ARTIFICIAL INTELLIGENCE, MACHINE LEARNING & DATA SCIENCE

Abstract: Estimating H I Mass Fraction in Galaxies with Bayesian Neural Networks

Neutral atomic hydrogen (H I) regulates galaxy growth and quenching, but direct 21 cm measurements remain observationally expensive and affected by selection biases. We develop Bayesian neural networks (BNNs)—a type of neural model that returns both a prediction and an associated uncertainty—to infer the H I mass, log10(MHI), from widely available optical properties (e.g., stellar mass, apparent magnitudes, and diagnostic colors) and simple structural parameters. For continuity with the photometric gas fraction (PGF) literature, we also report the gas-to-stellar-mass ratio, log10(G/S), where explicitly noted. Our dataset is a reproducible cross-match of SDSS DR12, the MPA–JHU value-added catalogs, and the 100% ALFALFA release, resulting in 31,501 galaxies after quality controls. To ensure fair evaluation, we adopt fixed train/validation/test partitions and an additional sky-holdout region to probe domain shift, i.e., how well the model extrapolates to sky regions that were not used for training. We also audit features to avoid information leakage and benchmark the BNNs against deterministic models, including a feed-forward neural network baseline and gradient-boosted trees (GBTs, a standard tree-based ensemble method in machine learning). Performance is assessed using mean absolute error (MAE), root-mean-square error (RMSE), and probabilistic diagnostics such as the negative log-likelihood (NLL, a loss that rewards models that assign high probability to the observed H I masses), reliability diagrams (plots comparing predicted probabilities to observed frequencies), and empirical 68%/95% coverage. The Bayesian models achieve point accuracy comparable to the deterministic baselines while additionally providing calibrated prediction intervals that adapt to stellar mass, surface density, and color. This enables galaxy-by-galaxy uncertainty estimation and prioritization for 21 cm follow-up that explicitly accounts for predicted uncertainties (“risk-aware” target selection). Overall, the results demonstrate that uncertainty-aware machine-learning methods offer a scalable and reproducible route to inferring galactic H I content from widely available optical data.

Abstract: Multi-agent Reinforcement Learning in Networking

Multi-agent systems (MAS) deal with systems where multiple agents must make decisions to
achieve their goals in a shared environment. Despite their ubiquity, their effective
implementation still faces several challenges. We will address these challenges by drawing on
recent advances in multi-agent reinforcement learning (MARL), that is, research on data
driven decision-making within MAS. We will also present some recent applications of MARL
in the context of energy-efficient resource allocation in networks.

Abstract: AI-Enhanced Digital Twins for Intelligent Manufacturing and Predictive Maintenance in Hydrogen Systems

Green hydrogen plays a central role in global decarbonization efforts, serving as a versatile energy carrier for transportation, industry, and energy storage. However, its widespread adoption is hindered by critical safety and reliability challenges, including high flamma- bility, hydrogen embrittlement, leak risks, and accelerated degradation of electrochemical components in systems such as proton exchange membrane (PEM) electrolyzers, storage tanks, pipelines, and refueling stations. This presentation introduces an integrated Digital Twin (DT) framework augmented by Artificial Intelligence (AI) to enable real-time mon- itoring, anomaly detection, predictive maintenance, and enhanced operational safety in hydrogen infrastructures. The approach combines physics-based modeling with advanced AI techniques, including machine learning, deep learning, and Physics-Informed Neural Networks (PINNs). A key focus is a laboratory-scale PEM electrolyzer (two-cell Titan EZ120 stack) mirrored through a DT architecture. The virtual replica, developed using Unity and Blender for interactive 3D visualization, synchronizes with experimental polar- ization data acquired via potentiostat under controlled temperatures (20–80 ◦C). PINNs are embedded within the DT, incorporating electrochemical constraints such as Butler- Volmer kinetics, Arrhenius temperature dependence, and ohmic resistance into a hybrid loss function. A feedforward neural network (4 hidden layers × 64 neurons, ReLU acti- vation) achieves a 12 % reduction in mean squared error compared to conventional neural networks, with RMSE values of 0.05–0.12 V and R² > 0.98 across conditions.  Real- time visualization occurs via a Grafana dashboard integrated with InfluxDB, supporting 10 Hz sampling and low end-to-end latency (150–250 ms). Broader industrial applications include predictive leak detection in refueling stations, degradation forecasting in elec- trolyzers (e.g., simulated overpotential increase of 15–20 % after 1000 cycles), structural health monitoring in pipelines and tanks, and risk assessment in underground storage using federated AI models. The framework addresses limitations such as data sparsity, model explainability, and cybersecurity through standardized DT architectures and open datasets. This work demonstrates how AI-driven DTs foster intelligent, resilient hydrogen ecosystems, accelerating safe deployment of green hydrogen technologies and contributing to sustainable energy transitions in alignment with Industry 4.0 principles.

Abstract: Cultural Landscape Gene–Driven Multimodal 3D Perception and Knowledge-Graph Analytics for Cultural Heritage Landmarks: The Heluo Digital Intelligence Platform

Driven by globalization and rapid digital transformation, heritage conservation and utilization face increasing pressures, including architectural homogenization, uncontrolled urban–rural expansion that compresses cultural space, and the isolation of heritage sites in industrialized contexts. Meanwhile, within heritage tourism, the concept and value interpretation of “cultural landmarks” remain insufficiently articulated, and many digitalization efforts follow fragmented or single-track technical routes. Taking the Heluo region—an important cradle of Chinese civilization—as a case, this paper adopts the Cultural Landscape Gene (CLG) perspective to characterize cultural landmarks along three dimensions: historical accumulation, visual-form expression, and socio-cultural identity, and to explore a sustainable conservation–tourism synergy. We develop the “Heluo Digital Intelligence” platform, underpinned by a CLG management database and GIS-based spatial analytics, and implement a closed-loop workflow of collection, classification, evaluation, reconstruction, and presentation. Multi-source 3D digitization (UAV oblique photogrammetry, terrestrial/handheld 3D scanning, and SfM/MVS reconstruction with meshing and texturing) is integrated with AI to build gene profiles and to automatically extract features from images and 3D models. In addition, NLP-based named-entity recognition and knowledge graphs enable semantic interpretation and relational analysis of cultural genes, supporting digital exhibition, virtual restoration, and cultural IP development. A representative endangered-heritage case demonstrates the feasibility of long-term digital archiving and re-creation. The results indicate that the CLG perspective can bridge value interpretation, data governance, technical implementation, and decision support, offering a replicable pathway for heritage digitalization and sustainable use in comparable regions.