Probabilistic Learning Research Articles

PAL: Boosting Skin Lesion Segmentation via Probabilistic Attribute Learning.

Skin lesion segmentation is vital for the early detection, diagnosis, and treatment of melanoma, yet it remains challenging due to significant variations in lesion attributes (e.g., color, size, shape), ambiguous boundaries, and noise interference. Recent advancements have focused on capturing contextual information and incorporating boundary priors to handle challenging lesions. However, there has been limited exploration on the explicit analysis of the inherent patterns of skin lesions, a crucial aspect of the knowledge-driven decision-making process used by clinical experts. In this work, we introduce a novel approach called Probabilistic Attribute Learning (PAL), which leverages knowledge of lesion patterns to achieve enhanced performance on challenging lesions. Recognizing that the lesion patterns exhibited in each image can be properly depicted by disentangled attributes, we begin by explicitly estimating the distributions of these attributes as distinct Gaussian distributions, with mean and variance indicating the most likely pattern of that attribute and its variation. Using Monte Carlo Sampling, we iteratively draw multiple samples from these distributions to capture various potential patterns for each attribute. These samples are then merged through an effective attribute fusion technique, resulting in diverse representations that comprehensively depict the lesion class. By performing pixel-class proximity matching between each pixel-wise representation and the diverse class-wise representations, we significantly enhance the model's robustness. Extensive experiments on two public skin lesion datasets and one unified polyp lesion dataset demonstrate the effectiveness and strong generalization ability of our method. Codes are available at https://github.com/IsYuchenYuan/PAL.

Repeated 5-HT6 receptor activation facilitates flexible behavior in C57BL/6J mice.

Repeated 5-HT6 receptor activation facilitates flexible behavior in C57BL/6J mice.

Dance Technique Error Identification Method Based on Image Segmentation

The rapid advancement of image segmentation techniques has revolutionized motion analysis and error detection in dynamic activities. Conventional approaches to detecting technical errors in dance depend on human observation, making them inherently subjective and difficult to scale. This research presents a novel framework that combines image segmentation, deep learning, and probabilistic modeling to accurately identify and classify errors in dance movements. Our approach utilizes structured learning models, integrating feature extraction, anomaly detection, and neural network-based classification to effectively capture deviations in motion sequences. Unlike heuristic methods, our probabilistic learning framework dynamically adjusts to changing movement patterns, providing both robustness and scalability. To improve detection accuracy, we incorporate an adaptive error correction mechanism that utilizes reinforcement learning. This system refines detection over time by dynamically adjusting classification thresholds based on real-time feedback. This study introduces a probabilistic adaptive error identification model (PAEIM) and an adaptive error mitigation strategy (AEMS) to enhance dance error detection through deep learning and image segmentation. Our framework is designed for real-time analysis, utilizing edge computing to reduce latency and improve responsiveness in live training scenarios. Evaluations on benchmark datasets show that our approach outperforms conventional pose estimation and machine learning-based methods in error detection accuracy. The proposed approach not only improves precision in dance training but also extends accessibility to high-quality dance education and rehabilitation applications. By combining advanced image segmentation, deep learning, and adaptive learning strategies, our framework sets a new standard for AI-driven motion analysis, paving the way for intelligent, data-driven performance assessment and automated feedback systems.

Navigating a varying reward environment in childhood and adolescence

Optimal reward learning requires individuals to adjust their learning rates – the extent to which new information replaces old. Learning rates should be higher in volatile environments, where new information is more salient, and lower in stable environments, where the longer-term history of outcomes is more predictive. It is not known, however, whether this adjustment in learning rates changes with age or is associated with better mental health and social functioning. We administered a child-friendly probabilistic reinforcement learning task with both fixed and fluctuating reward schedules to 121 participants aged 8–16 years. Adjustment of learning rates across childhood and adolescence to suit the levels of uncertainty in the environment did not differ by age, nor was it associated with better mental health and social functioning. Instead we found that learning rates for worse-than-expected outcomes generally decreased with age, temperature increased with age and higher learning rates, specifically during positive stable environments, were associated with greater self-reported prosocial behaviour. Our results highlight the exaggerated impact of negative feedback on children and suggest an increase in exploratory behaviour between childhood and adolescence.

Food-specific decision-making in anorexia nervosa: a comparative study of clinical, at-risk, and healthy control groups

ABSTRACT This study compared individuals with Restrictive Anorexia Nervosa (R-AN; n = 40), Healthy Controls (HCs; n = 45), and individuals at risk for eating disorders (RI; n = 38) using a Reinforcement Learning (RL) paradigm. Participants completed a Probabilistic Reversal Learning (PRL) task involving food-related and neutral contexts. The study examined whether RL impairments in R-AN are context-specific and whether they reflect maintaining factors or preclinical markers. R-AN participants showed reduced learning rates in food-related contexts compared to HC and RI but performed similarly in neutral contexts. Only R-AN individuals showed within-group differences between food and neutral tasks, indicating a disorder-specific impairment. The RI group performed comparably to HCs, suggesting that RL deficits are unlikely to be risk markers. These findings highlight the context-specificity of RL deficits in R-AN, which may act as maintaining factors and could be targeted to improve cognitive flexibility and food-related decision-making.

Prior Expectations of Volatility Following Psychotherapy for Delusions

Persecutory delusions are common, distressing, and difficult to treat. Testing computational neuroscience models of delusions can identify new therapeutic targets. To determine whether change in delusion severity is associated with a corresponding change in volatility priors and brain activation estimated during a belief updating task. This randomized clinical trial was conducted from April 9, 2021, to December 5, 2023, within the Vanderbilt University Medical Center Psychiatric Hospital and at a community mental health center in Nashville, Tennessee. Participants were adults (aged between 18 and 65 years) with schizophrenia spectrum or delusional disorder and an active, persistent (≥3 months) persecutory delusion with strong conviction (>50%). Participants were randomly assigned 1:1 to either cognitive behavioral therapy for psychosis (CBTp)-based intervention or befriending therapy. Intention-to-treat analysis was performed from June 1 to October 31, 2024. The CBTp was a manualized intervention targeting persecutory delusions. The befriending therapy involved engaging in conversations and activities focused on neutral topics. Both interventions were provided in person, lasted for 8 weeks, and included standard care. Standard care consisted of medication management and ancillary services. Primary outcomes were volatility priors (ie, prior expectations of volatility) derived from a 3-option probabilistic reversal learning task; persecutory delusion severity measured by the Psychotic Symptom Rating Scales (PSYRATS delusion subscale; score range: 0-16, with the highest score indicating severe preoccupation, distress, conviction, and functioning impact); and brain activation in the striatum and prefrontal cortex measured by blood oxygenation level-dependent signal change. Associations between volatility priors, clinical improvement, and change in neural activation were examined. Sixty-two participants (median [range] age, 31 [19-63] years; 38 males [61%]) were randomly assigned to the CBTp (n = 32) or befriending therapy (n = 30) arms. A subgroup of 35 participants (57%) completed functional magnetic resonance imaging. Volatility priors decreased following treatment (F1,112 = 7.7 [P = .006]; Cohen d = 0.52 [95% CI, 0.15-0.90]), as did delusion severity (F1,112 = 59.7 [P < .001]; Cohen d = 1.50 [95% CI, 1.00-1.90]), across both groups. The decrease in volatility priors was not associated with clinical improvement in PSYRATS scores (F1,102.8 = 1.8 [P = .18]; Cohen d = 0.26 [95% CI, -0.12 to 0.65]). Activation in the caudate and prefrontal cortex significantly decreased following treatment. Decreased caudate activation was associated with decreased volatility priors (F1,58.3 = 16.6 [P < .001]; Cohen d = 1.07 [95% CI, 0.51-1.61]) but not with PSYRATS total scores. Associations remained significant after controlling for antipsychotic medication (F1,53 = 13.77; P < .001). This randomized clinical trial found that elevated volatility priors and associated activation in the caudate nucleus were amenable to change. Volatility priors could be a potential target for intervention in psychosis. ClinicalTrials.gov Identifier: NCT04748679.

H-calibration: Rethinking Classifier Recalibration with Probabilistic Error-Bounded Objective.

Deep neural networks have demonstrated remarkable performance across numerous learning tasks but often suffer from miscalibration, resulting in unreliable probability outputs. This has inspired many recent works on mitigating miscalibration, particularly through post-hoc recalibration methods that aim to obtain calibrated probabilities without sacrificing the classification performance of pre-trained models. In this study, we summarize and categorize previous works into three general strategies: intuitively designed methods, binning-based methods, and methods based on formulations of ideal calibration. Through theoretical and practical analysis, we highlight ten common limitations in previous approaches. To address these limitations, we propose a probabilistic learning framework for calibration called $h$-calibration, which theoretically constructs an equivalent learning formulation for canonical calibration with boundedness. On this basis, we design a simple yet effective post-hoc calibration algorithm. Our method not only overcomes the ten identified limitations but also achieves markedly better performance than traditional methods, as validated by extensive experiments. We further analyze, both theoretically and experimentally, the relationship and advantages of our learning objective compared to traditional proper scoring rule. In summary, our probabilistic framework derives an approximately equivalent differentiable objective for learning error-bounded calibrated probabilities, elucidating the correspondence and convergence properties of computational statistics with respect to theoretical bounds in canonical calibration. The theoretical effectiveness is verified on standard post-hoc calibration benchmarks by achieving state-of-the-art performance. This research offers valuable reference for learning reliable likelihood in related fields.

Enhancing asymmetric threat predictions via machine learning: insights from North Korea’s trash balloon incidents

North Korea’s trash balloons represent an unconventional asymmetric threat with psychological, environmental, and strategic implications. These balloons, launched under specific meteorological conditions, deliver propaganda materials, hazardous waste, or other threats to South Korea, exploiting natural weather patterns and making them difficult to predict and counteract. This study employs machine learning models, including Random Forest and Extreme Gradient Boosting (XGBoost), to analyze correlations between wind patterns and balloon incidents. Using a data set of meteorological variables and media reports, our analysis identifies key weather conditions conducive to provocations. The proposed models achieve an accuracy of 80% and a recall of 82%, effectively predicting incidents under challenging conditions. The findings highlight the critical role of weather factors in operational planning and demonstrate the value of artificial intelligence (AI)-driven methodologies in addressing low-cost, high-impact threats. By integrating probabilistic and machine learning approaches, this research provides a framework for the identification of pre-emptive threats and resource optimization. Beyond North Korea’s provocations, this methodology has potential applications in broader defense strategies, such as drone incursions and other asymmetric challenges. This study advances the understanding of unconventional threats and underscores the importance of predictive analytics in enhancing defense readiness and resource allocation.

Refined climatologies of future precipitation over High Mountain Asia using probabilistic ensemble learning

Abstract High Mountain Asia holds the largest concentration of frozen water outside the polar regions, serving as a crucial water source for more than 1.9 billion people. Precipitation represents the largest source of uncertainty for future hydrological modelling in this area. In this study, we propose a probabilistic machine learning framework to combine monthly precipitation from 13 regional climate models developed under the Coordinated Regional Downscaling Experiment (CORDEX) over High Mountain Asia via a mixture of experts (MoE). This approach accounts for seasonal and spatial biases within the models, enabling the prediction of more faithful precipitation distributions. The MoE is trained and validated against gridded historical precipitation data, yielding 32\% improvement over an equally-weighted average and 254\% improvement over choosing any single ensemble member. This approach is then used to generate precipitation projections for the near future (2036--2065) and far future (2066--2095) under RCP4.5 and RCP8.5 scenarios. Compared to previous estimates, the MoE projects wetter summers but drier winters over the western Himalayas and Karakoram and wetter winters over the Tibetan Plateau, Hengduan Shan, and South East Tibet.&amp;#xD;

Exploring the Synergy of Self-Supervised Learning and Bayesian Networks for Customer Profiling in the Insurance Industry

The insurance business is increasingly relying on sophisticated machine learning techniques to better assess risk and profile customers. Since conventional approaches suffer from the issues of sparsity in data and a change in the behavior of customers, solutions need to be more flexible and easier to understand. This research tries to see if using SSL and BN together, through sizable, unlabeled datasets, may improve profiling customers in the insurance industry; better risk prediction; and general decision-making. It uses Bayesian networks for probabilistic modeling and dependency learning to extract features from raw, unlabeled data and integrates SSL. The model is evaluated using metrics such as AUC, recall, accuracy, and precision. Better than the typical approaches in machine learning, the proposed hybrid SSL and BN performs with 93% accuracy, 91% performance, and 0.92 AUC, showing excellent performance in handling sparse data while making correct risk evaluation. In the context of the insurance industry, this method ensures an essential, flexible and readable solution to customer profiling from the integration of SSL and BN. Using unlabeled data and probabilistic reasoning for this method further enhances the field of risk management, decisions and customized solutions.

EXPRESS: Domain-specific cognitive flexibility: Shift-readiness adaptations for task- and attention-switching are non-transferrable.

Adaptive behavior in the real world involves navigating competing goals in a constantly changing environment. Doing so requires cognitive flexibility across multiple domains, including flexibility for switching between tasks, i.e., activating the appropriate rules for stimulus-response associations, and flexibility for shifting attention between different sources of sensory inputs. Previous work in task-switching and attention-shifting has separately shown that people are capable of strategically modulating both types of flexibility based on the relevant current demands. That is, people become better at switching between tasks (e.g., digit magnitude versus parity tasks) when switches are frequently cued and better at shifting attention between different stimulus locations when shifts are frequently cued. Across five experiments in the current study, we investigated the possibility of cognitive flexibility transfer between the domains of task switching and attention shifting when the frequency of switches/shifts are orthogonally manipulated within the same context. We implemented a novel paradigm that involved concurrent cued task switching and attentional shifting. We varied either the proportion of task-switches or attention-shifts across blocks of trials, while keeping the proportion of the other constant. If flexibility adaptations in biased contexts transferred across domains, switch/shift frequency manipulations in one domain should affect flexibility across both domains. Instead, in Experiments 1-3, we found that performance costs of attentional shifts remained constant across task-switch biased contexts despite adjustments in task-switch costs; likewise, in Experiments 4-5, we found that costs of switching between tasks remained constant across blocks that varied in attention-shift frequency despite adaptations in attention-shift costs. These results suggest that probabilistic learning and adjustments of attention-shifts and task-switches to meet contextual demands occur independently.

Forecasting Significant Wave Height Intervals Along China’s Coast Based on Hybrid Modal Decomposition and CNN-BiLSTM

As a renewable and clean energy source with abundant reserves, the development of wave energy relies on accurate predictions of significant wave height (Hs). The fluctuation of Hs is a non-stationary process influenced by seasonal variations in marine climate conditions, which poses significant challenges for accurate predictions. This study proposes a deep learning method based on buoy datasets collected from four research locations in China’s offshore waters over three years (2021–2023, 3-hourly). The hybrid modal decomposition CEEMDAN-VMD is employed for reducing non-stationarity of the Hs sequence, with peak information incorporated as a data augmentation strategy to enhance the performance of deep learning. A probabilistic deep learning model, QRCNN-BiLSTM, was developed using quantile regression, achieving 12-, 24-, and 36-h interval predictions of Hs based on 12 days of historical data with three input features (Hs and wave velocities only). Furthermore, an optimization algorithm that integrates the proposed innovative enhancement strategies is used to automatically adjust the network parameters, making the model more lightweight. Results demonstrate that under a 0.95 prediction interval nominal confidence (PINC), the prediction interval coverage probability (PICP) reaches 100% for at least 6 days across all datasets, indicating that the developed system exhibits superior performance in short-term wave forecasting.

Lower reward sensitivity in frontostriatal stroke: Influence of depression and resting-state functional connectivity.

Stroke patients have shown low reward sensitivity, which is a transdiagnostic dimension that defines the extent to which a person actively pursues rewarding stimuli. Low reward sensitivity has been related to depression and dysregulation of the frontostriatal network. To date, studies have addressed this dimension in heterogenic stroke lesions and the underlying mechanisms of frontostriatal stroke patients are still unknown. This study included 54 participants (32 chronic frontostriatal stroke patients and 22 healthy controls). Reward sensitivity was assessed using the probabilistic reversal learning task. Depressive symptoms were measured with the Adult Self-Report, and resting-state functional connectivity (rsFC) was examined using functional near-infrared spectroscopy (fNIRS) in prefrontal, motor, and parietal cortices. Group differences and predictors of reward sensitivity were analyzed using Bayesian ANCOVA and multiple regression models. Stroke patients displayed lower reward sensitivity, higher depressive problems, and lower resting-state functional connectivity between the right orbitrofrontal cortex and the left dorsolateral prefrontal cortex, the right orbitrofrontal cortex and the right dorsolateral prefrontal, and the right dorsolateral prefrontal cortex and right premotor cortex and supplementary motor area. In stroke patients, lower reward sensitivity was predicted by higher depressive problems and lower resting-state functional connectivity between the right dorsolateral prefrontal cortex and the right premotor cortex and the right supplementary motor area. This work showed the relevance of reward sensitivity in frontostriatal post-stroke patients and its relationship with depression, and supports the resting-state functional connectivity measurement for characterizing abnormalities in connectivity in stroke patients.

Orbitofrontal-hippocampal state coding dynamics during reversal learning.

To build an understanding of our world, we make inferences about the connections between our actions, experiences, and the environment. This process, state inference, requires an agent to guess the current state of the world given a set of observations. During value-based decision-making, a growing body of evidence implicates the orbitofrontal cortex (OFC) and the hippocampus (HPC) in the process of contextualizing information and identifying links between stimuli, actions, and outcomes. However, the neural mechanisms driving these processes in primates remain unknown. To investigate how OFC and HPC contribute to state inference, we recorded simultaneously from both regions while two male monkeys (Macaca mulatta) performed a probabilistic reversal learning task, where reward contingencies could be captured by two task states. Using population-level decoding, we found neural representations of task state in both OFC and HPC that remained stable within each trial but strengthened with learning as monkeys adapted to reversals. Subjects also appeared to use their understanding of task structure to anticipate reversals, evidenced by anticipatory neural representations of the upcoming task state.Significance Statement During value-based decision-making, a growing body of evidence implicates the orbitofrontal cortex (OFC) and the hippocampus (HPC) in the process of contextualizing information and identifying links between stimuli, actions, and outcomes. However, limited work has been done in nonhuman primates to bridge the gap between rodent and human models. In this study, we show that task state is represented in both OFC and HPC in nonhuman primates. These representations remain stable within each trial, but evolve with learning and anticipate upcoming task changes, equipping subjects to adapt their choice preferences to reversals in reward contingencies. Our results support the theory that OFC-HPC interactions are important for flexible, goal-directed decision making, and provide insight into how the OFC and HPC participate in decision making when information is not explicitly provided, but must instead be inferred.

Amortized template matching of molecular conformations from cryoelectron microscopy images using simulation-based inference

Characterizing the conformational ensemble of biomolecular systems is key to understand their functions. Cryoelectron microscopy (cryo-EM) captures two-dimensional snapshots of biomolecular ensembles, giving in principle access to thermodynamics. However, these images are very noisy and show projections of the molecule in unknown orientations, making it very difficult to identify the biomolecule's conformation in each individual image. Here, we introduce cryo-EM simulation-based inference (cryoSBI) to infer the conformations of biomolecules and the uncertainties associated with the inference from individual cryo-EM images. CryoSBI builds on simulation-based inference, a merger of physics-based simulations and probabilistic deep learning, allowing us to use Bayesian inference even when likelihoods are too expensive to calculate. We begin with an ensemble of conformations, templates from experiments, and molecular modeling, serving as structural hypotheses. We train a neural network approximating the Bayesian posterior using simulated images from these templates and then use it to accurately infer the conformation of the biomolecule from each experimental image. Training is only done once on simulations, and after that, it takes just a few milliseconds to make inference on an image, making cryoSBI suitable for arbitrarily large datasets and direct analysis on micrographs. CryoSBI eliminates the need to estimate particle pose and imaging parameters, significantly enhancing the computational speed compared to explicit likelihood methods. Importantly, we obtain interpretable machine learning models by integrating physics-based approaches with deep neural networks, ensuring that our results are transparent and reliable. We illustrate and benchmark cryoSBI on synthetic data and showcase its promise on experimental single-particle cryo-EM data.

Intact habit learning in work addiction: Evidence from a probabilistic sequence learning task.

Intact habit learning in work addiction: Evidence from a probabilistic sequence learning task.

Probabilistic quantitative inversion of wheel out-of-roundness using probabilistic learning on manifolds and dynamic model

Probabilistic quantitative inversion of wheel out-of-roundness using probabilistic learning on manifolds and dynamic model

Enhanced risk mapping for hurricane surge losses using synthesized probabilistic and machine learning models

Enhanced risk mapping for hurricane surge losses using synthesized probabilistic and machine learning models

Characterizing Uncertainty in Deep Convection Triggering Using Explainable Machine Learning

Abstract Realistically representing deep atmospheric convection is important for accurate numerical weather and climate simulations. However, parameterizing where and when deep convection occurs (“triggering”) is a well-known source of model uncertainty. Most triggers parameterize convection deterministically, without considering the uncertainty in the convective state as a stochastic process. In this study, we develop a machine learning model, a random forest, that predicts the probability of deep convection, and then apply clustering of Shapley additive explanations (SHAP) values, an explainable machine learning method, to characterize the uncertainty of convective events. The model uses observed large-scale atmospheric variables from the Atmospheric Radiation Measurement constrained variational analysis dataset over the Southern Great Plains, United States. The analysis of feature importance shows which mechanisms driving convection are most important, with large-scale vertical velocity providing the highest predictive power for more certain, or easier to predict, convective events, followed by the dynamic generation rate of dilute convective available potential energy. Predictions of uncertain, or harder to predict, convective events instead rely more on other features such as precipitable water or low-level temperature. The model outperforms conventional convective triggers. This suggests that probabilistic machine learning models can be used as stochastic parameterizations to improve the occurrence of convection in weather and climate models in the future. Significance Statement Convective storms, which produce clouds and precipitation, are difficult to represent in models since they occur at scales smaller than a model grid box. The purpose of this study is to better understand why convection is sometimes easier or harder to predict with certainty. This is important because predicting where and when convection occurs in atmospheric models affects the energy, moisture, and momentum processes in these models, which is known to lead to errors in weather forecasts and climate projections. This work highlights the importance of representing uncertainty in processes like convection.

Brain Activation Associated with Response to Psychotherapies for Late-life Depression: A Task-based fMRI Study

Brain Activation Associated with Response to Psychotherapies for Late-life Depression: A Task-based fMRI Study