cs.LGMar 27, 2026

GLU: Global-Local-Uncertainty Fusion for Scalable Spatiotemporal Reconstruction and Forecasting

Authors: Linzheng Wang, Jason Chen, Nicolas Tricard, Zituo Chen, Sili Deng

Organizations: Department of Mechanical Engineering, Massachusetts Institute of Technology, Cambridge, 02139, MA, United States of America.

Read PDF
Ask AI
Cite

Copy citation

DOINot available for this paper

Abstract

Digital twins of complex physical systems are expected to infer unobserved states from sparse measurements and predict their evolution in time, yet these two functions are typically treated as separate tasks. Here we present GLU, a Global-Local-Uncertainty framework that formulates sparse reconstruction and dynamic forecasting as a unified state-representation problem and introduces a structured latent assembly to both tasks. The central idea is to build a structured latent state that combines a global summary of system-level organization, local tokens anchored to available measurements, and an uncertainty-driven importance field that weights observations according to the physical informativeness. For reconstruction, GLU uses importance-aware adaptive neighborhood selection to retrieve locally relevant information while preserving global consistency and allowing flexible query resolution on arbitrary geometries. Across a suite of challenging benchmarks, GLU consistently improves reconstruction fidelity over reduced-order, convolutional, neural operator, and attention-based baselines, better preserving multi-scale structures. For forecasting, a hierarchical Leader-Follower Dynamics module evolves the latent state with substantially reduced memory growth, maintains stable rollout behavior and delays error accumulation in nonlinear dynamics. On a realistic turbulent combustion dataset, it further preserves not only sharp fronts and broadband structures in multiple physical fields, but also their cross-channel thermo-chemical couplings. Scalability tests show that these gains are achieved with substantially lower memory growth than comparable attention-based baselines. Together, these results establish GLU as a flexible and computationally practical paradigm for sparse digital twins.

Figures & tables

Figure 1: Bridging the reality gap between sparse sensing and dynamic digital twins. a) Complex physical systems monitored by sparse, irregular sensors. b) A continuous stream of sparse, partial observations serves as the input. c) The GLU framework decomposes measurements into three complementary streams. Global tokens extract system-level patterns and long-range correlations, while local tokens preserve fine-grained sensor fidelity. A learned uncertainty signal guides the assimilation of local data. d) The GLU-assembled representations drive the digital twin, enabling both full-state inference of unobserved quantities and stable forecasting of future trajectories.
Figure 2: Architecture and performance of the GLU framework. a, Sparse encoding via global-local-aware encoder with bidirectional attention to transform arbitrary sparse inputs into a latent state. b, Soft adaptive reconstruction. For any query location, the model dynamically aggregates top- kk most relevant sensors using an importance-warped distance metric. This process decodes both the mean field value and a spatial uncertainty estimate. c, Leader-Follower-Dynamic structure to model latent temporal evolution. The global leader drives the dynamics by attending to important regions via IMP-Gating, while sensor followers update their local states by attending to the leader. d, Normalized reconstruction error across seven datasets ranging from canonical flows to chaotic reaction-diffusion and multi-physics turbulent combustion. e, Forecasting error accumulation for periodic (Collinear Re40) and chaotic (Collinear Re100, Reaction-Diffusion) dynamics.
Figure 3: Adaptive spatial reconstruction mechanisms and information scaling performances. a, Learned importance score distributions ( ϕ\phi ) for representative datasets. b, Multi-scale detail recovery: (1) Ground truth and scale-separated reconstructions (large/medium/small) from global-only ablation, Senseiver, and GLU; (2) Spectral and scale-wise summaries. GLU best matches the ground-truth energy spectrum and retains the most energy across scales (bar chart), yielding the lowest error at each scale. c, Mechanism of adaptive selection: (1) Visualization of the adaptive neighborhood on Collinear Flow; (2) Reconstruction error vs. local field complexity; (3) Probability density function of reconstruction errors; d, Reconstruction error versus number of sensors across six datasets.
Figure 4: Long-term forecasting stability on complicated dynamics. a, Latent dynamics visualization. 2D PCA projection of latent trajectories for (1) periodic (Collinear Re40, 32 sensors) and (2) chaotic (Collinear Re100, 96 sensors) flows. Observation windows contain 16 steps. (3-4) Corresponding pointwise velocity forecasting confirms that GLU-LFD avoids the phase shifts and amplitude collapses seen in baselines. b, Generalization to unseen dynamics. (1) Error evolution of the reaction-diffusion system from unseen random initial condition during training from 16 steps of sparse observations (64 sensors). (2-3) Quantitative stability metrics. GLU-LFD preserves the correct spatial ( grg_{r} ) and temporal correlation ( gtg_{t} ) structures over time, preventing the physical degradation.
Figure 5: Performance of GLU in learning realistic multi-physics systems (turbulent combustion). a, Instantaneous snapshots and error maps for all channels from 1% sparse sensors. GLU minimizes structural errors at the flame front (sharp interfaces) and in turbulent regions compared to baselines, which suffer from blurring. b, Energy spectra (bottom) and log-spectral distance (LSD) scores (top). GLU adheres to the ground truth spectrum at high wavenumbers, capturing fine-scale turbulence that baselines filter out. c, Joint probability density functions visualizing cross-channel correlations. The low Jensen-Shannon divergence (JSD) scores confirm that GLU accurately preserves the thermodynamic and kinematic couplings inherent in combustion.
Figure 6: Scalability and memory comparison for spatial and temporal reconstruction. a, Peak GPU memory versus the number of reconstruction points, NptsN_{\mathrm{pts}} , for Senseiver and GLU with different information-bottleneck top- KK sensor settings. Lower top- KK reduces GLU memory, enabling a tunable efficiency–accuracy trade-off while maintaining substantially better memory scaling than Senseiver. b, Corresponding relative L2L_{2} norm for the models on turbulent channel flow dataset. c, Peak GPU memory for temporal reconstruction as a function of the number of reference frames, NrefN_{\mathrm{ref}} , and window size, NwindowN_{\mathrm{window}} . GLU-LFD achieves the lowest memory usage across all temporal settings.
More in this topic

Spatiotemporal Forecasting

Spatiotemporal forecasting predicts how quantities change across locations and time, from traffic and weather to disease and environmental systems.

See all 423 papers

Explore similar work

CardsList
Sep 30, 2026

Grounding Time-Series Foundation Models in Digital Twin Topology for Predictive Maintenance

Digital twins increasingly support downstream analytical tasks that depend on time-series data, motivating interest in time-series foundation models (TSFMs) as scalable backbones. However, TSFMs are primarily pretrained for temporal continuation and often underperform on unseen tasks such as regression, and systematic empirical comparisons against state-of-the-art dedicated models in digital twin contexts remain limited. This paper makes three contributions. First, we benchmark five well-known TSFMs with frozen backbones on remaining useful life (RUL) prediction using the C-MAPSS dataset, finding that multivariate architectures substantially outperform univariate ones, particularly under varying operating conditions. This raises a deeper question: when cross-channel dependencies can be modeled through pretrained weights, target-task adaptation, and digital twin-derived representations, how much does each contribute, and are they complementary? Second, we propose a topology-informed fusion approach in which topological constraints, derived from the asset structure the digital twin stores among its information models, explicitly shape cross-attention, so that fused representations respect the physical system's local connectivity rather than relying on unconstrained all-to-all interactions. Third, we conduct an ablation study across C-MAPSS subsets of varying operational complexity that isolates the three sources and their interactions. The sources prove complementary rather than redundant, and topology-constrained attention outperforms unconstrained fusion, though by a small margin, enabling a frozen TSFM informed by digital twin representations to remain competitive or in some cases exceed state-of-the-art performance on this regression task.

Sizhe Ma, Katherine A. Flanigan, Mario Bergés

Department of Civil & Environmental Engineering, Carnegie Mellon University 5000 Forbes Ave., Pittsburgh, PA USA 15213

RUL EstimationDigital TwinsTime-Series Foundation Models
AskPDF
Oct 17, 2026

GTSM: Gaussian Splatting with Thermal State-Space Dynamics for Spatio-Temporal Thermal Field Reconstruction and Prediction

Temperature-field reconstruction and prediction are important for condition monitoring, process analysis, and state assessment in industrial and dynamic environments. This work aims to recover a four-dimensional spatio-temporal thermal state field directly from two-dimensional observations, without requiring a predefined 3D model, prescribed thermodynamic equation, or predefined heating/cooling trajectory. To this end, we propose the Gaussian Thermal State-Space Model (GTSM), which represents thermal evolution through a shared scene-level latent process and Gaussian-specific local thermal states. The resulting continuous dynamics are propagated using differentiable RK2 integration, allowing thermal evolution to be inferred from multi-view and multi-temporal observations and continuously extrapolated beyond the observed interval. GTSM is designed as a backbone-agnostic thermal dynamics module for Gaussian representations and can be integrated with both static 3D Gaussian Splatting and dynamic 4D Gaussian Splatting, supporting thermal reconstruction and prediction under both static and time-varying target geometry. The reconstructed representation further enables novel-view thermal rendering and future-time thermal-state prediction. Experiments on real infrared sequences and synthetic dynamic thermal data demonstrate improved reconstruction accuracy and substantially stronger temporal extrapolation than representative baselines, while also validating the applicability of GTSM to deforming targets.

Yunpeng Zhao, Haoxiang Zhang, Chao Liu

Key Laboratory for Information Science of Electromagnetic Waves (MoE), Fudan University, Shanghai 200433, China · Department of Electronic Engineering, Tsinghua University, Beijing, China

Spatiotemporal ForecastingThermal ImagingState Space Models3D Gaussian Splatting
AskPDF
Feb 17, 2026

Uncertainty-Calibrated Spatiotemporal Field Diffusion with Sparse Supervision

Physical fields are typically observed only at sparse, time-varying sensor locations, making forecasting and reconstruction ill-posed and uncertainty-critical. We present SOLID, a mask-conditioned diffusion framework that learns spatiotemporal dynamics from sparse observations alone: training and evaluation use only observed target locations, requiring no dense fields and no pre-imputation. Unlike prior work that trains on dense reanalysis or simulations and only tests under sparsity, SOLID is trained end-to-end with sparse supervision only. SOLID conditions each denoising step on the measured values and their locations, and introduces a dual-masking objective that (i) emphasizes learning in unobserved void regions while (ii) upweights overlap pixels where inputs and targets provide the most reliable anchors. This strict sparse-conditioning pathway enables posterior sampling of full fields consistent with the measurements, achieving up to an order-of-magnitude improvement in probabilistic error and yielding calibrated uncertainty maps (\r{ho} > 0.7) under severe sparsity.

Kevin Valencia, Xihaier Luo, Shinjae Yoo +1

University of California, Los Angeles · Brookhaven National Laboratory

Spatiotemporal ForecastingDiffusion-Based Inverse ProblemsMasked Diffusion ModelsUncertainty CalibrationTime Series Forecasting
AskPDF