMG.
**B**: Our goal is to estimate the marginal medians of the variables.
**C**: In both scenarios, missing values are introduced under MC... | CAB | CBA | CBA | CBA | Selection 4 |
In Figure 2, we additionally plot the distribution obtained for the policy πθ\pi_{\theta} maximizing criteria j(x)=xj(x)=\sqrt{x}, as well as the policies obtained by optimizing IPS and LS. Looking in the left plot, the first observation is that IPS is overly-confident, predicting an impossible outcome and suffers an ... | **A**: These methods converge to policies that play a diverse set of actions, which are still good enough to increase the reward of π0\pi_{0}.
.
**B**: IPS converges to a nearly deterministic policy, consistently choosing the same action.
**C**: The right plot in Figure 2 displays the entropies of these policies. ... | CBA | CBA | CBA | ACB | Selection 2 |
<|MaskedSetence|> <|MaskedSetence|> Note that Fig. 2 only reports the MSE of OPE-based methods—PWLL-based methods exhibit extremely high MSE that distorts the plot scale and makes visual comparison uninformative. Still, PWLL-based methods consistently achieve strong reward despite such poor OPE performance, which is ... | **A**: Effect of batch size and learning rate schedule on final validation reward.
**B**: More broadly, good OPE performance (i.e., low MSE) does not correlate with good OPL performance (i.e., high validation reward), and the converse also holds.
**C**: We observe this even within OPE-based methods: greater optimizat... | CBA | CBA | CBA | CBA | Selection 3 |
These four variables may be under the influence of three groups of factors. <|MaskedSetence|> Clearly, patient age affects many physiological and epidemiological variables, so that age affects almost all variables of the perioperative process. <|MaskedSetence|> <|MaskedSetence|> For instance, hip replacements are ... | **A**: Also, there are many procedures that affect different age groups in a different way.
**B**: For instance, in old age, recovery from anesthesia takes longer.
**C**: First, there are patient factors such as age, sex, and weight.
| CBA | CAB | CBA | CBA | Selection 1 |
In Figure 17, we show the LRPC analysis of one of the regions highlighted in Table 2. <|MaskedSetence|> Regressions were statistically significant, but only those highlighted were clinically significant. We also notice how the correlation increases throughout the AE. <|MaskedSetence|> <|MaskedSetence|> Notably, sign... | **A**: A summary of the methods and the rest of the parameters and the LRPC analysis reveals regions established as part of the pattern of neurodegeneration, especially in the NOR vs MCI comparison, tables 4 and 6.
Table 4 summarizes the number of significant and non-significant brain regions identified using PCA and... | BAC | BCA | BAC | BAC | Selection 4 |
Method
Data Sources and Analysis
Hospital admission data were obtained from the New York State (NYS) Department of Health’s Statewide Planning and Research Cooperative System (SPARCS) [13], which captures approximately 95% of all hospital discharges in NYS. The SPARCS database includes deidentified administrative inf... | **A**: Hospitalization rate was calculated by dividing the number of hospital admissions by the total population and expressed as rates per 100,000 population.
**B**: Frequency was selected based on prominent amplitudes in the periodogram and their alignment with known human activity patterns.
**C**: For this study, ... | CAB | ACB | CAB | CAB | Selection 1 |
<|MaskedSetence|> It helps preserve context continuity—preventing important sentences from being split —at the cost of indexing more passages. Too little overlap risks breaking up meaningful context across chunk boundaries.
Top-k specifies how many chunks are returned to the LLM prompt. Higher values increase covera... | **A**: Standard RAG retrieves the KK most relevant chunks by maximising the total score:.
**B**: Lower values reduce noise but may miss critical context.
**C**:
Chunk overlap determines how much text is shared between adjacent chunks (typically 25-50% of chunk size).
| CBA | ABC | CBA | CBA | Selection 3 |
Many methods have been developed over the past fifteen years aiming to leverage node covariates to improve estimation of latent network structure such as community memberships, latent positions or graphon structure (Xu et al. 2012; Yang et al. 2013; Zhang et al. 2016; Binkiewicz et al. 2017; Su et al. 2020; Chandna et ... | **A**: 2023; Li et al.
**B**: 2024).
Since they are focused on estimation of latent network structure, these methods do not directly yield a way to assess whether node features are predictive of (or predicted by) network structure.
Indeed, most of these methods make the implicit assumption that some or all observed no... | BAC | BAC | BAC | BAC | Selection 4 |
One of the recent PinT advances, the so-called GParareal, introduced in Pentland et al. (2023b), is especially relevant for this work. <|MaskedSetence|> This approach was later extended to utilize nearest-neighbor GPs (nnGPs) in Gattiglio et al. (2025), resulting in the nnGParareal algorithm, offering improved scalabi... | **A**: (2024), uses shallow random weights neural networks to learn the solvers’ discrepancy, allowing its application to partial differential equations (PDEs), accommodating up to 10510^{5} spatial discretization points.
**B**: Although its origins can be traced back to the pioneering works of Suldin (1959) and Larki... | CAB | BAC | CAB | CAB | Selection 1 |
<|MaskedSetence|> They serve as the foundation for major corporate revenue models, as seen in companies like Google, Microsoft, Amazon, and Meta. <|MaskedSetence|> While substantial research has focused on optimizing platform performance, relatively little is known about how to leverage real data to evaluate alternat... | **A**: This stands in contrast to real-time bidding (RTB) (Choi
et al., 2020), where a well-developed causal inference literature exists.
**B**:
Advertising platforms play a fundamental role in the global economy.
**C**: These platforms connect advertisers with content providers—whether search engines, newspapers, r... | BCA | BCA | BCA | CBA | Selection 3 |
The central objective of this work is to establish rigorous statistical guarantees for the spatial empirical measure as an estimator of its population counterpart and, by extension, for the empirical KRD as an estimator of the population KRD. <|MaskedSetence|> By computing the KRD for varying choices of CC, the KRD e... | **A**: For the former assumption, we develop in Subsection 4.2 a relation to time-reduced factorial covariance measures of point processes..
**B**: The KRD constitutes a conceptually appealing functional for comparing finite measures, as it naturally incorporates the geometry of the underlying spatial domain.
**C**: ... | CBA | BCA | BCA | BCA | Selection 4 |
<|MaskedSetence|> In general, for any given training dataset, we could always construct opposing causal models which are equally compatible with the observed data [5] and would render the same model or metric appropriate or inappropriate. This is the crux of this section. <|MaskedSetence|> <|MaskedSetence|> We illus... | **A**: In medical imaging, however, we deal with a wide range of dataset characteristics stemming from the use of various imaging modalities, different patient populations, clinical tasks, diagnostic processes and workflows, each contributing to the underlying causal processes with different potential sources of bias [... | CBA | BAC | CBA | CBA | Selection 3 |
5.1 Criteo Uplift Prediction Dataset Application
First, we apply the proposed method to the Criteo Uplift Prediction Dataset, which is released by the Criteo AI Lab along with the paper [4]. <|MaskedSetence|> The binary treatment variable shall be independent of the features by design based on the data description i... | **A**: The average visit rate is 0.047.
**B**: There are two binary outcome variables: visit and conversion, and we consider to model the visit outcome variable here as suggested in the paper.
**C**: This dataset has nearly 14MM rows in total with 12 anonymized feature values.
| CBA | ACB | CBA | CBA | Selection 4 |
Pretraining has been the most computationally expensive component in the training pipeline for large language models, accounting for over 95% of the cost in DeepSeek V3 DeepSeek-AI et al. <|MaskedSetence|> (2025a) is also comparatively much smaller.
Until recently, AdamW has been the standard optimizer. Recent studie... | **A**: (2025); Ma et al.
**B**: (2025b), and the additional RL training cost in DeepSeek R1 DeepSeek-AI et al.
**C**: (2025); Wang et al.
| BCA | BCA | BCA | BCA | Selection 4 |
<|MaskedSetence|> For personality traits (left), alignment ranges from 45–62%, with agreeableness showing the highest alignment (62%) and neuroticism the lowest (45%). In all cases, the estimated 95% CIs overlap with 50% level expected by chance under random directional alignment. Behavioral tasks (middle) show even m... | **A**: Model-level results (right) reveal that the alignment for most model is no better than chance (e.g., 43–50% for smaller LLaMA and Qwen models).
**B**:
Alignment Across Traits, Tasks and Models.
In Figure 3, alignment proportions vary across traits, tasks, and models.
**C**: These patterns suggest no alignme... | BAC | BAC | BAC | BAC | Selection 4 |
Fully sequential IZ FCPs introduce one tolerance level for each constraint, and this tolerance level specifies how much the decision-maker is willing to be off from a constant threshold for checking the feasibility of the systems with respect to the constraint. This is the least absolute difference in the performance m... | **A**: (2025) examined the impact of the first-stage sampling size and proposed improved versions of the IZ-free procedures for selecting the best system.
**B**: Nevertheless, when the tolerance level is too small relative to the difference between the true expected performance measure and the threshold, computational... | BCA | BCA | BCA | ABC | Selection 1 |
<|MaskedSetence|> <|MaskedSetence|> It evaluates how much new input data deviates from the training distribution using empirical cumulative distribution function (ECDF)-based distance metrics. This helps estimate potential accuracy drops and prompts human intervention when data shifts are detected [8].
SafeML is on... | **A**: Threshold tuning remains a challenge; automated threshold adjustment methods have been proposed [14].
.
**B**: Runtime monitoring and human oversight are essential for safe ML deployment.
**C**: SafeML supports this by applying statistical methods to assess model reliability during operation.
| BCA | BAC | BCA | BCA | Selection 4 |
<|MaskedSetence|> <|MaskedSetence|> As long as the statistical formula and the clinician receive the same information, the data can be anything, be it interviews, life history, mental tests, or other biometrics.
Obviously, such data has to be transformed into a machine-readable format somehow. Here’s another place ... | **A**: To make the decision, Meehl assumes the clinician has the same data as the statistical rule.
**B**: He belabors the distinction between the kind of data and the mode of combining the data.
**C**: Now, patient data is charted in detailed electronic health records, and these can be processed by large language mo... | ABC | BCA | ABC | ABC | Selection 4 |
<|MaskedSetence|> <|MaskedSetence|> Lower case letters are instances of these variables. White nodes correspond to the PM, gray nodes to C and striped nodes are common to both. <|MaskedSetence|> The arcs involving the screening results are characterized in [7], as well as an optimal screening strategy with respect t... | **A**: The variables involved are the covariates XX, the state of CRC, the decision of screening 𝒮\mathcal{S}, the result of screening ℛ\mathcal{R}, the incentive ℐ\mathcal{I} and the utilities UU of the PM and the citizen.
**B**: The biagent influence diagram (BAID) in Figure 3(a) captures the essence of the problem... | BAC | BAC | ABC | BAC | Selection 4 |
The approach proposed in this paper builds off of meta-analytic prediction intervals (Riley et al., 2011), which are commonly used to estimate a range of potential values for an average effect in a new study. <|MaskedSetence|> We apply this prediction interval approach to the CATE, and we extend the method to the case... | **A**: Notably, this idea of a new study can also be considered a “target setting” or a set of future patients (IntHout et al., 2016); we use these concepts interchangeably in this paper.
**B**: This group represents patients for whom clinicians are interested in understanding treatment effects to aid with decision-ma... | ACB | CBA | ACB | ACB | Selection 4 |
In this setting, the non-conformity score is obtained from the model output. The model has been trained using part of the data, but the conformal predictor is using a separate calibration dataset. <|MaskedSetence|> First, the efficiency-confidence trade-off, as characterized in [CMLB24] in terms of conditional entropy... | **A**: Through this interpretation, the authors provided various information-theoretic inequalities that relate the efficiency of conformal prediction, characterized by the averaged prediction set size, to the information-theoretic measures of uncertainty given by the conditional entropy.
Two questions follow from th... | BAC | ACB | ACB | ACB | Selection 3 |
Future research could also explore applying C-ILM to the geographically dependent ILM (GD-ILM) proposed by Mahsin et al. (2022), which incorporates individual-level spatial location, spatially varying regional-level risk factors (e.g., socioeconomics, environment), and unobserved spatial structure into the susceptibili... | **A**: Another potential application is integrating C-ILM into the behavioural change ILM (BC-ILM), a framework that incorporates individual-level information and behavioural change effects modeled by “alarm” functions (Ward et al., 2023, 2025).
**B**: The susceptibility function could vary by cluster, accounting for ... | CAB | BAC | BAC | BAC | Selection 4 |
<|MaskedSetence|> In practice, the standard SIMEX algorithm assumes the measurement error follows an additive structure with errors distributed as N(0,σ2)N(0,\sigma^{2}), where σ2\sigma^{2} is either known or estimated. The standard SIMEX method involves three main steps: simulation, estimation, and extrapolation. <... | **A**: In the estimation step, the model’s parameters are estimated repeatedly with varying levels of introduced error.
**B**: In the simulation step, additional measurement error is artificially introduced into the data.
**C**:
2.2 The SIMEX algorithm for Poisson distributed surrogates
The SIMEX algorithm, introdu... | CBA | BAC | CBA | CBA | Selection 4 |
<|MaskedSetence|> We
focus on a particular family of deterministic ensemble Kalman-Bucy filters (EnKBF) where the model noise has been replaced by an interacting particle term that reflects the ensemble spread. Furthermore, we consider a localised variant which is given by the solution of a coupled system of SDEs.
The... | **A**: This is necessary because the data assimilation update and contraction of the system often cause the particles to become very similar.
**B**: Here, we consider accurate partial observations with linear time-dependent observation operator, h(t,Xt):=HtXth(t,X_{t}):=H_{t}X_{t}.
**C**: Inflation is utilized to p... | BAC | BAC | BCA | BAC | Selection 2 |
The ML estimation method of non-centered stationary discrete- and continuous-time Gaussian models with long memory, short memory, or anti-persistence, referred to as general Gaussian processes, is the focus of this paper. <|MaskedSetence|> The choice of ML estimation is motivated by the practical need for accurate es... | **A**: Since these memory properties are relevant across various applications, our goal is to develop an estimation method that does not impose prior restrictions on the memory type of the process.
**B**: (2024)—are not recommended.
**C**: For example, Corsi (2009) criticized semi-parametric methods for producing sig... | ABC | ABC | ABC | ABC | Selection 2 |
The top panel shows the ordered eigenvalues on the log-scale, colored by tjt_{j}. The lighter colors (toward yellow) are at earlier dates, and the darker (brown) are later. <|MaskedSetence|> ARB5 and ETH5 consistently stabilize as linear afterward, with stable slopes and slightly shifted intercepts over time periods... | **A**: The stability across dates suggests that a lower rank structure is suitable for ARB5 and ETH5, while ETH30 may be too unstable.
.
**B**: Moving to ETH30, it has a similar shape to the other two prior to 2023 (yellow and light orange), but exhibits a persistent regime shift in late 2023, with a slower decay bot... | CBA | CBA | CAB | CBA | Selection 1 |
Recently, Gizewski et al. [15] studied spectral algorithms (4) under covariate shift, assuming that the density ratio is uniformly bounded. Ma et al. [26] and Feng et al. <|MaskedSetence|> Gogolashvili et al. <|MaskedSetence|> Fan et al. <|MaskedSetence|> All these works are confined to well-specified settings, wher... | **A**: [12] investigated kernel ridge regression—a special case of spectral algorithms (4)—under the condition that the density ratio is either bounded or unbounded but has finite second moment.
**B**: [10] adopted the moment condition from [16] and analyzed spectral algorithms (4) within covariate shift.
**C**: [16]... | ACB | BAC | ACB | ACB | Selection 3 |
In order to identify the mode shape patterns, we can determine the signs of the mode shape components using engineering judgment. These identified mode shape patterns are then normalized to ensure a maximum value of one for comparison purpose. The resulting normalized mode shapes are compared to the true mode shapes o... | **A**: It is observed that the identified mode shapes obtained using the variance of the measurements (labeled as "Estimated (SD)" in the figure) are matching quite well with the true mode shapes for both TR-I and TR-II scenarios.
**B**: This discrepancy is likely due to the higher velocity of the moving sensor in the... | BCA | ACB | ACB | ACB | Selection 2 |
<|MaskedSetence|> <|MaskedSetence|> Fig. 6 shows that different seeds applied to the same data lead to different local minima and (numerically) quite different basis images. <|MaskedSetence|> To address this question, we are working on semantic analysis tools, which can apply to NMF. (2) Information theoretic tools ... | **A**: (1) We identified several scenarios of nonidentifiability related to NMF; e.g.
**B**:
There are several problems left for the future.
**C**: This does not necessarily mean that these different basis images refer to different semantic features.
| CAB | BAC | BAC | BAC | Selection 2 |
Both neural network-based and parametric approaches have notable limitations. Neural methods, while flexible and capable of quantifying predictive uncertainty as demonstrated in the Extended Deep Triangle (EDT) model, often lack interpretability, particularly in how they represent dependence structures (Cai et al.,, 20... | **A**: Building on this, the Extended Deep Triangle (EDT) incorporates dependence between two LOBs by modeling pairwise and sequential relationships in loss ratios (Cai et al.,, 2025).
**B**: Parametric models, on the other hand, are more interpretable but may suffer from model misspecification and fail to fully lever... | BAC | ABC | BAC | BAC | Selection 3 |
<|MaskedSetence|> These posterior draws are typically sampled through MCMC approaches. <|MaskedSetence|> Stan requires full specification of the prior, likelihood and data. <|MaskedSetence|> These algorithms provide a relatively efficient sampling approach when considering the time investment needed to self-code a s... | **A**: The software package, Stan, was used to produce the posterior chains presented in the remainder of this work.
**B**: Stan utilizes the no-U-turn sampler (NUTS) and Hamiltonian Monte Carlo (HMC) algorithms.
**C**:
3.2 Posterior Sampling
A sample from the posterior distribution is the most common approach for ... | CAB | CAB | ABC | CAB | Selection 4 |
2.2.1. <|MaskedSetence|> <|MaskedSetence|> Input features include user demographics and past consumption behavior across content types aggregated over various time windows, content representations, contextual signals (e.g., device type). <|MaskedSetence|> We encode temporal signals (e.g., time of day and day of wee... | **A**: Learning Model
To learn the reward function hh, we use a fully connected neural network (MLP) with two hidden layers (sizes 256 and 64), ReLU activations, and a binary cross-entropy loss function.
**B**: We learn embeddings for categorical features such as country to learn a richer representation of similar it... | ACB | ACB | ACB | ACB | Selection 1 |
<|MaskedSetence|> They are used to efficiently study the effects of endogenous time-varying covariates on the survival response; we can study the strength of association between the hazard of the event and the time-varying covariate by implementing joint models. <|MaskedSetence|> That is, as time progresses and addit... | **A**: In this work, we propose a new method as an extension of JLS modelling by using a similarity-based approach to improve the dynamic predictive performance of JLS models.
.
**B**:
Joint longitudinal-survival (JLS) models were developed to accommodate the unique characteristics of longitudinal and time-to-event... | BCA | ABC | BCA | BCA | Selection 1 |
In the present work, we propose a broader definition of discrete stability by replacing location shifts with Poisson translation, which has been previously described in [9]. <|MaskedSetence|> These broadly discrete stable distributions form a natural generalization of the strictly discrete stable family of [16]. They ... | **A**: Finally, we show that a subset of the broadly discrete stable distributions are discretely self-decomposable according to the definition of [16], and are therefore unimodal.
**B**: We prove that the mixed Poisson-stable distributions are the unique family with this property.
**C**: On the other hand, the rest ... | BAC | BAC | BAC | CAB | Selection 2 |
The approach is also readily extendable to familiar econometric designs. In regression analysis, it allows researchers to model potential confounders directly within the hypothesis space rather than assuming they are absent. <|MaskedSetence|> <|MaskedSetence|> <|MaskedSetence|> More broadly, panel regressions, instr... | **A**: In difference-in-differences models, for example, the framework relaxes the strict parallel-trends assumption by allowing for latent group-specific heterogeneity, yielding treatment effect posteriors that are typically wider but more credible.
**B**: When only partial information about the covariance structure ... | BAC | BAC | BAC | BCA | Selection 3 |
2.5.2 The ADRENAL trial
The ADRENAL trial was a multicentre, placebo-controlled, double-blind, randomised clinical trial conducted to examine the primary hypothesis that hydrocortisone, compared to placebo, decreases 90-day all-cause mortality in patients admitted to an intensive care unit with septic shock. <|Maske... | **A**: (2018).
.
**B**: The primary endpoint was all-cause mortality 90 days after randomisation.
**C**: Given a 33% 90-day all-cause mortality in patients with septic shock receiving placebo, ADRENAL aimed to establish a clinically meaningful 5% reduction in those receiving hydrocortisone, with a two-sided type I er... | BCA | BCA | BCA | CAB | Selection 2 |
This approach allows us to compute gradient-like updates to the distribution over biomolecular structures using the geometry of optimal transport, providing a natural descent direction even in infinite-dimensional spaces. <|MaskedSetence|> This scheme retains flexibility in representing conformations while benefiting ... | **A**: The resulting gradient flow is discretized using a particle-based approximation, allowing us to represent and evolve a set of structural samples that collectively approximate the optimal distribution.
**B**: By contrast, our method follows an optimize-then-discretize (OTD) paradigm: we first solve a variational... | ABC | CBA | ABC | ABC | Selection 4 |
∙\bullet Finally, computer code with GPU acceleration is made available to users. <|MaskedSetence|> Section 2 presents the inverse modeling and the bilevel optimization problem. <|MaskedSetence|> Numerical examples are presented in Section 4 to demonstrate the performance of the proposed approach. Conclusions and dis... | **A**: For the sensor allocation problem that requires computationally heavy algorithms, it is unrealistic for engineers to adopt the solution unless code is provided.
The remainder of this paper is organized as follows.
**B**: Section 3 investigates two optimization algorithms for solving the proposed bilevel optim... | ABC | ABC | BCA | ABC | Selection 4 |
<|MaskedSetence|> The CDR framework operates directly on the simplex, naturally handles zeros without artificial imputation, and features dual visualization, where the joint display of reduction matrices provides an immediate understanding of the reduction. <|MaskedSetence|> For estimation, we develop the CKDR method... | **A**: Within this framework, we formalize compositional SDR as an identifiable optimality criterion.
**B**:
6 Discussions
This paper proposes a novel approach for interpretable dimension reduction of compositional data.
**C**: Python codes for the proposed method and experiments are available at https://github.co... | BAC | BAC | CAB | BAC | Selection 4 |
This work establishes that, under non-parametric mixture models with Gaussian or Poisson components, the behavior of the likelihood ratio test (LRT) is governed by the structure of the null mixing distribution g0g_{0}. <|MaskedSetence|> <|MaskedSetence|> This divergence is based on a new and general divergence mecha... | **A**: When g0g_{0} is finitely discrete, the LRT converges, exhibiting an effectively finite-dimensional behavior despite the non-parametric model class.
**B**: Those results substantially advance our fundamental understanding of likelihood ratio statistics in non-parametric mixture models and will be useful for futu... | ACB | BAC | ACB | ACB | Selection 4 |
<|MaskedSetence|> In our framework, each sensor stream (e.g., pressure mats, depth cameras, accelerometers) is represented as a modality-specific graph and processed through residual GCNs. <|MaskedSetence|> <|MaskedSetence|> Extensive experiments demonstrate that graph-based multimodal modeling is significantly more... | **A**:
In this article, we propose GraMFedDHAR, a Graph-based Multimodal Federated Learning framework for differentially private HAR.
**B**: Their embeddings are fused via attention-based weighting, enabling robust multimodal activity classification.
**C**: To protect sensitive information, DP is integrated into the... | ABC | BAC | ABC | ABC | Selection 4 |
In this paper, we developed novel estimators for time-dependent ROC analysis using LTRC data, including the regression estimators and inverse truncation-and-censoring probability weighting (IPW) estimators. The regression estimators are direct extensions of Li 2017, while our IPW approach is inspired by the general w... | **A**: We perform comprehensive simulation studies to evaluate the performance of the AUC estimators in Section 4.
**B**: The rest of the paper is organized as follows.
**C**: In Section 5, we demonstrate the proposed ROC analysis by evaluating the risk score in Chow et al.
| CAB | BAC | BAC | BAC | Selection 2 |
<|MaskedSetence|> <|MaskedSetence|> Severe class imbalance is another, arising when the minority class—often the outcome of greatest interest (e.g., fraud cases, diseased patients)—is dwarfed by the majority class. To address this, we previously introduced Precision–Recall Curve (PRC) classification trees and their e... | **A**: Beyond scalability to “big” data, the autoencoder–random-forest combination is notably robust: it maintains performance in the face of noise, missing entries, and only weakly labeled samples—conditions under which many traditional methods falter.
**B**: Consequently, the approach has gained traction across fina... | ABC | ABC | ABC | ABC | Selection 3 |
Fig.5 shows the ROC and AUC of the best and the worst class based on the difference in the AUC scores. <|MaskedSetence|> <|MaskedSetence|> <|MaskedSetence|> When we decrease the risk level to a relatively more appropriate value, c=0.35c=0.35, the risk-averse method begins to perform better compared to the risk-neut... | **A**: For the same risk level, we consider the class to be the best when the difference between the AUC scores of the risk-averse method and the score of the risk-neutral method is the largest among the differences of all 10 classes.
**B**: Likewise, when this difference is the smallest, we consider this class to be ... | ABC | ABC | ABC | ABC | Selection 3 |
To illustrate more clearly the meaning of the values presented in Tables 8 and 8, consider the following example: if, from Table 6, one sums the number of times DCV detected significant differences for each pair of methods that also showed a significant difference in the benchmark truth, and divides this sum by the tot... | **A**: This is mentioned here merely as a reflection, which could be further explored through additional experiments if deemed relevant.
.
**B**: This provides an estimate of the frequency with which CV falsely detects differences not supported by the reference data.
It is worth noting that interpreting these cases ... | CBA | CBA | CBA | CBA | Selection 1 |
Our approach has a few general limitations. First, the validity of Bayesian post-selection inference depends on the use of well-calibrated priors, and our ability to express realistic beliefs about treatment effect heterogeneity. We show that shrinkage priors can improve coverage, but performance deteriorates when di... | **A**: Second, while the use of decision trees enhances interpretability, the inherent instability of such models remains a concern, especially when small perturbations in the data yield markedly different tree structures.
**B**: In addition, although our results suggest that approximate optimization strategies are of... | BAC | ABC | ABC | ABC | Selection 2 |
End of preview. Expand in Data Studio
README.md exists but content is empty.
- Downloads last month
- 36