An official website of the United States government
Missing value imputation is a fundamental problem in spatiotemporal modeling, from motion tracking to the dynamics of physical systems. Deep autoregressive models suffer from error propagation which becomes catastrophic for imputing long-range sequences. In this paper, we take a non-autoregressive approach and propose a novel deep generative model: Non-AutOregressive Multiresolution Imputation (NAOMI) to impute long-range sequences given arbitrary missing patterns. NAOMI exploits the multiresolution structure of spatiotemporal data and decodes recursively from coarse to fine-grained resolutions using a divide-and-conquer strategy. We further enhance our model with adversarial training. When evaluated extensively on benchmark datasets from systems of both deterministic and stochastic dynamics. NAOMI demonstrates significant improvement in imputation accuracy (reducing average prediction error by 60% compared to autoregressive counterparts) and generalization for long range sequences.
more »
« less
SDYN-GANs: Adversarial learning methods for multistep generative models for general order stochastic dynamics
We introduce adversarial learning methods for data-driven generative modeling of dynamics of n-th-order stochastic systems. Our approach builds on Generative Adversarial Networks (GANs) with generative model classes based on stable m-step stochastic numerical integrators. From observations of trajectory samples, we introduce methods for learning long-time predictors and stable representations of the dynamics. Our approaches use discriminators based on Maximum Mean Discrepancy (MMD), training protocols using both conditional and marginal distributions, and methods for learning dynamic responses over different time-scales. We show how our approaches can be used for modeling physical systems to learn force-laws, damping coefficients, and noise-related parameters. Our adversarial learning approaches provide methods for obtaining stable generative models for dynamic tasks including long-time prediction and developing simulations for stochastic systems.
more »
« less
- Award ID(s):
- 2306101
- PAR ID:
- 10611634
- Publisher / Repository:
- Elsevier
- Date Published:
- Journal Name:
- Journal of Computational Physics
- Volume:
- 519
- Issue:
- C
- ISSN:
- 0021-9991
- Page Range / eLocation ID:
- 113442
- Subject(s) / Keyword(s):
- Machine Learning Generative AI Adversarial Learning Stochastic Processes Dynamical Systems Data-Driven Modeling Scientific Computation
- Format(s):
- Medium: X
- Sponsoring Org:
- National Science Foundation
More Like this
-
-
Long-term fairness in sequential decision-making is critical yet challenging, as decisions at each time step influence future opportunities and outcomes, potentially exacerbating existing disparities over time. While existing methods primarily achieve fairness by directly adjusting decision models, in this work, we study a complementary perspective based on sequential algorithmic recourse, in which fairness is pursued through actionable interventions for individuals. We introduce Sequential Causal Algorithmic Recourse for Fairness (SCARF), a causally grounded framework that generates temporally coherent recourse trajectories by integrating structural causal modeling with sequential generative modeling. By explicitly incorporating both short-term and long-term fairness constraints, as well as practical budget limitations, SCARF generates personalized recourse plans that effectively mitigate disparities over multiple decision cycles. Through experiments on synthetic and semi-synthetic datasets, we empirically examine how different recourse strategies influence fairness dynamics over time, illustrating the trade-offs between short-term and long-term fairness under sequential interventions. The results demonstrate that SCARF provides a practical and informative framework for analyzing long-term fairness in dynamic decision-making settings.more » « less
-
Li, J.; Spanos, P. D.; Chen, J. B.; Peng, Y. B. (Ed.)Infrastructure networks offer critical services to modern society. They dynamically interact with the environment, operators, and users. Infrastructure networks are unique engineered systems, large in scale and high in complexity. One fundamental issue for their reliability assessment is the uncertainty propagation from stochastic disturbances across interconnected components. Monte Carlo simulation (MCS) remains approachable to quantify stochastic dynamics from components to systems. Its application depends on time efficiency along with the capability of delivering reliable approximations. In this paper, we introduce Quasi Monte Carlo (QMC) sampling techniques to improve modeling efficiency. Also, we suggest a principled Monte Carlo (PMC) method that equips the crude MCS with Probably Approximately Correct (PAC) approaches to deliver guaranteed approximations. We compare our proposed schemes with a competitive approach for stochastic dynamic analysis, namely the Probability Density Evolution Method (PDEM). Our computational experiments are on ideal but complex enough source-terminal (S-T) dynamic network reliability problems. We endow network links with oscillators so that they can jump across safe and failed states allowing us to treat the problem from a stochastic process perspective. We find that QMC alone can yield practical accuracy, and PMC with a PAC algorithm can deliver accuracy guarantees. Also, QMC is more versatile and efficient than the PDEM for network reliability assessment. The QMC and PMC methods provide advanced uncertainty propagation techniques to support decision makers with their reliability problems.more » « less
-
The discovery of new materials is essential for enabling technological advancements. Computational approaches for predicting novel materials must effectively learn the manifold of stable crystal structures within an infinite design space. We introduce Open Materials Generation (OMatG), a unifying framework for the generative design and discovery of inorganic crystalline materials. OMatG employs stochastic interpolants (SI) to bridge an arbitrary base distribution to the target distribution of inorganic crystals via a broad class of tunable stochastic processes, encompassing both diffusion models and flow matching as special cases. In this work, we adapt the SI framework by integrating an equivariant graph representation of crystal structures and extending it to account for periodic boundary conditions in unit cell representations. Additionally, we couple the SI flow over spatial coordinates and lattice vectors with discrete flow matching for atomic species. We benchmark OMatG's performance on two tasks: Crystal Structure Prediction (CSP) for specified compositions, and de novo generation (DNG) aimed at discovering stable, novel, and unique structures. In our ground-up implementation of OMatG, we refine and extend both CSP and DNG metrics compared to previous works. OMatG establishes a new state of the art in generative modeling for materials discovery, outperforming purely flow-based and diffusion-based implementations. These results underscore the importance of designing flexible deep learning frameworks to accelerate progress in materials science. The OMatG code is available at https://github.com/FERMat-ML/OMatG.more » « less
-
To monitor the dynamic behavior of degrading systems over time, a flexible hierarchical discrete-time state-space model (SSM) is introduced that can mathematically characterize the stochastic evolution of the latent states (discrete, continuous, or hybrid) of degrading systems, dynamic measurements collected from condition monitoring sources (e.g., sensors with mixed-type out-puts), and the failure process. This flexible SSM is inspired by Bayesian hierarchical modeling and recurrent neural networks without imposing prior knowledge regarding the stochastic structure of the system dynamics and its variables. The temporal behavior of degrading systems and the relationship between variables of the corresponding system dynamics are fully characterized by stochastic neural networks without having to define parametric relationships/distributions between deterministic and stochastic variables. A Bayesian filtering-based learning method is introduced to train the structure of the proposed framework with historical data. Also, the steps to utilize the proposed framework for inference and prediction of the latent states and sensor outputs are dis-cussed. Numerical experiments are provided to demonstrate the application of the proposed framework for degradation system modeling and monitoring.more » « less
- Free Publicly Accessible Full Text
-
Accepted Manuscript
- Journal Article:
- https://doi.org/10.1016/j.jcp.2024.113442
