The family Turdidae (Aves, Passeriformes) is a diverse clade of passerine birds that presents a global distribution across a wide range of environments. While numerous studies have addressed the biogeography of specific genera, the family's origins remain largely unresolved. To reconstruct this taxon's biogeographic history, we compiled genetic and geographic range data from public sources. This information was used to build a time-calibrated phylogeny based on a concatenated dataset of mitochondrial and nuclear genes for 155 species across 16 genera. Ancestral area reconstructions were performed using PhyGeo, a geographically explicit, diffusion-based method that integrates species distributions with a dynamic palaeogeographic model applied to a pixelated spherical model of the Earth. Our analysis produced contrasting results depending on model assumptions. Under the Global Model, which treats all land features as equally habitable, Turdidae likely originated in West Antarctica-South America, with later colonization of Africa, Australia, and Asia via an Antarctic route. In contrast, the Restricted Model, which excludes unglaciated Antarctic land, shifted the inferred origin to northern North America, placing ancestral ranges closer to extant distributions and more consistent with current evidence. These results underscore the importance of considering multiple lines of evidence and spatially explicit approaches when reconstructing biogeographic histories.

BackgroundSparse-view CT reduces radiation dose by decreasing the number of projections, yet the resulting undersampling introduces severe artifacts in images reconstructed with traditional analytical algorithms. Recent diffusion posterior sampling (DPS)-based methods enhance image quality but frequently generate spurious details and incur prohibitive computational cost, limiting clinical adoption.ObjectiveTo enable high-fidelity, low-dose CT imaging from sparse projections while suppressing hallucinated details and reducing computational burden.MethodsWe propose a novel diffusion-based method that synergizes null-space restoration with Filtered Back-Projection (FBP) pseudoinverse approximation. Specifically, by employing range-null space decomposition, we use diffusion models to restore null-space image components while ensuring data consistency through the FBP algorithm approximating the pseudoinverse of projection matrix in the range image space. Moreover, we provide a theoretical analysis of the rationality of this approximation. This novel approach effectively combines the strengths of diffusion models and traditional CT reconstruction techniques, optimizing the inverse diffusion trajectory to enable high-fidelity image recovery from sparse data.ResultsExperimental results show that the proposed method achieves significant improvements in image quality and computational efficiency. Compared with the DPS method, it yields an average PSNR gain of 5.32 dB, an average SSIM increase of 0.083, and a 41.9% reduction in computation time.ConclusionIn summary, this framework provides a practical and effective solution for high-quality, low-dose CT imaging, effectively balancing reconstruction accuracy and computational efficiency in practical applications.

Projector photometric compensation corrects color distortions introduced by surface texture, reflection, and ambient lighting. Existing deep learning-based methods usually require professional scene-specific data collection and lack consideration for perceptual quality. To address this limitation, we present a diffusion-based photometric compensation method that reconstructs compensation images under photometric and content-aware guidance. Specifically, we fi rst mo del th e ph otometric distortions introduced during projection as environment-dependent additive noise, thereby reformulating the photometric compensation problem as a denoising task with physical constraints. Next, we introduce a diffusion model, which generates compensation images by following an additive trajectory to iteratively remove the noise. Finally, to accurately estimate the noise at each timestep, by analyzing the factors that contribute to distortions in the physical process of projection and capturing, we design a noise estimation network that incorporates features of both photometry-aware and content conditions. Experiments show that our method achieves superior visual performance in unknown scenarios, thereby exhibiting significant practical advantages over prior art. Our source code is available at https://github.com/cyxwang/DiffPC.

To overcome the high computational expense of conventional quantum chemistry techniques and the limited incorporation of physical constraints in machine learning models, we present SphereDiff-TS: a diffusion-based method for predicting 3D transition state (TS) structures using a spherical coordinate system with flexible boundary and dynamic radius constraints. Evaluated against true transition states, the model achieves chemical accuracy in both geometry (median RMSD: 0.048 Å; median of 0.022 Å on selected cross-system cases) and energy (median absolute error: 0.55 kcal/mol; 0.328 kcal/mol on the same cases). Moreover, comparative analysis with the literature-reported structures confirms that the model accurately reproduces barrier heights, with deviations generally below 1.5 kcal/mol. These results highlight the potential of SphereDiff-TS as a robust computational tool for exploring reaction mechanisms and aiding in computer-driven reaction design.

From missing data to meaningful rankings: Multi-criteria performance evaluation of six global iron ore producers.

The rapid evolution of generative and diffusion-based method in artificial intelligence (AI) has transformed de novo peptide and protein design, enabling the creation of sequence–structure–function relationships beyond conventional heuristic exploration. In this work, we developed an AI-guided molecular design pipeline that integrates diffusion-based generative modeling, inverse folding algorithms, and atomistic simulation to design and evaluate novel peptide inhibitors of the LasR quorum-sensing receptor of Pseudomonas aeruginosa. Peptides were generated using multiple complementary AI frameworks, including DiffPepBuilder (a diffusion-based denoising generator), ProteinMPNN (an inverse folding network), RFdiffusion2, Evobind2, and BindCraft (a hybrid sequence–structure generative model). These peptides were then subjected to physicochemical profiling and molecular docking to estimate binding affinities (Kd) and interface complementarity. The top candidates were subjected to 400 ns all-atom molecular dynamics (MD) simulations, performed in triplicate, to assess stability, flexibility, and interaction persistence, and were compared against the experimentally validated peptides benchmark. Post-simulation metrics, including RMSD, RMSF, and radius of gyration (Rg), revealed reproducible but architecture-specific dynamic behaviors, with BindCraft 4 maintaining the most compact and stable configuration, exhibiting minimal fluctuations and consistent binding-site retention. MMGBSA, performed at different time intervals, revealed strong van der Waals and electrostatic stabilization, correlating with sub-micromolar dissociation constants predicted computationally. In contrast, DiffPepBuilder 1–2 exhibited balanced flexibility and moderate stabilization, while ProteinMPNN 2 and the control (Aqs1C) displayed comparatively weaker energetic convergence. Principal Component Analysis (PCA) and Free Energy Landscape (FEL) projections confirmed that BindCraft 4 sampled a single deep energy basin, signifying a thermodynamically optimized conformational state. Collectively, this integrative AI-simulation hybrid computational framework demonstrates the capability of generative diffusion and inverse folding models to design peptides with superior structural stability and binding thermodynamics, offering a rational framework for AI-driven peptide therapeutics targeting Pseudomonas aeruginosa communication networks.

Generating realistic, subject-specific eye movement signals is crucial for data augmentation, gaze-based authentication, privacy-preserving analytics, and robust gaze-based interfaces. Recent advances in deep learning have enabled the generation of synthetic gaze data, but individualized gaze sequence generation has been less explored, with most approaches relying on random noise distributions or predefined latent embeddings. In this work, our main contribution is the introduction of explicit subject-aware modifications to both diffusion and generative adversarial networks (GANs)-based methods for generating realistic, individualized synthetic gaze data. In particular, we equip the diffusion-based method with compact user embeddings to capture per-subject traits and enhance the GAN-based generator with a subject-specific synthesis module to better retain idiosyncratic gaze information. We comprehensively assess these subject-aware methods with standard eye-tracking signal quality metrics, including spatial accuracy and precision. This work defines synthetic signal quality, realism, and subject specificity, and advances the potential of subject-aware gaze-based applications.

Monitoring and understandingthe aggregation kinetics of n-typepolymers provide strategies to favorably control aggregation and therebyoptimize conjugated polymers ink shelf life, printability, and thin-filmproperties. Here, the in situ characterization of n-type copolymeraggregates employing ensemble absorbance and fluorescence spectroscopyfor spectral characterization, in combination with single-moleculefluorescence spectroscopy methods to identify subcategories of aggregates,is reported. Specifically, we utilize a diffusion-based single-moleculeburst method that resolves individual aggregates as they traversethe observation volume, allowing us to determine the aggregate size,concentration, and chain conformation through the statistical analysisof single-aggregate fluorescence data. Base-stable P­(EO-NDIT2) withbranched ether-based side chains self-assembles from molecularly dissolvedchains into nano- to micrometer-sized aggregates over the course ofweeks to months, depending on the solvent used. Through spectral decompositionand polarization-sensitive single-molecule fluorescence spectroscopy,the aggregates were categorized by their size into small (Rh approximately 60 nm) and large (Rh approximately 300 nm) aggregates and monitored withtime. An increase in size was correlated with enhanced fluorescencebrightness and red-shifted emission as well as an increase in internalorder, as revealed by emission anisotropy. This increase in orderwithin the aggregates may be related to alignment of crystalline domainsand a planarization of the polymer backbone torsion.

Modelling human gaze behaviour on 360 \({}^{\circ}\) images is important for various human–computer interaction applications. However, existing methods are limited to predicting discrete fixation sequences or aggregated saliency maps, thereby neglecting fine-grained gaze behaviour such as saccadic eye movements that can be captured by commercial eye-trackers. We introduce a more challenging task— fine-grained gaze sequence generation . This task aims to generate eye-tracker-like gaze data for given stimuli. We propose DiffGaze , a diffusion-based method for generating realistic and diverse fine-grained human gaze sequences conditioned on 360 \({}^{\circ}\) images. We evaluate DiffGaze on two 360 \({}^{\circ}\) image benchmarks for fine-grained gaze sequence generation as well as two downstream tasks, scanpath prediction and saliency prediction. Our evaluations show that DiffGaze outperforms the fine-grained gaze generation baselines in all tasks on both benchmarks. We also report a 21-participant survey study showing that our method generates gaze sequences that are indistinguishable from real human sequences. Taken together, our evaluations not only demonstrate the effectiveness of DiffGaze but also point towards a new generation of methods that faithfully model the rich spatial and temporal nature of natural human gaze behaviour.

STF-FlowDiff: A Diffusion-Based SpatioTemporal Fusion Method with Optical Flow Guidance

ConDNS: A novel conditional diffusion-based negative sampling method for knowledge graph embedding

Mobile robotic machining systems (MRMS) often suffer from limited overall motion accuracy due to the accumulation and coupling of pose errors among multi-units during operation. To address this issue, this study analyzes the motion transmission chain of MRMS, and decouples the geometric parameter errors and joint compliance errors using the nonsingular line representation method. Subsequently, the linear propagation-decoupling equations for generalized kinematic errors are derived, revealing the transmission mechanisms of different types of errors among various units across coordinate frames. On this basis, an information diffusion-based dataset augmentation method is developed to address the problem of limited sample data, and it is integrated with gradient boosting decision trees to predict the residual uncertainty errors. Furthermore, a hierarchical calibration method is proposed to progressively calibrate joint compliance errors, generalized geometric parameter errors, and residual uncertainty errors of MRMS. The effectiveness and superiority of the proposed method are validated through experiments.

Cardiovascular diseases have long represented one of the most significant global health challenges. Due to its non-contact and non-invasive characteristics, the ballistocardiogram (BCG) shows considerable promise for various applications, particularly in heart rate variability (HRV) monitoring. Accurate J-peak detection is a critical prerequisite for effective BCG-based analysis. However, BCG signal quality is affected by multiple factors—including sensor type, measurement posture, and individual differences—posing challenges to reliable J-peak detection. In this paper, a diffusion-based J-peak detection method, named J-Diff model, is proposed. To capture the underlying data distribution and address uncertainties in feature variations, a discriminative model termed the conditional detector is employed to generate initial segmentation results. These results are subsequently refined by a conditional diffusion probabilistic model (CDDPM)-based diffusion refiner that models the data distribution across varying noise levels, thereby yielding more accurate and robust segmentation outcomes. Specifically, the conditional cross transformer (CCT) is introduced to integrate features from the condition detector and the diffusion refiner, thereby reducing the domain gap and directing the network’s attention toward J-peaks. The experimental evaluations were conducted on two BCG datasets collected under different postures and using different sensor types, achieving average Precision of 88.02% and 97.32%, Recall of 86.79% and 97.10%, and F1-scores of 0.874 and 0.972, respectively. Correlation analysis revealed significantly high correlations between all HRV indices derived from BCG and those obtained from ECG. These results highlight the effectiveness of applying the proposed J-Diff model in improving the reliability and generalization of BCG measurement systems.

FA-Seg: A fast and accurate diffusion-based method for open-vocabulary segmentation

SuperDiff: A diffusion super-resolution method for digital pathology with comprehensive quality assessment.

Fetal electrocardiography (FECG) is a crucial tool for assessing fetal cardiac health and pregnancy status. Direct invasive FECG provides reliable fetal heart rate signals, but poses risks and is limited to use during labor. Conversely, non-invasive monitoring of the fetal heart is possible via abdominal electrocardiography (AECG), which detects fetal heart waveforms using electrodes positioned on the mother’s abdomen. However, this method is often subject to interference from maternal cardiac activity and other external sources. To address this issue, we propose a novel diffusion method, DIFF-FECG, aimed at improving the extraction of FECG signals from AECG recordings. This method leverages a condition-driven diffusion process to learn specific conditional probability distributions, enabling the effective separation of high-quality FECG signals from noisy AECG data. By adaptively managing the inherent non-Gaussian noise characteristics of MECG within the AECG, DIFF-FECG achieves more effective FECG reconstruction. Furthermore, the quality of the generated FECG signals is also enhanced by adding reconstruction loss and multiple reconstructions. Experimental results on two public databases demonstrate that the proposed DIFF-FECG method yields satisfactory results, with an average Pearson correlation coefficient of 0.922 for the estimated FECG. These findings underscore the potential of diffusion probabilistic models in advancing FECG signal extraction techniques, thereby contributing to improved fetal health monitoring.

Generating temporal data under conditions is crucial for forecasting, imputation, and generative tasks. Such data often has metadata and partially observed signals that jointly influence the generated values. However, existing methods face three key limitations: (1) they condition on either the metadata or observed values, but rarely both together; (2) they adopt either training-time approaches that fail to generalise to unseen scenarios, or inference-time approaches that ignore metadata; and (3) they suffer from trade-offs between generation speed and temporal coherence across time windows, choosing either slow but coherent autoregressive methods or fast but incoherent parallel ones. We propose WaveStitch, a novel diffusion-based method to overcome these hurdles through: (1) dual-sourced conditioning on both metadata and partially observed signals; (2) a hybrid training-inference architecture, incorporating metadata during training and observations at inference via gradient-based guidance; and (3) a novel pipeline-style paradigm that generates time windows in parallel while preserving coherence through an inference-time conditional loss and a stitching mechanism. Across diverse datasets, WaveStitch demonstrates adaptability to arbitrary patterns of observed signals, achieving 1.81x lower mean-squared-error compared to the state-of-the-art, and generates data up to 166.48x faster than autoregressive methods while maintaining coherence. Our code is available at: https://github.com/adis98/WaveStitch.

A diffusion-based machine learning method for 3D architectural form-finding

With growing concerns about data privacy, differential privacy (DP) has become a central technique in privacy-preserving data sharing and machine learning. While many DP generative models have been developed for continuous data such as images, generating synthetic tabular dataespecially with mixed data typesremains a challenging task. In this work, we present TableDiffusion, the first diffusion-based generative model specifically designed for differentially private tabular data synthesis. TableDiffusion introduces a hybrid data generation framework capable of handling both continuous and categorical features. It uses a cosine noise scheduler to control the diffusion process and adopts a composite loss function that combines mean squared error (MSE) and Kullback-Leibler (KL) divergence to better adapt to different data types. The model is implemented using the Opacus library to provide formal DP guarantees, and privacy accounting is performed through the Rnyi Differential Privacy (RDP) mechanism. To assess the effectiveness of TableDiffusion, we implemented a set of benchmark DP generative models, including attention-based methods such as DPattentionVAE and DPattentionGAN, as well as classical approaches such as PATE-GAN and DP-WGAN. Experiments are conducted on a real-world tabular dataset in the cybersecurity domain, containing more than 50,000 samples and 45 features. We evaluated model performance using metrics such as marginal distribution similarity, PMSE ratio, and Alpha-Beta membership inference scores, focusing on both data utility and privacy preservation. The experimental results are currently being analyzed. Preliminary findings suggest that TableDiffusion offers promising performance in a range of privacy budgets and shows advantages in training stability and diversity over GAN-based methods. Our main contributions include: (1) proposing the first diffusion-based method for differentially private tabular data generation; (2) designing a unified architecture for mixed-type feature synthesis; and (3) providing a comprehensive implementation and evaluation framework to facilitate further research.

Diffusion-based adversarial attack method against person re-identification