Diffusion bridge models have recently become a powerful tool in the field of generative modeling. In this work, we leverage their power to address another important problem in machine learning and information theory, the estimation of the mutual information (MI) between two random variables. Neatly framing MI estimation as a domain transfer problem, we construct an unbiased estimator for data posing difficulties for conventional MI estimators. We showcase the performance of our estimator on three standard MI estimation benchmarks, i.e., low-dimensional, image-based and high MI, and on real-world data, i.e., protein language model embeddings.

The integration of Artificial Intelligence (AI) with satellite remote sensing has fundamentally transformed environmental monitoring from classical physical models to modern deep learning (DL). This review addresses a central question: when, why, and to what extent do AI-driven approaches outperform classical methods in monitoring environmental hazards, and what fundamental limitations persist? We systematically analyze over 200 studies across four critical domains where satellite observations and AI are most effectively combined: wildfires, floods, droughts, and land cover change. Our analysis traces the evolution from index-based and physical models through classical machine learning (ML) to contemporary DL architectures, including convolutional neural networks, recurrent networks (LSTM, ConvLSTM), transformers, and emerging geospatial foundation models. DL demonstrates several important advantages, including automatic feature learning from multispectral imagery, spatial context awareness, temporal modeling of slow-onset and rapid hazards, and integration of heterogeneous satellite data (optical, SAR, LiDAR). However, persistent limitations constrain operational deployment, including data scarcity and domain shift across regions, inadequate uncertainty quantification, physical inconsistency with governing laws, and underutilization of remote sensing data in operational forecasting. Physical boundaries, including chaotic fire behavior, intrinsic predictability limits of meteorological forcing, and spectral saturation in dense forests, impose fundamental constraints regardless of algorithmic sophistication. This review provides a roadmap for next-generation environmental intelligence, identifying hybrid physics-AI modeling, foundation models, and uncertainty-aware architectures as the most promising frontiers. By bridging current capabilities and future potential, the review offers guidance for researchers developing trustworthy and scalable solutions at the intersection of Earth observation and AI. This figure illustrates the conceptual structure for review of DL in environmental hazard monitoring using satellite remote sensing. Data section shows that there are numerous data sources, including satellite-based Earth observation missions (i.e., Sentinel, Landsat), as well as various sensor modalities employed in environmental applications, such as optical, SAR, thermal, and LiDAR Data. Analysis Stage illustrates methodology employed within the review, which includes: literature review, taxonomy of methods, comparison of approaches. Model layer indicates the differences between classical ML algorithms and the DL architectures, including Convolutional Neural Networks, Transformers, and Foundation Models adapted for geospatial Data. The application section lists the primary environmental monitoring tasks examined in the review, including wildfire prediction, flood monitoring, drought assessment, forest monitoring, land cover classification and change detection. The conclusion summarizes the results of the review by providing key insights identifying major challenges, limitations, and future research directions.

Topological Data Analysis (TDA) provides powerful tools to explore the shape and structure of data through topological features such as clusters, loops, and voids. Persistence diagrams are a cornerstone of TDA, capturing the evolution of these features across scales. While effective for analyzing individual manifolds, persistence diagrams do not account for interactions between pairs of them. Cross-persistence diagrams (cross-barcodes), introduced recently, address this limitation by characterizing relationships between topological features of two point clouds. In this work, we present the first systematic study of the density of cross-persistence diagrams. We prove its existence, establish theoretical foundations for its statistical use, and design the first machine learning framework for predicting cross-persistence density directly from point cloud coordinates and distance matrices. Our statistical approach enables the distinction of point clouds sampled from different manifolds by leveraging the linear characteristics of cross-persistence diagrams. Interestingly, we find that introducing noise can enhance our ability to distinguish point clouds, uncovering its novel utility in TDA applications.We demonstrate the effectiveness of our methods through experiments on diverse datasets, where our approach consistently outperforms existing techniques in density prediction and achieves superior results in point cloud distinction tasks. Our findings contribute to a broader understanding of cross-persistence diagrams and open new avenues for their application in data analysis, including potential insights into time-series domain tasks and the geometry of AI-generated texts. Our code is publicly available at https://github.com/Verdangeta/TDA_experiments.

Artificial intelligence (AI) has emerged as a transformative force in production technology (PT) within the oil and gas industry, driving advancements in prediction, optimization, and real-time decision-making. This review systematically examines the evolution, current applications, opportunities and challenges of AI techniques in PT. It considers a wider range of relevant topics than previous reviews, providing an historical perspective to the current state of AI approaches applied to or proposed for PT. Key PT applications of AI include predicting physical properties of produced oil, gas, condensate, and natural gas hydrates. The performance and characterization of production wells are analyzed, focusing on flowrate predictions, well placement, reservoir connectivity, pressure gradient estimation, and petrophysical and geomechanical characterization. Optimization techniques for production well pumping and the prediction of operational challenges are discussed. The review highlights the role of AI techniques in real-time production optimization and intelligent workflow management, emphasizing its integration in operating company workflows. Hybrid AI techniques, incorporating optimization algorithms and committee machine approaches, are evaluated for their effectiveness in enhancing PT workflows. Economic considerations regarding the adoption of AI techniques contribute to guidelines proposed for selecting and evaluating AI techniques tailored to PT-related issues. A summary of key findings, challenges, and opportunities for future developments, provides a comprehensive framework for leveraging AI to advance PT in the oil and gas sector.

The integration of artificial intelligence (AI) into reservoir simulation (RS) has rapidly evolved, offering new solutions to long-standing challenges in predicting subsurface behavior. Although recent advances have improved the accuracy of fluid-flow modeling and reservoir performance forecasting, issues related to computational cost, data complexity, and model scalability persist. This review, therefore, explores the transformative role of AI in enhancing RS, with a particular focus on three core areas: history matching (HM), production optimization, and computational acceleration. To achieve this, the study systematically analyzes AI-driven approaches, including the historical development of RS and its multiple challenges with data obstacles. These include data volume and quality, input variable prioritization by feature selection, and objective function refinement. Reservoir property assessment across clastic and carbonate formations, alongside AI-enabled strategies for optimizing well placement and well production and injection planning pose additional challenges. The review finds that AI significantly enhances prediction accuracy and reduces simulation time while offering scalable solutions to big-data challenges in RS workflows, leading to better decision-making. Despite these advancements, obstacles remain in the real field deployment of AI models, the interpretation of their outputs results, and the assurance of consistent performance across various reservoirs. This highlights the necessity for hybrid physics and AI methodologies, in conjunction with automated, high-performance computing workflows, and learning from real field applications to address these deficiencies. This review provides a structured perspective on leveraging AI to advance reservoir engineering practices, enhance predictive capabilities, and maximize hydrocarbon recovery in an economically and operationally viable manner. It also identifies key challenges and research gaps and offers recommendations for future directions.

While achieving exceptional generative quality, modern diffusion, flow, and other matching models suffer from slow inference, as they require many steps of iterative generation. Recent distillation methods address this by training efficient one-step generators under the guidance of a pre-trained teacher model. However, these methods are often constrained to only one specific framework, e.g., only to diffusion or only to flow models. Furthermore, these methods are naturally data-free, and to benefit from the usage of real data, it is required to use an additional complex adversarial training with an extra discriminator model. In this paper, we present \textbf{RealUID}, a unified distillation framework for all matching models that seamlessly incorporates real data into the distillation procedure without GANs. Our \textbf{RealUID} approach offers a simple theoretical foundation that covers previous distillation methods for Flow Matching and Diffusion models, and is also extended to their modifications, such as Bridge Matching and Stochastic Interpolants.

We introduce PersonalAI 2.0 (PAI-2), a novel framework, designed to enhance large language model (LLM) based systems through integration of external knowledge graphs (KG). The proposed approach addresses key limitations of existing Graph Retrieval-Augmented Generation (GraphRAG) methods by incorporating a dynamic, multistage query processing pipeline. The central point of PAI-2 design is its ability to perform adaptive, iterative information search, guided by extracted entities, matched graph vertices and generated clue-queries. Conducted evaluation over six benchmarks (Natural Questions, TriviaQA, HotpotQA, 2WikiMultihopQA, MuSiQue and DiaASQ) demonstrates improvement in factual correctness of generating answers compared to analogues methods (LightRAG, RAPTOR, and HippoRAG 2). PAI-2 achieves 4

Nowadays, the creation of a process flow diagram (PFD) and its subsequent transformation into a piping and instrumentation diagram (P ID) is predominantly performed manually. Applying artificial intelligence in the task could potentially lead not only to process automation and time savings, but also to financial gains by exploring numerous diagram's topology options and reducing manual labor. This research presents P ID Pilot - a practical end-to-end AI pipeline capable of handling flowsheet developing for both stages. The first stage focuses on PFD synthesis, whereas the second is directed toward modifying the generated PFD into P ID. After comparing four different methods, the hybrid approach combining genetic algorithms (GA) and large language models (LLM) is shown to generate the optimal valid PFD topology, achieving the lowest loss value among all the methods, while satisfying the required outlet flow parameters without engineering-rule violations. For the second stage, the proposed LLM-based agent successfully transforms the generated PFD into a source-grounded P ID by producing validated, executable modifications through a restricted engineering software development kit, achieving 100

Personalizing language models by effectively incorporating user interaction history remains a central challenge in the development of adaptive AI systems. While large language models (LLMs), combined with Retrieval-Augmented Generation (RAG), have improved factual accuracy, they often lack structured memory and fail to scale in complex, long-term interactions. To address this, we propose a flexible external memory framework based on a knowledge graph that is constructed and updated automatically by the LLM. Building upon the AriGraph architecture, we introduce a novel hybrid graph design that supports both standard edges and two types of hyper-edges, enabling rich and dynamic semantic and temporal representations. Our framework also supports diverse retrieval mechanisms, including A*, WaterCircles traversal, beam search, and hybrid methods, making it adaptable to different datasets and LLM capacities. We evaluate our system on TriviaQA, HotpotQA, DiaASQ benchmarks and demonstrate that different memory and retrieval configurations yield optimal performance depending on the task. Additionally, we extend the DiaASQ benchmark with temporal annotations and internally contradictory statements, showing that our system remains robust and effective in managing temporal dependencies and context-aware reasoning.

Managed forest lands are key contributors to the carbon balance assessment needed for the greenhouse gas inventories on local, regional, national, and global levels. However, forest lands, due to size and complexity, are challenging for detailed spatially-explicit monitoring and, therefore, reliable and automatic assessment of spatial-temporal changes of carbon stocks is limited. This study presents an effective methodology for estimating key forest structure characteristics relevant to sequestration potential by combining management-level inventory data with remote sensing covariates. It primarily focuses on a machine learning (ML) pipeline that integrates an uncertainty quantification stage to support reliable decision-making and environmental analysis. We evaluated three algorithms - Random Forest (RF), Extreme Gradient Boosting (XGBoost), and TabNet-applied for multispectral satellite measurements. Predictions were made at the stand level inventory data, addressing classification tasks for dominant species and age group, and regression tasks for timber stock, stand height, and average basal area. Forest carbon stock was also derived as a target variable. XGBoost achieved the best overall performance across tasks. For regression, it reached mean absolute percentage error (MAPE) equal to 0.18 for height; 0.24 for basal area; 0.47 for timber stock and 0.37 for carbon stock. The coefficient of determination ([Formula: see text]) of 0.68 across all regression tasks. For classification, XGBoost achieved an average F1-score of 0.70 for age group prediction and 0.83 for dominant species prediction. To address the 'black-box' nature of machine learning models and enhance interpretability, we incorporated a refinement of conformal prediction to quantify predictive uncertainty at a nominal 90% coverage level. As a result, a geospatial mapping tool was developed, enabling the generation of stand-level forest attributes at 10 m spatial resolution, together with corresponding uncertainty estimates, supporting more informed forest management and carbon accounting.

Precise and reliable predictions of filtrate loss (FL) from drilling fluids/muds under high-pressure-high-temperature (HPHT) conditions are required to prevent formation damage, maintain wellbore stability, and optimize drilling efficiency in complex reservoirs. Traditional laboratory-based measurements are too time consuming to provide HPHT FL values that are usable for decision-making while drilling. This study, therefore, develops machine learning (ML) models that enable frequent and prompt predictions of HPHT FL with high accuracy and reliable precision. To this end, a large dataset was compiled containing five input parameters, namely mud temperature (MT), mud weight (MW), funnel viscosity (FV), mud alkalinity (MA), and electrical stability (ES), and a single target variable HPHT FL. Following meticulous data preprocessing, four predictive models were developed using established ML algorithms: multilayer perceptron neural network (MLPNN), support vector regression (SVR), extreme gradient boosting (XGBoost), and extreme learning machine (ELM). For each ML algorithm, five separate model instances were developed and evaluated, with XGBoost providing the best performance in predicting the target parameter (root mean square error (RMSE) = 0.096 cc/30 min for the testing subset). The superior performance of the XGBoost model was further confirmed by residual, uncertainty, overfitting, and robustness evaluations, indicating its excellent generalization capability. Shapley additive explanations (SHAP) analysis identified ES as the most influential input parameter and FV as the least influential for XGBoost’s HPHT FL predictions. The XGBoost model as developed offers a substantial improvement compared to laboratory FL analysis, enabling more efficient and quicker monitoring of FL in challenging HPHT downhole environments.

Learning conditional distributions π^*(·|x) is a central problem in machine learning, which is typically approached via supervised methods with paired data (x,y) ∼ π^*. However, acquiring paired data samples is often challenging, especially in problems such as domain translation. This necessitates the development of semi-supervised models that utilize both limited paired data and additional unpaired i.i.d. samples x ∼ π^*_x and y ∼ π^*_y from the marginal distributions. The usage of such combined data is complex and often relies on heuristic approaches. To tackle this issue, we propose a new learning paradigm that integrates both paired and unpaired data seamlessly using the data likelihood maximization techniques. We demonstrate that our approach also connects intriguingly with inverse entropic optimal transport (OT). This finding allows us to apply recent advances in computational OT to establish an end-to-end learning algorithm to get π^*(·|x). In addition, we derive the universal approximation property, demonstrating that our approach can theoretically recover true conditional distributions with arbitrarily small error. Furthermore, we demonstrate through empirical tests that our method effectively learns conditional distributions using paired and unpaired data simultaneously.

Large Language Models (LLMs) are powerful tools for modern applications, but their computational demands limit accessibility. Quantization offers efficiency gains, yet its impact on safety and trustworthiness remains poorly understood. To address this, we introduce OpenMiniSafety, a human-curated safety dataset with 1.067 challenging questions to rigorously evaluate model behavior. We publicly release human safety evaluations for four LLMs (both quantized and full-precision), totaling 4.268 annotated question-answer pairs. By assessing 66 quantized variants of these models using four post-training quantization (PTQ) and two quantization-aware training (QAT) methods across four safety benchmarks including human-centric evaluations we uncover critical safety performance trade-offs. Our results show both PTQ and QAT can degrade safety alignment, with QAT techniques like QLORA or STE performing less safely. No single method consistently outperforms others across benchmarks, precision settings, or models, highlighting the need for safety-aware compression strategies. Furthermore, precision-specialized methods (e.g., QUIK and AWQ for 4-bit, AQLM and Q-PET for 2-bit) excel at their target precision, meaning that these methods are not better at compressing but rather different approaches.

High-resolution regional numerical weather prediction (NWP) models are essential tools for capturing atmospheric dynamics, including critical mesoscale weather phenomena, yet they often suffer from systematic biases caused by errors in initial conditions, parameterization schemes, and numerical integration. One established way to address these errors is statistical bias correction, which adjusts model outputs toward reference datasets to reduce systematic deviations. Such methods commonly rely on observational datasets that can be sparse in space and time or relatively coarse. However, naively tuning forecasts to these datasets may provoke loss of vital small-scale features present in the original high-resolution predictions. To address this, we propose a statistical bias-correction method designed to improve atmospheric forecast accuracy while preserving the small-scale dynamics of the original predictions. We introduce a novel loss-function term that minimizes the difference between small-scale features in the original and corrected forecasts. A convolutional U-net with a Transformer in the latent space is trained on European Centre for Medium-Range Weather Forecasts (ECMWF) Reanalysis Version 5 (ERA5) reanalysis data, along with meteorological station observations and scatterometer measurements over the Kara and Barents Seas. The proposed Bidirectional Encoder Representations from Transformers and U-net model achieves accuracy comparable with state-of-the-art neural network methods, while offering superior performance in perceptual metrics describing mesoscale dynamics. The method shows season-dependent skill, with the strongest improvements observed during the sea-ice season. Incorporating specific observational datasets during training leads to consistent error reductions when evaluated against the corresponding sources. The proposed approach is readily applicable in operational weather forecasting.

The design of modern neural architectures has converged through incremental empirical choices, yet the mechanisms governing their training dynamics remain only partially understood. We identify and analyze a negative weight drift induced by the interaction between standard losses and positively biased activation functions. We prove that under MSE or cross-entropy loss, the gradient with respect to positive pre-activations is non-negative in expectation at initialization, driving downstream weights toward negative values during early training. The drift is intrinsic to optimization rather than data, and persists across architectures (MLP, ResNet, ViT, GPT-nano, MP-SENe) and asymmetric activation functions (ReLU, GELU, SiLU). Coupled with ReLU, weight drift produces activation sparsity reaching up to 90% in GPT-nano. We characterize the sparsity-accuracy tradeoff across 79 configurations and identify a sharp accuracy cliff above ∼70% activation sparsity. While ReLU^2 achieves a good sparsity–accuracy ratio in GPT-nano, it pathologically amplifies identified activation spikes in intermediate transformer layers. Clipping resolves this while preserving the representational benefits of squaring: clipped ReLU^2 outperforms its unclipped version, and GELU^2 achieves the lowest validation loss on GPT-nano. Code is available at https://github.com/On-Point-RND/BugOrFeature.

The Iterative Markovian Fitting (IMF) procedure, which iteratively projects onto the space of Markov processes and the reciprocal class, successfully solves the Schrödinger Bridge (SB) problem. However, an efficient practical implementation requires a heuristic modification - alternating between fitting forward and backward time diffusion at each iteration. This modification is crucial for stabilizing training and achieving reliable results in applications such as unpaired domain translation. Our work reveals a close connection between the modified version of IMF and the Iterative Proportional Fitting (IPF) procedure - a foundational method for the SB problem, also known as Sinkhorn's algorithm. Specifically, we demonstrate that the heuristic modification of the IMF effectively integrates both IMF and IPF procedures. We refer to this combined approach as the Iterative Proportional Markovian Fitting (IPMF) procedure. Through theoretical and empirical analysis, we establish the convergence of IPMF procedure under various settings, contributing to developing a unified framework for solving SB problems. Moreover, from a practical standpoint, the IPMF procedure enables a flexible trade-off between image similarity and generation quality, offering a new mechanism for tailoring models to specific tasks.

Efficient simulations of complex physical systems described by partial differential equations (PDEs) require computational methods that can reduce the resource demands without sacrificing the accuracy. Here, we introduce a framework based on graph neural networks for learnable self-supervised differentiable coarsening of unstructured computational grids. We leverage graph-based representation of the physical system and offer a computational grid coarsening method which preserves the underlying physical properties together with the stability of the chosen numerical scheme with the help of the designed loss terms. The coarsening model is trained in a self-supervised way by minimizing the error between the output of the simulations on the coarsened and original grids. We evaluate the approach on four PDE systems of different types, covering both linear and nonlinear regimes, including subsurface flow and wave propagation. We demonstrate that the proposed framework exhibits the ability to maintain high fidelity in simulation outputs even after 95% reduction in the number of nodes, significantly reducing computational overhead. We show that the model exhibits generalizability to unseen scenarios, thereby outperforming the baselines. The generality of the developed framework is also proven by its possibility to adapt to implicit numerical schemes used to model stiff systems of PDEs. Thus, the developed approach demonstrates the ability to accelerate physical simulations without compromising accuracy.

The complexity and sustainability demands of geoenergy science (GS) systems necessitate predictive models that are both accurate and computationally efficient. Traditional numerical simulators, while physically rigorous, are often prohibitively expensive, whereas purely data-driven machine learning models lack interpretability and physical fidelity, especially under data scarcity. Physics-informed machine learning (PIML) bridges this gap by integrating deep learning with governing physical laws to deliver reliable, efficient, and interpretable predictions. This review provides a comprehensive analysis of PIML applications in GS, with emphasis on reservoir modeling, subsurface multiphase flow, thermal dynamics, and inter-well connectivity estimation. It critically evaluates core PIML paradigms, including physics-informed neural networks (PINNs), neural operators (NO), and physics-guided machine learning (PGML), across key domains: underground gas storage (natural gas, COQ, and HQ), geothermal energy systems, and real-time reservoir optimization. The review highlights how PIML embeds conservation laws (e.g., mass, energy) to solve both forward and inverse problems, and assesses advances in uncertainty quantification, transfer learning, and multiscale hierarchical modeling. Real-world case studies and bibliometric trends illustrate practical deployment and emerging methodological directions. PIML offers the potential to provide robust, generalizable frameworks that enhance prediction performance, ensure physical consistency, and substantially reduce computational cost, even with sparse or noisy field data. The principal outcome of this review is that PIML enables generalizable, physics-compliant surrogate models that outperform conventional approaches. However, future work is required to prioritize scalable architectures, resilience to data noise, and tighter integration with downhole sensing to support autonomous, physics-guided decision-making in the next-generation of GS operations.

Entropic optimal transport (EOT) in continuous spaces with quadratic cost is a classical tool for solving the domain translation problem. In practice, recent approaches optimize a weak dual EOT objective depending on a single potential, but doing so is computationally not efficient due to the intractable log-partition term. Existing methods typically resolve this obstacle in one of two ways: by significantly restricting the transport family to obtain closed-form normalization (via Gaussian-mixture parameterizations), or by using general neural parameterizations that require simulation-based training procedures. We propose Variational Entropic Optimal Transport (VarEOT), based on an exact variational reformulation of the log-partition log𝔼[exp(·)] as a tractable minimization over an auxiliary positive normalizer. This yields a differentiable learning objective optimized with stochastic gradients and avoids the necessity of MCMC simulations during the training. We provide theoretical guarantees, including finite-sample generalization bounds and approximation results under universal function approximation. Experiments on synthetic data and unpaired image-to-image translation demonstrate competitive or improved translation quality, while comparisons within the solvers that use the same weak dual EOT objective support the benefit of the proposed optimization principle.

As AI-code assistant tools become widespread, automatic assessment of the correctness of generated code becomes a significant challenge. Code LLMs are prone to hallucinations, which may lead to code that does not solve the required problem, or even to code with severe security vulnerabilities. In this paper, we introduce CodeTD – the first approach to pre-execution assessment of code correctness based on topological data analysis (TDA) of Code LLMs' attention maps. Our method quantifies prompt-generation mismatch using topological patterns of attention maps. We carry out experiments with common benchmarks (HumanEval, MBPP, BigCodeBench, MultiPL-E), 5 programming languages and 10 Code LLMs of size up to 34B parameters. The experimental results show that the proposed method outperforms recent baselines. Moreover, CodeTD is transferable between coding benchmarks.