Inference Model Research Articles

Particle Gibbs for likelihood-free inference of stochastic volatility models

Particle Gibbs for likelihood-free inference of stochastic volatility models

Directed Acyclic Graphs in Decision-Analytic Modeling: Bridging Causal Inference and Effective Model Design in Medical Decision Making.

Our commentary proposes the application of directed acyclic graphs (DAGs) in the design of decision-analytic models, offering researchers a valuable and structured tool to enhance transparency and accuracy by bridging the gap between causal inference and model design in medical decision making.The practical examples in this article showcase the transformative effect DAGs can have on model structure, parameter selection, and the resulting conclusions on effectiveness and cost-effectiveness.This methodological article invites a broader conversation on decision-modeling choices grounded in causal assumptions.

MD-Unet for tobacco leaf disease spot segmentation based on multi-scale residual dilated convolutions

Identification and diagnosis of tobacco diseases are prerequisites for the scientific prevention and control of these ailments. To address the limitations of traditional methods, such as weak generalization and sensitivity to noise in segmenting tobacco leaf lesions, this study focused on four tobacco diseases: angular leaf spot, brown spot, wildfire disease, and frog eye disease. Building upon the Unet architecture, we developed the Multi-scale Residual Dilated Segmentation Model (MD-Unet) by enhancing the feature extraction module and integrating attention mechanisms. The results demonstrated that MD-Unet achieved 92.75%, 90.94%, 84.93%, and 91.81% for the lesion CPA, recall, IoU, and F1 metrics, respectively, with an overall Dice score of 94.67%. Furthermore, the model parameters, floating-point operations, and inference time per single image for MD-Unet were 4.65 × 107, 2.3392 × 1011, and 65.096 ms, respectively. Compared to Unet, PSP, DeepLab v3+, FCN, SegNet, UNET++, and DoubleU-Net, MD-Unet significantly improved accuracy while effectively managing model complexity, achieving optimal overall performance. This work provides the theoretical foundations and technical support for precise segmentation of tobacco lesions, with potential applications in the segmentation of other plant diseases.

Efficient and accurate variational inference for multilevel threshold autoregressive models in intensive longitudinal data.

Recent technological advancements have enabled the collection of intensive longitudinal data (ILD), consisting of repeated measurements from the same individual. The threshold autoregressive (TAR) model is often used to capture the dynamic outcome process in ILD, with autoregressive parameters varying based on outcome variable levels. For ILD from multiple individuals, multilevel TAR (ML-TAR) models have been proposed, with Bayesian approaches typically used for parameter estimation. However, fitting ML-TAR models can be computationally challenging. This study introduces a mean-field variational Bayes (MFVB) algorithm as an alternative to traditional Bayesian inference. By optimizing to approximate posterior densities, variational Bayes aims to find the best approximation within a defined set of distributions. Simulation results demonstrate that our MFVB algorithm is significantly faster than the standard Markov chain Monte Carlo (MCMC) approach. Moreover, increasing the number of individuals or time points enhances the accuracy of the parameter estimates using MFVB, suggesting that sufficient data are crucial for accurate estimation in complex models like ML-TAR models. When applied to real-world data, the MFVB algorithm was significantly more efficient than MCMC and maintained similar accuracy. Thus, the MFVB algorithm is a faster and more consistent alternative to MCMC for large-scale inference in ILD models.

Architectures and Optimization Strategies for Real-Time Machine Learning Recommendation Systems: A Systematic Review of Scalability Challenges

This article comprehensively analyzes the challenges and solutions in deploying real-time machine learning recommendation systems at scale. The article examines the critical trade-offs between model complexity, inference latency, and system scalability that impact modern recommendation architectures. The article investigates three primary dimensions: infrastructure optimization, model serving strategies, and resource utilization patterns. The article proposes a novel framework for balancing these competing requirements through a combination of distributed computing architectures, hybrid model deployment approaches, and intelligent caching mechanisms. The findings demonstrate that implementing a multi-tiered serving architecture with dynamic resource allocation significantly improves system performance while maintaining recommendation quality. The article also explores emerging optimization techniques, including model quantization, feature store architectures, and adaptive serving strategies. The article contributes to the field by providing a systematic approach to designing and implementing real-time recommendation systems that can effectively handle high-concurrency workloads while delivering personalized suggestions within strict latency constraints. The results offer valuable insights for practitioners and researchers working on large-scale recommendation systems, particularly in environments where real-time performance is crucial.

Bayesian empirical likelihood inference and order shrinkage for a hysteretic autoregressive model

Bayesian empirical likelihood inference and order shrinkage for a hysteretic autoregressive model

An Interpretable Deep Inference Model With Dynamic Constraints for Forecasting the Evolution of Sea Surface Variables in the South China Sea

An Interpretable Deep Inference Model With Dynamic Constraints for Forecasting the Evolution of Sea Surface Variables in the South China Sea

Deciphering allosteric mechanisms in KRAS activation: insights from GTP-induced conformational dynamics and interaction network reorganization.

The conformational dynamics and activation mechanisms of KRAS proteins are of great importance for targeted cancer therapy. However, the detailed molecular mechanics of KRAS activation induced by GTP binding remains unclear. In this study, we systematically investigated how GTP/GDP exchange affects the thermodynamic and kinetic properties of KRAS and explored the activation mechanism using molecular dynamics (MD) simulations, Markov state models (MSMs), and neural relational inference (NRI) models. Our MD simulation results show that GTP binding significantly enhances the conformational flexibility of KRAS, and thus promotes its transition to an active conformation with more open switch I and II regions. MSMs analyses show that KRAS in the GTP-bound state can be transitioned to the active state more efficiently during the simulation than in the GDP-bound state. In addition, NRI model calculations showed that GTP binding enhanced residue-residue interactions within the KRAS protein, especially when the long-range interactions were significantly enhanced. Furthermore, the allosteric signaling pathways from the P-loop to switch I and II as well as the key amino acid sites along the pathways were obtained using a graph-based shortest path analysis. Our results can contribute to a deeper understanding of the mechanism of KRAS allosteric activation and provide a foundation for the development of targeted therapeutic drugs to regulate KRAS activity.

A Revealed Preference Approach to Identification and Inference in Producer-Consumer Models

This article provides a new identification result for a large class of models in which consumers participate in production. I show that consumer preferences are necessary and sufficient to identify production functions through cross-equation restrictions implied by first-order conditions. In addition, I derive a nonparametric revealed preference characterization of the class of models that exhausts its empirical implications. Finally, I use a novel and easy-to-apply inference method that is valid under partial identification. This method can be used to statistically test the model, can deal with any type of latent variables (e.g., measurement error), and can be combined with standard exclusion restrictions. Using data on shopping expenditures and shopping intensity from the NielsenIQ Homescan Dataset, I show that a doubling of shopping intensity decreases prices paid by about 15 % . At the same time, I find that search costs are significant, hence, largely diminishing benefits of price search.

Analysis of fractional Black-Scholes, Ornstein-Uhlenbeck, and Langevin models: a minimum distance estimation approach

The present research investigates parameter estimation in fractional stochastic models, specifically the Fractional Black-Scholes, Fractional Ornstein-Uhlenbeck, and Fractional Langevin models. These models, driven by fractional Brownian motion (fBm), capture long-memory effects and complex dependencies often observed in financial and physical systems. To efficiently estimate the parameters of these models, we apply the Minimum Distance Estimation (MDE) method. This approach tackles the challenges posed by long-range dependence, ensuring both asymptotic consistency and normality. We analyze the asymptotic properties of the estimators, including their consistency, asymptotic normality, and efficiency. By providing a unified theoretical framework, the current research highlights the robustness and applicability of the MDE method for parameter inference in fractional models. The findings contribute significantly to diverse fields such as quantitative finance, econometrics, and statistical physics.

Posttraining Network Compression for 3D Medical Image Segmentation: Reducing Computational Efforts via Tucker Decomposition.

"Just Accepted" papers have undergone full peer review and have been accepted for publication in Radiology: Artificial Intelligence. This article will undergo copyediting, layout, and proof review before it is published in its final version. Please note that during production of the final copyedited article, errors may be discovered which could affect the content. Purpose To investigate whether the computational effort of 3D CT-based multiorgan segmentation with TotalSegmentator can be reduced via Tucker decomposition-based network compression. Materials and Methods In this retrospective study, Tucker decomposition was applied to the convolutional kernels of the TotalSegmentator model, an nnU-Net model trained on a comprehensive CT dataset for automatic segmentation of 117 anatomic structures. The proposed approach reduced the floating-point operations (FLOPs) and memory required during inference, offering an adjustable trade-off between computational efficiency and segmentation quality. This study utilized the publicly available TotalSegmentator dataset containing 1228 segmented CTs and a test subset of 89 CTs, employing various downsampling factors to explore the relationship between model size, inference speed, and segmentation accuracy, evaluated using the Dice score. Results The application of Tucker decomposition to the TotalSegmentator model substantially reduced the model parameters and FLOPs across various compression ratios, with limited loss in segmentation accuracy. Up to 88% of the model's parameters were removed, with no evidence of differences in performance compared with the original model for 113 of 117 classes after fine-tuning. Practical benefits varied across different graphics processing unit architectures, with more distinct speed-ups on less powerful hardware. Conclusion The study demonstrates that posthoc network compression via Tucker decomposition presents a viable strategy for reducing the computational demand of medical image segmentation models without substantially impacting model accuracy. ©RSNA, 2025.

Building an analytical model for inference, detection and early warning in UST measurements

Building an analytical model for inference, detection and early warning in UST measurements

An improved U-net and attention mechanism-based model for sugar beet and weed segmentation.

Weeds are a major factor affecting crop yield and quality. Accurate identification and localization of crops and weeds are essential for achieving automated weed management in precision agriculture, especially given the challenges in recognition accuracy and real-time processing in complex field environments. To address this issue, this paper proposes an efficient crop-weed segmentation model based on an improved UNet architecture and attention mechanisms to enhance both recognition accuracy and processing speed. The model adopts the encoder-decoder structure of UNet, utilizing MaxViT (Multi-Axis Vision Transformer) as the encoder to capture both global and local features within images. Additionally, CBAM (Convolutional Block Attention Module) is incorporated into the decoder as a multi-scale feature fusion module, adaptively adjusting feature map weights to enable the model to focus more accurately on the edges and textures of crops and weeds. Experimental results show that the proposed model achieved 84.28% mIoU and 88.59% mPA on the sugar beet dataset, representing improvements of 3.08% and 3.15% over the baseline UNet model, respectively, and outperforming mainstream models such as FCN, PSPNet, SegFormer, DeepLabv3+, and HRNet. Moreover, the model's inference time is only 0.0559 seconds, reducing computational overhead while maintaining high accuracy. Its performance on a sunflower dataset further verifies the model's generalizability and robustness. This study, therefore, provides an efficient and accurate solution for crop-weed segmentation, laying a foundation for future research on automated crop and weed identification.

The Impact of Farmland Transfer on Urban–Rural Integration: Causal Inference Based on Double Machine Learning

Urban–rural fragmentation represents a significant challenge encountered by nations globally, particularly in both developing and developed contexts, during the modernisation process. This study examines the effects of rural land system reform on facilitating integrated development between urban and rural areas. The analysis of the impact of the 2010 liberalisation of the land transfer policy employs a dual machine learning model, utilising provincial-level data from China spanning 2005 to 2022, to address the limitations of traditional causal inference models while ensuring estimation accuracy. The findings indicate that the reform of the rural land system significantly enhances integrated urban–rural development, particularly in demographic, economic, and ecological dimensions. The mechanisms encompass the facilitation of extensive land and agricultural service operations, the development of new business entities, the migration of rural labour, and the enhancement of agricultural capital. Furthermore, notable disparities exist in the effects of reforms across various regions, particularly concerning urban–rural integration development and land transfer levels. The policy effects of land transfer exhibit a marginally diminishing trend. The influence of land transfer on urban–rural integration varies with economic development levels, demonstrating a nonlinear relationship, with the most pronounced effects observed in regions with moderate economic development. Additionally, the policy effects of land transfer differ based on geographic location. The impact of land transfer policies varies across geographic regions, with the central region exhibiting the most significant effect, followed by the north-eastern region, the western region, and the eastern region, which shows the least effect. This study provides a reference for advancing the reform of the marketisation of land factors, improving the efficiency of land resource allocation, and regionally and in multiple layers advancing the reform of the rural land system.

Introducing ActiveInference.jl: A Julia Library for Simulation and Parameter Estimation with Active Inference Models.

We introduce a new software package for the Julia programming language, the library ActiveInference.jl. To make active inference agents with Partially Observable Markov Decision Process (POMDP) generative models available to the growing research community using Julia, we re-implemented the pymdp library for Python. ActiveInference.jl is compatible with cutting-edge Julia libraries designed for cognitive and behavioural modelling, as it is used in computational psychiatry, cognitive science and neuroscience. This means that POMDP active inference models can now be easily fit to empirically observed behaviour using sampling, as well as variational methods. In this article, we show how ActiveInference.jl makes building POMDP active inference models straightforward, and how it enables researchers to use them for simulation, as well as fitting them to data or performing a model comparison.

DSF-GAM: a location inference model in social network Twitter

Analyzing the social interactions and texts on Twitter can provide valuable insights into users' behavior, opinions, and even their geographical locations. Location inference of users within a social network finds applications in various domains, such as targeted advertising and public health monitoring. Inference techniques through social interactions, such as co-mentions, remain a favored method for constructing user social networks. Co-mentions, however, introduce a substantial number of noisy relationships that impact the overall network structure and prediction of the location of a user. Furthermore, previous approaches have overlooked the examination of cohesive groups or community structures within an individual's ego network for location prediction. This paper introduces a predictive model called DSF-GAM (DS stand for document similarity, F stands for frequencies and GAM stands for Generalized Additive Model). which redesigns social interactions based on mentions to infer user location on Twitter. DSF-GAM introduces a novel and innovative social graph, termed a user relation graph. DSF-GAM employs a generalized additive model (GAM) for location inference. The fundamental contribution of DSF-GAM lies in the representation of data using a new graph structure with enriched community features and edges. The experimental results on well-known datasets demonstrate that DSF-GAM outperforms other state-of-the-art methods based on both content and network approaches and mixed results in hybrid approaches.

Beyond what was said: Neural computations underlying pragmatic reasoning in referential communication.

The ability to infer a speaker's utterance within a particular context for the intended meaning is central to communication. Yet, little is known about the underlying neurocomputational mechanisms of pragmatic inference, let alone relevant differences among individuals. Here, using a reference game combined with model-based functional magnetic resonance imaging (fMRI), we showed that an individual-level pragmatic inference model was a better predictor of listeners' performance than a population-level model. Our fMRI results showed that Bayesian posterior probability was positively correlated with activity in the ventromedial prefrontal cortex (vmPFC) and ventral striatum and negatively correlated with activity in dorsomedial PFC, anterior insula (AI), and inferior frontal gyrus (IFG). Importantly, individual differences in higher-order reasoning were correlated with stronger activation in IFG and AI and positively modulated the vmPFC functional connectivity with AI. Our findings provide a preliminary neurocomputational account of how the brain represents Bayesian belief inferences and the neural basis of heterogeneity in such reasoning.

Where A/B Testing Goes Wrong: How Divergent Delivery Affects What Online Experiments Cannot (and Can) Tell You About How Customers Respond to Advertising

Marketers use online advertising platforms to compare user responses to different ad content. But platforms’ experimentation tools deliver different ads to distinct and undetectably optimized mixes of users that vary across ads, even during the test. Because expo­sure to ads in the test is nonrandom, the estimated comparisons confound the effect of the ad content with the effect of algorithmic targeting. This means that experimenters may not be learning what they think they are learning from ad A/B tests. The authors document these “divergent delivery” patterns during an online experiment for the first time. They explain how algorithmic targeting, user heterogeneity, and data aggregation conspire to confound the magnitude, and even the sign, of ad A/B test results. Analytically, the authors extend the potential outcomes model of causal inference to treat random assignment of ads and user exposure to ads as separate experimental design elements. Managerially, the authors explain why platforms lack incentives to allow experimenters to untangle the effects of ad content from proprietary algorithmic selection of users when running A/B tests. Given that experimenters have diverse reasons for comparing user responses to ads, the au­thors offer tailored prescriptive guidance to experimenters based on their specific goals.

Contribution of Open Access Databases to Intensive Care Medicine Research: Scoping Review.

Intensive care units (ICUs) handle the most critical patients with a high risk of mortality. Due to those conditions, close monitoring is necessary and therefore, a large volume of data is collected. Collaborative ventures have enabled the emergence of large open access databases, leading to numerous publications in the field. The aim of this scoping review is to identify the characteristics of studies using open access intensive care databases and to describe the contribution of these studies to intensive care research. The research was conducted using 3 databases (PubMed-MEDLINE, Embase, and Web of Science) from the inception of each database to August 1, 2022. We included original articles based on 4 open databases of patients admitted to ICUs: Amsterdam University Medical Centers Database, eICU Collaborative Research Database, High time resolution ICU dataset, Medical Information Mart for Intensive Care (II to IV). A double-blinded screening for eligibility was performed, first on the title and abstract and subsequently on the full-text articles. Characteristics relating to publication journals, study design, and statistical analyses were extracted and analyzed. We observed a consistent increase in the number of publications from these databases since 2016. The Medical Information Mart for Intensive Care databases were the most frequently used. The highest contributions came from China and the United States, with 689 (52.7%) and 370 (28.3%) publications respectively. The median impact factor of publications was 3.8 (IQR 2.8-5.8). Topics related to cardiovascular and infectious diseases were predominant, accounting for 333 (25.5%) and 324 (24.8%) articles, respectively. Logistic regression emerged as the most commonly used statistical model for both inference and prediction questions, featuring in 396 (55.5%) and 281 (47.5%) studies, respectively. A majority of the inference studies yielded statistically significant results (84.0%). In prediction studies, area under the curve was the most frequent performance measure, with a median value of 0.840 (IQR 0.780-0.890). The abundance of scientific outputs resulting from these databases, coupled with the diversity of topics addressed, highlight the importance of these databases as valuable resources for clinical research. This suggests their potential impact on clinical practice within intensive care settings. However, the quality and clinical relevance of these studies remains highly heterogeneous, with a majority of articles being published in low-impact factor journals.

Behind the scenes of cellular organization: Quantifying spatial phenotypes of puncta structures with statistical models including random fields.

The cellular interior is a spatially complex environment shaped by non-trivial stochastic and biophysical processes. Within this complexity, spatial organizational principles-also called spatial phenotypes-often emerge with functional implications. However, identifying and quantifying these phenotypes in the stochastic intracellular environment is challenging. To overcome this challenge for puncta, we discuss the use of inference of point-process models that link the density of points to other imaged structures and a random field that captures hidden processes. We apply these methods to simulated data and multiplexed immunofluorescence images of Vero E6 cells. Our analysis suggests that peroxisomes are likely to be found near the perinuclear region, overlapping with the ER, and located within a distance of 1 μm to mitochondria. Moreover, the random field captures a hidden variation of the mean density in the order of 15 μm. This length scale could provide critical information for further developing mechanistic hypotheses and models. By using spatial statistical models including random fields, we add a valuable perspective to cell biology.