Inference Framework Research Articles

Active Inference Models of AV Takeovers: Relating Model Parameters to Trust, Situation Awareness, and Fatigue.

Our objectives were to assess the efficacy of active inference models for capturing driver takeovers from automated vehicles and to evaluate the links between model parameters and self-reported cognitive fatigue, trust, and situation awareness. Control transitions between human drivers and automation pose a substantial safety and performance risk. Models of driver behavior that predict these transitions from data are a critical tool for designing safer, human-centered, systems but current models do not sufficiently account for human factors. Active inference theory is a promising approach to integrate human factors because of its grounding in cognition and translation to a quantitative modeling framework. We used data from a driving simulation to develop an active inference model of takeover performance. After validating the model's predictions, we used Bayesian regression with a spike and slab prior to assess substantial correlations between model parameters and self-reported trust, situation awareness, fatigue, and demographic factors. The model accurately captured driving takeover times. The regression results showed that increases in cognitive fatigue were associated with increased uncertainty about the need to takeover, attributable to mapping observations to environmental states. Higher situation awareness was correlated with a more precise understanding of the environment and state transitions. Higher trust was associated with increased variance in environmental conditions associated with environmental states. The results align with prior theory on trust and active inference and provide a critical connection between complex driver states and interpretable model parameters. The active inference framework can be used in the testing and validation of automated vehicle technology to calibrate design parameters to ensure safety.

Statistical inference with a manifold-constrained RNA velocity model uncovers cell cycle speed modulations.

Across biological systems, cells undergo coordinated changes in gene expression, resulting in transcriptome dynamics that unfold within a low-dimensional manifold. While low-dimensional dynamics can be extracted using RNA velocity, these algorithms can be fragile and rely on heuristics lacking statistical control. Moreover, the estimated vector field is not dynamically consistent with the traversed gene expression manifold. To address these challenges, we introduce a Bayesian model of RNA velocity that couples velocity field and manifold estimation in a reformulated, unified framework, identifying the parameters of an explicit dynamical system. Focusing on the cell cycle, we implement VeloCycle to study gene regulation dynamics on one-dimensional periodic manifolds and validate its ability to infer cell cycle periods using live imaging. We also apply VeloCycle to reveal speed differences in regionally defined progenitors and Perturb-seq gene knockdowns. Overall, VeloCycle expands the single-cell RNA sequencing analysis toolkit with a modular and statistically consistent RNA velocity inference framework.

Early remission of deliberate self-harm predicts emotion regulation capacity in adulthood: 12.4 years follow-up of a randomized controlled trial of adolescents with repeated self-harm and borderline features

AbstractEmotion regulation capacity, critical for adult functioning and mental health, develops strongly during adolescence in healthy individuals. Deficits in emotion regulation is often referred to as emotion dysregulation [ED] and is associated with various mental health problems, including repeated deliberate self-harm [DSH] which peaks in adolescence. Dialectical Behaviour Therapy for adolescents [DBT-A] systematically targets ED through strategies such as changing coping behaviours and has previously been shown to effectively induce DSH remission in adolescents. However, whether such remission is associated with improved emotion regulation capacity in adulthood, and whether this effect is mediated by changes in use of coping strategies has not been previously studied. Prospective long-term follow-up study of an RCT comparing DBT-A with enhanced usual care [EUC] for adolescents presenting to community child and adolescent psychiatric outpatient clinics with borderline personality features and repeated self-harm. Assessments included both structured interviews and self-report at baseline and 1.6, 3.1 and 12.4 years follow-up. In the final follow-up, adult ED was measured and data were collected for 61 (80%) of the original 77 participants. DSH remission was assessed at 1.6 years follow-up, and use of coping strategies at 3.1 and 12.4 years follow-ups. A mediation analysis was conducted within a causal inference framework. Both treatment groups increased their use of functional coping skills from adolescence to adulthood, while only DBT-A was associated with decreases in dysfunctional coping. There was a direct effect of DSH remission 1 year after treatment on adult ED, particularly for participants who did not receive DBT-A. There was a negative association between reductions in dysfunctional coping and adult ED, however this did not mediate the effect of DSH remission. This is the first study to report that early DSH remission in adolescence predicted lower ED in adulthood. These results highlight the importance of early DSH remission and provides new insight into the long-term relationship between DSH and ED. Clinical trial registration information: “Treatment for Adolescents with Deliberate Self-harm”; http://ClinicalTrials.gov/;NCT00675129

Unlocking adolescent dietary behaviours: the causal role of social contexts

Abstract Background Using a causal inference framework, our study aimed to extend previous work in providing a comprehensive understanding of the role various social contexts across different levels can play in predicting adolescent dietary behaviours. Methods Data came from the 2017/18 Health Behaviour in School-aged Children (HBSC) survey in Flanders (Belgium). The sample included representative data from 8,702 adolescents in 165 schools. Social factors at individual-level (family meals, family, peer, student and teacher support) and at contextual-level (school culture of student and teacher support and school-level socioeconomic status [SES]) were considered. Multilevel logistic regression analyses were performed. Results At individual-level, family meals and family and student support were positively associated with fruit and vegetable consumption, family meals with sweet consumption too. Student and teacher support showed positive associations solely with fruit consumption. At contextual-level, adolescents attending schools with a lower school-SES demonstrated lower vegetable consumption. Culture of teacher support was negatively associated with sweets but also vegetable consumption. For culture of student support no associations were found. Conclusions The study mainly provide cues for strengthening family social ties as dietary health promotion initiatives, with some evidence supporting the role of beneficial peer and school social factors. Mixed results for contextual school-level social factors warrant further research to identify additional school-related influences on adolescent dietary behaviours to advance current understandings. Key messages • Family meals and support promote healthy eating in adolescents, emphasising the health promoting role of family social ties in dietary habits. • Mixed findings on school-level social factors call for more research on school health policies.

GLP1-RA versus DPP4 inhibitors on cardiovascular outcomes: A three-year real-world analysis

Abstract Background Cardiovascular outcome trials of GLP1-RAs have massively impacted the clinical practice in the management of patients with type 2 diabetes. The cardiovascular protective properties of GLP1-RA compared to DPP4 have been demonstrated in an observational setting before, however recent methodological developments have enabled clinically relevant effect estimation. Aim This study aimed to evaluate the real-world cardiovascular outcomes of GLP1-RA versus DPP4 inhibitors in patients with type 2 diabetes and high cardiovascular risk. Methods In this study, we utilized the Danish nationwide registers to identify new users of GLP1-RA or DPP4 therapies from 2012 to 2022. Participants were required to be either at least 50 years of age and have one or more of the following conditions: hypertension, chronic obstructive pulmonary disease, cardiovascular disease, or chronic renal disease or be at least 60 years of age and have hypertension. Additionally, their glycated hemoglobin (HbA1c) levels needed to be 53 mmol/mol or higher within the year preceding their first prescription date for these therapies. The index date was defined as the initial prescription date, and comorbidities and medicine use were evaluated up to 10 years prior to this date. Individuals who had filled prescriptions for SGLT2 inhibitors within six months prior to their index date were excluded from the study. The outcome of interest was a composite of cardiovascular death, nonfatal myocardial infarction, or nonfatal stroke (MACE). Patients were followed for up to three years until MACE, all-cause mortality, emigration, or until the end of follow-up on December 31, 2022. To assess the three-year risk of MACE, the Aalen Johansson estimator was used. A casual inference framework was employed estimating the three-year average treatment effect using g-formula based on Cox regression with competing risk. Results A total of 26,190 individuals were included, with 19,494 initiating DPP4 inhibitor therapy and 6,696 initiating GLP1-RA therapy. The three-year risk of MACE was 8.56% (95% confidence interval [CI]: 8.11 to 9.00) for the DPP4 initiator group, compared to 4.76% (95% CI: 4.08 to 5.43) for the GLP1-RA initiator group. Emulating what would have happened had all included individuals initiated GLP1-RA treatment versus had all initiated treatment with DPP4, accounting for age, sex, income, education, degree of urbanization, diabetes duration, baseline hba1c level, co-medicine, and co-morbidities, we estimated an three-year average treatment effect of -2.05% (95% CI:-2.92;-1.18) , p-value<0.001) for GLP1-RA initiation versus DPP4 initiation for MACE. Conclusion Our findings suggest that GLP1-RA use is associated with a significantly lower risk of MACE compared to DPP4 inhibitors in a real-world population of patients with type 2 diabetes and high cardiovascular risk. This supports the cardiovascular protective role of GLP1-RA in clinical practice.

DeepZoning: Re-accelerate CNN Inference with Zoning Graph for Heterogeneous Edge Cluster

Parallelizing CNN inference on heterogeneous edge clusters with data parallelism has gained popularity as a way to meet real-time requirements without sacrificing model accuracy. However, existing algorithms struggle to find optimal parallel granularity for complex CNNS, the structure of which is a directed acyclic graph (DAG) rather than a chain, and the parallel dimension is inflexible. To distribute the workload of modern CNNs on heterogeneous devices is also proven as NP-hard problem. In this paper, we introduce DeepZoning , a versatile and cooperative inference framework that combines both model and data parallelism to accelerate CNN inference. DeepZoning employs two algorithms at different levels: (1) a low-level Adaptive Workload Partition algorithm that uses linear programming and takes spatial and channel dimensions into optimization during the search for feature map distribution on heterogeneous devices, and (2) a high-level Model Partition algorithm that finds the optimal model granularity and organizes complex CNNs into sequential zones to balance communication and computation during execution. Our experimental evaluations show that DeepZoning is effective, achieving up to a 3.02 × speed improvement on our experimental prototype compared to state-of-the-art algorithms.

Unequal effects of adolescent health behaviors on adult cardiometabolic conditions: a cohort study

Abstract Background Health behaviors in adolescence could have a life-long effect on cardiometabolic health and differ by parental socioeconomic conditions. Adopting a life course perspective and a causal inference framework, we aimed to quantify the differential effect of adolescent unhealthy behaviors on adult cardiometabolic conditions by parental financial situation. Methods Using the National Longitudinal Study of Adolescent to Adult Health in the United States (n = 3,772), we estimated conditional causal effect by parental financial situation through inverse probability-weighted marginal structural models. Exposures were adolescent health behaviors (dietary habits, cigarette smoking, alcohol consumption and physical activity) and the effect modifier was parental financial situation (difficulty paying bills vs. no difficulty), both measured at ages 12-19 (1994-1995). Outcomes were cardiometabolic conditions (hypertension, stroke, cardiac failure, diabetes, chronic kidney disease), measured at ages 33 - 43 (2016-2018) using biomarkers and self-reports. Results There were 1,206 participants with cardiometabolic conditions after 20 years of follow-up. Alcohol consumption among adolescents with parental financial difficulty led to 325 more cardiometabolic cases (95% confidence interval: 2 - 647) compared to alcohol consumption by their peers without financial difficulty. This differential effect is driven more by cardiovascular conditions (hypertension, stroke, or cardiac failure) than by metabolic conditions (diabetes or chronic kidney disease). There were no differential effects of dietary habits, cigarette smoking and physical inactivity on the occurrence of adult cardiometabolic conditions. Conclusions Public health initiatives targeting alcohol consumption during adolescence could mitigate socioeconomic inequalities in adult cardiometabolic conditions. Key messages • Socioeconomically disadvantaged adolescents are more susceptible to the long-term detrimental health effects of alcohol consumption. • These susceptible groups may need targeted interventions to mitigate the cardiometabolic effect of alcohol consumption.

Fisher Meets Bayes: The Value of Randomisation for Bayesian Inference of Causal Effects

SummaryFor a Bayesian agent with beliefs about the relationship between covariates and potential outcomes, deterministically selecting an assignment that yields optimal covariate balance rationally dominates randomisation. However, randomisation—by enabling control over the probabilities of erroneous causal conclusions due to unknown covariate imbalances—offers insurance against the possibility that an agent's beliefs may be misleading. For the most part, such rational justifications for optimum assignment have presupposed the framework of Bayesian inference, while such epistemic justifications for randomisation have presupposed the framework of significance testing. In this paper, I build on a conception of balance that seems inextricable from the significance testing framework, Fisherian balance, to show that it implies an analogous epistemic justification for randomisation within the framework of Bayesian inference. Consequently, for the choice between optimum and random assignment, this paper shows that epistemic justifications need not be wedded to significance testing nor must Bayesian inference be wedded to rational justifications.

Causal relationships between diseases mined from the literature improve the use of polygenic risk scores.

Identifying causal relations between diseases allows for the study of shared pathways, biological mechanisms, and inter-disease risks. Such causal relations can facilitate the identification of potential disease precursors and candidates for drug re-purposing. However, computational methods often lack access to these causal relations. Few approaches have been developed to automatically extract causal relationships between diseases from unstructured text, but they are often only focused on a small number of diseases, lack validation of the extracted causal relations, or do not make their data available. We automatically mined statements asserting a causal relation between diseases from the scientific literature by leveraging lexical patterns. Following automated mining of causal relations, we mapped the diseases to the International Classification of Diseases (ICD) identifiers to allow the direct application to clinical data. We provide quantitative and qualitative measures to evaluate the mined causal relations and compare to UK Biobank (UKB) diagnosis data as a completely independent data source. The validated causal associations were used to create a directed acyclic graph that can be used by causal inference frameworks. We demonstrate the utility of our causal network by performing causal inference using the do-calculus, using relations within the graph to construct and improve polygenic risk scores, and disentangle the pleiotropic effects of variants. The data is available through https://github.com/bio-ontology-research-group/causal-relations-between-diseases. Supplementary data are available at Bioinformatics online.

Evidence to inform spatiotemporal management of a western Pacific Ocean tuna purse seine fishery.

Fisheries can profoundly impact co-occurring species exposed to incidental capture. Spatiotemporal fisheries management holds substantial potential to balance socioeconomic benefits with ecological costs to threatened bycatch species. This study estimated the effect of the spatial and temporal distribution of effort by a western Pacific Ocean tuna purse seine fishery on catch rates of target and at-risk species by fitting spatially explicit generalized additive multilevel regression models within a Bayesian inference framework to observer data. Mean field prediction surfaces defined catch rate hotspots for tunas, silky sharks, rays, and whale sharks, informing the design of candidate area-based management strategies. Due to limited sample sizes, odontocete and marine turtle catch rate geospatial patterns were summarized using simple 2D hexagonal binning. Effort could be focused in two areas within core fishing grounds to reduce overlap with hotspots for silky sharks, rays, and whale sharks without affecting target catch. Effort could be shifted outside of core fishing areas to zones with higher target tuna catch rates to reduce overlap with hotspots for at-risk species. Sparse and small marine turtle and whale shark hotspots occurred across the fishing grounds. Results did not identify opportunities for temporally dynamic spatial management to balance target and at-risk catch rates. Research on the economic and operational viability of alternative spatial management strategies is a priority. A small subset of sets had disproportionately large odontocete captures. Real-time fleet communication, move-on rules, and avoiding sets on dolphin schools might reduce odontocete catch rates. Managing set association type and mesh size present additional opportunities to balance catch rates of at-risk and target species. Employing output controls that effectively constrain the fishery would alter the spatial management strategy to focus fishing within zones with the lowest ratio of at-risk bycatch to target tuna catch. Findings inform the design of alternative spatial management strategies to avoid catch rate hotspots of at-risk species without compromising the catch of principal market species.

Bayesian Multilayered Mediation Analysis for Cancer Pharmacogenomics

ABSTRACTMultiomic data from multilevel biological systems are becoming common and motivate integrative modelling approaches to decipher within‐ and cross‐platform dependencies. Mediation analysis aims to identify mediating mechanisms that regulate the effect of an exposure on an outcome. In multiomic contexts, identification of genomic mediators of disease outcomes provides a deeper understanding of mechanisms of disease and corresponding therapeutic targets. While there has been significant work on joint modelling of high‐dimensional potential mediators, approaches that can identify individual mediators in presence of high‐dimensional potential mediators are lacking. We posit that the multiomic data are interrelated following multilayered Gaussian graphical models that include undirected and directed acyclic graphs as special cases. We develop a Bayesian inferential framework for multilayered mediation analysis with continuous, binary, and ordinal outcomes using probit models. As opposed to existing approaches focusing on identifying joint mediation effects, we decompose the joint effect into effects attributable to individual mediators in the framework of interventional mediation analysis. Simulations demonstrate our method outperforms other existing approaches to identify mediators that have nonzero indirect effects to the outcome. We apply our method to multiomic analysis on drug sensitivity outcomes of palbociclib and agents for endocrine therapy, standard care for breast cancer.

Bayesian finite element model inversion of offshore wind turbine structures for joint parameter-load estimation

Operating in harsh and unsteady marine environment, offshore wind turbine (OWT) structures are exposed to unpredictable wind and wave loads. Identifying the structural loads and their effects on the OWTs allow for predicting the remaining fatigue life of these structures and improving the structural design procedure. In this paper, a finite element (FE) model inversion method is presented to estimate the unknown loads and model parameters of OWTs using sparse measurement data. A realistic FE model of an OWT structure with jacket substructure is created in the open-source simulation platform, OpenSees. A Bayesian inference framework is presented to integrate the measured data with the FE model to estimate unknown wind loads and mass of rotor-nacelle assembly. To evaluate the performance of this data assimilation framework, the effect of sensor type, number of sensors, and modeling errors on the estimation accuracy of wind loads and model parameters are investigated through different case studies where synthetic data are used as measurements. The results of this study are important to guide instrumentation of new OWT structures, and to understand the potential limitations and sources of error in the real-world application of this data assimilation framework for joint model parameter and input load estimation.

Discovering dynamics and parameters of nonlinear oscillatory and chaotic systems from partial observations

Despite rapid progress in data acquisition techniques, many complex physical, chemical, and biological systems remain only partially observable, thus posing the challenge to identify valid theoretical models and estimate their parameters from an incomplete set of experimentally accessible time series. Here, we combine sensitivity methods and ranked-choice model selection to construct an automated hidden dynamics inference framework that can discover predictive nonlinear dynamical models for both observable and latent variables from noise-corrupted incomplete data in oscillatory and chaotic systems. After validating the framework for prototypical FitzHugh-Nagumo oscillations, we demonstrate its applicability to experimental data from squid neuron activity measurements and Belousov-Zhabotinsky reactions, as well as to the Lorenz system in the chaotic regime. Published by the American Physical Society 2024

GraphPPL.jl: A Probabilistic Programming Language for Graphical Models

This paper presents GraphPPL.jl, a novel probabilistic programming language designed for graphical models. GraphPPL.jl uniquely represents probabilistic models as factor graphs. A notable feature of GraphPPL.jl is its model nesting capability, which facilitates the creation of modular graphical models and significantly simplifies the development of large (hierarchical) graphical models. Furthermore, GraphPPL.jl offers a plugin system to incorporate inference-specific information into the graph, allowing integration with various well-known inference engines. To demonstrate this, GraphPPL.jl includes a flexible plugin to define a Constrained Bethe Free Energy minimization process, also known as variational inference. In particular, the Constrained Bethe Free Energy defined by GraphPPL.jl serves as a potential inference framework for numerous well-known inference backends, making it a versatile tool for diverse applications. This paper details the design and implementation of GraphPPL.jl, highlighting its power, expressiveness, and user-friendliness. It also emphasizes the clear separation between model definition and inference while providing developers with extensibility and customization options. This establishes GraphPPL.jl as a high-level user interface language that allows users to create complex graphical models without being burdened with the complexity of inference while allowing backend developers to easily adopt GraphPPL.jl as their frontend language.

Seismic Discrimination Using Regional Distance Coda-Wave Ratios

Abstract This study utilizes regional seismic coda-wave amplitude ratios to distinguish earthquakes and declared nuclear explosions in the Democratic People’s Republic of Korea (DPRK). We compute the coda spectral ratios at common stations to isolate source information from propagation and site effects. The resulting frequency-dependent amplitude ratios help identify differences between source types at DPRK. We use parametric models to validate observations, linking them to seismic source parameters. Our method effectively distinguishes the spectral signatures of explosions, nearby earthquakes, and a post-explosion collapse event. The measurements show clear differences between source-type combinations and agree with estimated seismic source models. Notably, the coda spectral ratios between explosions and earthquakes indicate that explosive sources exhibit relatively lower high-frequency S-wave energy and corner frequency compared to P waves. We leverage these spectral differences to develop a Bayesian inference framework for seismic source discrimination at the DPRK test site. This study demonstrates that spectral differences between source types can be exploited in coda spectral ratios to provide critical information for seismic discrimination.

The good, the bad and the ugly: What do we really do when we identify the best and the worst organisations?

Identifying high and poorly performing organisations is common practice in healthcare. Often this is done within a frequentist inferential framework where statistical techniques are used that acknowledge that observed performance...

Precision psychiatry needs causal inference.

Psychiatric research applies statistical methods that can be divided in two frameworks: causal inference and prediction. Recent proposals suggest a down-prioritisation of causal inference and argue that prediction paves the road to 'precision psychiatry' (i.e., individualised treatment). In this perspective, we critically appraise these proposals. We outline strengths and weaknesses of causal inference and prediction frameworks and describe the link between clinical decision-making and counterfactual predictions (i.e., causality). We describe three key causal structures that, if not handled correctly, may cause erroneous interpretations, and three pitfalls in prediction research. Prediction and causal inference are both needed in psychiatric research and their relative importance is context-dependent. When individualised treatment decisions are needed, causal inference is necessary. This perspective defends the importance of causal inference for precision psychiatry.

Statistical inference for Hicks–Moorsteen productivity indices

AbstractThe statistical framework for the Malmquist productivity index (MPI) is now well-developed and emphasizes the importance of developing such a framework for its alternatives. In this paper, we try to fill this gap in the literature for another popular measure, known as the Hicks–Moorsteen productivity index (HMPI). Unlike MPI, the HMPI has a total factor productivity interpretation in the sense of measuring productivity as the ratio of aggregated outputs to aggregated inputs and has other useful advantages over MPI. In this work, we develop a novel framework for statistical inference for HMPI in various contexts: when its components are known or when they are replaced with non-parametric envelopment estimators. This will be done for a particular firm’s HMPI as well as for the simple mean (unweighted) HMPI and the aggregate (weighted) HMPI. Our results further enrich the recent theoretical developments of nonparametric envelopment estimators for the various efficiency and productivity measures. We also examine the performance of these theoretical results for both the unweighted and weighted mean of HMPI for a finite sample, using Monte-Carlo simulations and also provide an empirical illustration along with the computation code.

Salient object detection with bayesian inference based on radar and camera fusion used in UAV obstacle avoidance

Most existing salient object detection methods are sensitive to background noise and rely on prior information in UAV obstacle avoidance applications despite detection methods witnessing rapid progress. In this paper, we propose an efficient framework for salient object detection based on radar-camera fusion and iterative Bayesian optimization. A rough salient object (RSO) image is first built through radar and camera calibration. Next, the RSO image is used to calculate the edge response based on the receptive field mechanism of the primary visual cortex to construct the contour image. Finally, the above two images and the 2D Gaussian distribution are jointly integrated into an iterative Bayesian optimization scheme to get the final salient object image. Different from typical detection methods, this method suppresses background noise by filtering out irrelevant pixels using fusion information. The Bayesian inference framework’s detection performance is improved by precise spatial prior, consisting of optimized contour and RSO images. Experimental results indicate that the presented algorithm performs well against state-of-the-art salient object detection methods on several reference datasets in different evaluation metrics.

A unified EM framework for estimation and inference of normal ogive item response models.

Normal ogive (NO) models have contributed substantially to the advancement of item response theory (IRT) and have become popular educational and psychological measurement models. However, estimating NO models remains computationally challenging. The purpose of this paper is to propose an efficient and reliable computational method for fitting NO models. Specifically, we introduce a novel and unified expectation-maximization (EM) algorithm for estimating NO models, including two-parameter, three-parameter, and four-parameter NO models. A key improvement in our EM algorithm lies in augmenting the NO model to be a complete data model within the exponential family, thereby substantially streamlining the implementation of the EM iteration and avoiding the numerical optimization computation in the M-step. Additionally, we propose a two-step expectation procedure for implementing the E-step, which reduces the dimensionality of the integration and effectively enables numerical integration. Moreover, we develop a computing procedure for estimating the standard errors (SEs) of the estimated parameters. Simulation results demonstrate the superior performance of our algorithm in terms of its recovery accuracy, robustness, and computational efficiency. To further validate our methods, we apply them to real data from the Programme for International Student Assessment (PISA). The results affirm the reliability of the parameter estimates obtained using our method.