SCIENTIFIC COMPUTING AND IMAGING INSTITUTE
at the University of Utah

This article summarizes the vision, roadmap, and implementation plan for a National Artificial Intelligence Research Resource that aims to provide a widely accessible cyberinfrastructure for artificial intelligence R&D, with the overarching goal of bridging the resource–access divide.

Physics-informed neural networks (PINNs) as a means of discretizing partial differential equations (PDEs) are garnering much attention in the Computational Science and Engineering (CS&E) world. At least two challenges exist for PINNs at present: an understanding of accuracy and convergence characteristics with respect to tunable parameters and identification of optimization strategies that make PINNs as efficient as other computational science tools. The cost of PINNs training remains a major challenge of Physics-informed Machine Learning (PiML) – and, in fact, machine learning (ML) in general. This paper is meant to move towards addressing the latter through the study of PINNs on new tasks, for which parameterized PDEs provides a good testbed application as tasks can be easily defined in this context. Following the ML world, we introduce metalearning of PINNs with application to parameterized PDEs. By introducing metalearning and transfer learning concepts, we can greatly accelerate the PINNs optimization process. We present a survey of model-agnostic metalearning, and then discuss our model-aware metalearning applied to PINNs as well as implementation considerations and algorithmic complexity. We then test our approach on various canonical forward parameterized PDEs that have been presented in the emerging PINNs literature.

Physics-informed neural networks (PINNs) as a means of solving partial differential equations (PDE) have garnered much attention in the Computational Science and Engineering (CS&E) world. However, a recent topic of interest is exploring various training (i.e., optimization) challenges – in particular, arriving at poor local minima in the optimization landscape results in a PINN approximation giving an inferior, and sometimes trivial, solution when solving forward time-dependent PDEs with no data. This problem is also found in, and in some sense more difficult, with domain decomposition strategies such as temporal decomposition using XPINNs. To address this problem, we first enable a general categorization for previous causality methods, from which we identify a gap (e.g., opportunity) in the previous approaches. We then furnish examples and explanations for different training challenges, their cause, and how they relate to information propagation and temporal decomposition. We propose a solution to fill this gap by reframing these causality concepts into a generalized information propagation framework in which any prior method or combination of methods can be described. This framework is easily modifiable via user parameters in the open-source code accompanying this paper. Our unified framework moves toward reducing the number of PINN methods to consider and the reimplementation and retuning cost for thorough comparisons rather than increasing it. Using the idea of information propagation, we propose a new stacked-decomposition method that bridges the gap between time-marching PINNs and XPINNs. We also introduce significant computational speed-ups by using transfer learning concepts to initialize subnetworks in the domain and loss tolerance-based propagation for the subdomains. Finally, we formulate a new time-sweeping collocation point algorithm inspired by the previous PINNs causality literature, which our framework can still describe, and provides a significant computational speed-up via reduced-cost collocation point segmentation. The proposed methods overcome training challenges in PINNs and XPINNs for time-dependent PDEs by respecting the causality in multiple forms and improving scalability by limiting the computation required per optimization iteration. Finally, we provide numerical results for these methods on baseline PDE problems for which unmodified PINNs and XPINNs struggle to train.

Spherical harmonics glyphs are an established way to visualize high angular resolution diffusion imaging data. Starting from a unit sphere, each point on the surface is scaled according to the value of a linear combination of spherical harmonics basis functions. The resulting glyph visualizes an orientation distribution function. We present an efficient method to render these glyphs using ray tracing. Our method constructs a polynomial whose roots correspond to ray-glyph intersections. This polynomial has degree 2k + 2 for spherical harmonics bands 0, 2, . . . , k. We then find all intersections in an efficient and numerically stable fashion through polynomial root finding. Our formulation also gives rise to a simple formula for normal vectors of the glyph. Additionally, we compute a nearly exact axis-aligned bounding box to make ray tracing of these glyphs even more efficient. Since our method finds all intersections for arbitrary rays, it lets us perform sophisticated shading and uncertainty visualization. Compared to prior work, it is faster, more flexible and more accurate.

Processing, managing and publishing the substantial volume of data collected through modern remote sensing technologies in a format that is easy for researchers - across broad skill levels and scientific domains - to view and use presents a formidable challenge. As a prime example, the massive scale of image mosaics produced by NEON’s Airborne Observation Platform (AOP), often several to hundreds of gigabytes in volume, demands efficient data management strategies. Additionally, these aerial mosaics frequently exhibit seams due to variations in lighting conditions during the data acquisition process. These seams undermine the integrity of subsequent scientific analyses, introducing distortions that hinder accurate interpretation of ecological patterns. Finally, one of NEON’s core objectives is to make these data broadly accessible to users, including those who are not yet versed in working with remote sensing data or who wish to view the datasets without needing to download and process them.In response to these challenges, we have developed a comprehensive data management pipeline that enables interactive access for analysis and visualization of NEON’s aerial mosaic collection. This pipeline automates data ingestion, conversion, and publication in a streamable format, facilitating seamless user interaction through web viewers and programming APIs. Moreover, we have implemented a portable blending algorithm aimed at eliminating these problematic seams from large aerial mosaics. This algorithm, grounded in the Conjugate Gradient (CG) method, has been implemented both in CUDA and using the modern SYCL programming model for enhanced portability across diverse computing platforms.Experimental results demonstrate scalable performance across both CPU and GPU architectures. This work not only addresses the challenges of large aerial data management and seam removal but also opens avenues for more accurate and comprehensive scientific investigations within the NEON ecosystem.

This letter proposes multitask learning as a regularization method for segmentation tasks in seismic images. We examine application-specific auxiliary tasks, such as the estimation/detection of horizons, dip angle, and amplitude that geophysicists consider relevant for identification of channels (a geological feature), which is currently done through painstaking outlining by qualified experts. We show that multitask training helps in better generalization on test datasets with very similar and different structure/statistics. In such settings, we also show that multitask learning performs better on unseen datasets relative to the baseline.

The variational autoencoder (VAE) [1] is a well-studied, deep, latent-variable model (DLVM) that efficiently optimizes the variational lower bound of the log marginal data likelihood and has a strong theoretical foundation. However, the VAE’s known failure to match the aggregate posterior often results in pockets/holes in the latent distribution (i.e., a failure to match the prior) and/or posterior collapse, which is associated with a loss of information in the latent space. This paper addresses these shortcomings in VAEs by reformulating the objective function associated with VAEs in order to match the aggregate/marginal posterior distribution to the prior. We use kernel density estimate (KDE) to model the aggregate posterior in high dimensions. The proposed method is named the aggregate variational autoencoder (AVAE) and is built on the theoretical framework of the VAE. Empirical evaluation of the proposed method on multiple benchmark data sets demonstrates the effectiveness of the AVAE relative to state-of-the-art (SOTA) methods.

Homography estimation is a basic image-alignment method in many applications. Recently, with the development of convolutional neural networks (CNNs), some learning based approaches have shown great success in this task. However, the performance across different domains has never been researched. Unlike other common tasks (e.g., classification, detection, segmentation), CNN based homography estimation models show a domain shift immunity, which means a model can be trained on one dataset and tested on another without any transfer learning. To explain this unusual performance, we need to determine how CNNs estimate homography. In this study, we first show the domain shift immunity of different deep homography estimation models. We then use a shallow network with a specially designed dataset to analyze the features used for estimation. The results show that networks use low-level texture information to estimate homography. We also design some experiments to compare the performance between different texture densities and image features distorted on some common datasets to demonstrate our findings. Based on these findings, we provide an explanation of the domain shift immunity of deep homography estimation.

Asynchronous task-based systems offer the possibility of making it easier to take advantage of scalable heterogeneous architectures.
This paper extends the National Institute of Standards and Technology’s Hedgehog dataflow graph models, which target a single high-end
compute node, to run on a cluster by borrowing aspects of Uintah’s cluster-scale task graphs and applying them to a sample implementation
of matrix multiplication. These results are compared to implementations using the leading libraries, SLATE and DPLASMA, for illustrative purposes only. The motivation behind this work is to demonstrate that using general purpose high-level abstractions, such as Hedgehog’s dataflow graphs, does not negatively impact performance.

Deep neural operators, such as DeepONets, have changed the paradigm in high-dimensional nonlinear regression from function regression to (differential) operator regression, paving the way for significant changes in computational engineering applications. Here, we investigate the use of DeepONets to infer flow fields around unseen airfoils with the aim of shape optimization, an important design problem in aerodynamics that typically taxes computational resources heavily. We present results which display little to no degradation in prediction accuracy, while reducing the online optimization cost by orders of magnitude. We consider NACA airfoils as a test case for our proposed approach, as their shape can be easily defined by the four-digit parametrization. We successfully optimize the constrained NACA four-digit problem with respect to maximizing the lift-to-drag ratio and validate all results by comparing them to a high-order CFD solver. We find that DeepONets have low generalization error, making them ideal for generating solutions of unseen shapes. Specifically, pressure, density, and velocity fields are accurately inferred at a fraction of a second, hence enabling the use of general objective functions beyond the maximization of the lift-to-drag ratio considered in the current work.

Deep neural operators, such as DeepONet, have changed the paradigm in high-dimensional nonlinear regression, paving the way for significant generalization and speed-up in computational engineering applications. Here, we investigate the use of DeepONet to infer flow fields around unseen airfoils with the aim of shape constrained optimization, an important design problem in aerodynamics that typically taxes computational resources heavily. We present results that display little to no degradation in prediction accuracy while reducing the online optimization cost by orders of magnitude. We consider NACA airfoils as a test case for our proposed approach, as the four-digit parameterization can easily define their shape. We successfully optimize the constrained NACA four-digit problem with respect to maximizing the lift-to-drag ratio and validate all results by comparing them to a high-order CFD solver. We find that DeepONets have a low generalization error, making them ideal for generating solutions of unseen shapes. Specifically, pressure, density, and velocity fields are accurately inferred at a fraction of a second, hence enabling the use of general objective functions beyond the maximization of the lift-to-drag ratio considered in the current work. Finally, we validate the ability of DeepONet to handle a complex 3D waverider geometry at hypersonic flight by inferring shear stress and heat flux distributions on its surface at unseen angles of attack. The main contribution of this paper is a modular integrated design framework that uses an over-parametrized neural operator as a surrogate model with good generalizability coupled seamlessly with multiple optimization solvers in a plug-and-play mode.

TEMA project is a Horizon Europe funded project that aims at addressing Natural Disaster Management by the use of sophisticated Cloud-Edge Continuum infrastructures by means of data analysis algorithms wrapped in Serverless functions deployed on a distributed infrastructure according to a Federated Learning scheduler that constantly monitors the infrastructure in search of the best way to satisfy required QoS constraints. In this paper, we discuss the advantages of Serverless workflow and how they can be used and monitored to natively trigger complex algorithm pipelines in the continuum, dynamically placing and relocating them taking into account incoming IoT data, QoS constraints, and the current status of the continuum infrastructure. Therefore we presented the Urgent Function Enabler (UFE) platform, a fully distributed architecture able to define, spread, and manage FaaS functions, using local IOT data managed using the Fiware ecosystem and a computing infrastructure composed of mobile and stable nodes.

Continuum Computing encompasses the integration of diverse infrastructures, including cloud, edge, and fog, to facilitate seamless
migration of applications based on their specific needs, ensuring optimal satisfaction of their requirements. The primary obstacles in this particular context mostly pertain to the incapacity to promptly respond to changes in the environment or the quality of service (QoS) constraints of the application, as well as the incapability to maintain an application in a stateless manner, hence impeding its relocation without the risk of data loss. The objective of this research is to tackle the aforementioned issues through the introduction of a framework based on Function-as-a-Service (FaaS) and event-driven architecture. This framework enables the decomposition, localization, and relocation of applications inside a Continuum infrastructure, facilitated by a rule engine that is both system and data-aware

Accurate assessment of left atrial fibrosis in patients with atrial fibrillation relies on high-quality 3D late gadolinium enhancement (LGE) MRI images. However, obtaining such images is challenging due to patient motion, changing breathing patterns, or sub-optimal choice of pulse sequence parameters. Automated assessment of LGE-MRI image diagnostic quality is clinically significant as it would enhance diagnostic accuracy, improve efficiency, ensure standardization, and contributes to better patient outcomes by providing reliable and high-quality LGE-MRI scans for fibrosis quantification and treatment planning. To address this, we propose a two-stage deep-learning approach for automated LGE-MRI image diagnostic quality assessment. The method includes a left atrium detector to focus on relevant regions and a deep network to evaluate diagnostic quality. We explore two training strategies, multi-task learning, and pretraining using contrastive learning, to overcome limited annotated data in medical imaging. Contrastive Learning result shows about 4%, and 9% improvement in F1-Score and Specificity compared to Multi-Task learning when there’s limited data.

Accurate assessment of left atrial fibrosis in patients with atrial fibrillation relies on high-quality 3D late gadolinium enhancement (LGE) MRI images. However, obtaining such images is challenging due to patient motion, changing breathing patterns, or sub-optimal choice of pulse sequence parameters. Automated assessment of LGE-MRI image diagnostic quality is clinically significant as it would enhance diagnostic accuracy, improve efficiency, ensure standardization, and contributes to better patient outcomes by providing reliable and high-quality LGE-MRI scans for fibrosis quantification and treatment planning. To address this, we propose a two-stage deep-learning approach for automated LGE-MRI image diagnostic quality assessment. The method includes a left atrium detector to focus on relevant regions and a deep network to evaluate diagnostic quality. We explore two training strategies, multi-task learning, and pretraining using contrastive learning, to overcome limited annotated data in medical imaging. Contrastive Learning result shows about 4%, and 9% improvement in F1-Score and Specificity compared to Multi-Task learning when there's limited data.

The decentralized stochastic gradient method emerges as a promising solution for solving large-scale machine learning problems. This paper studies the decentralized Markov chain gradient descent (DMGD), a variant of the decentralized stochastic gradient method, which draws random samples along the trajectory of a Markov chain. DMGD arises when obtaining independent samples is costly or impossible, excluding the use of the traditional stochastic gradient algorithms. Specifically, we consider the DMGD over a connected graph, where each node only communicates with its neighbors by sending and receiving the intermediate results. We establish both ergodic and nonergodic convergence rates of DMGD, which elucidate the critical dependencies on the topology of the graph that connects all nodes and the mixing time of the Markov chain. We further numerically verify the sample efficiency of DMGD.

In this paper, we propose an adaptive online gradient descent method with momentum for pairwise learning, in which the step size is determined by historical information. Due to the structure of pairwise learning, the sample pairs are dependent on the parameters, causing difficulties in the convergence analysis. To this end, we develop novel techniques for the convergence analysis of the proposed algorithm. We show that the proposed algorithm can output the desired solution in strongly convex, convex, and nonconvex cases. Furthermore, we present theoretical explanations for why our proposed algorithm can accelerate previous workhorses for online pairwise learning. All assumptions used in the theoretical analysis are mild and common, making our results applicable to various pairwise learning problems. To demonstrate the efficiency of our algorithm, we compare the proposed adaptive method with the non-adaptive counterpart on the benchmark online AUC maximization problem.

We have developed UncertainSCI (UncertainSCI, 2020) as an open-source tool designed to make modern uncertainty quantification (UQ) techniques more accessible in biomedical simulation applications. UncertainSCI is implemented in Python with a noninvasive interface to meet our software design goals of 1) numerical accuracy, 2) simple application programming interface (API), 3) adaptability to many applications and methods, and 4) interfacing with diverse simulation software. Using a Python implementation in UncertainSCI allowed us to utilize the popularity and low barrier-to-entry of Python and its common packages and to leverage the built-in integration and support for Python in common simulation software packages and languages. Additionally, we used noninvasive UQ techniques and created a similarly noninvasive interface to external modeling software that can be called in diverse ways, depending on the complexity and level of Python integration in the external simulation pipeline. We have developed and included examples applying UncertainSCI to relatively simple 1D simulations implemented in Python, and to bioelectric field simulations implemented in external software packages, which demonstrate the use of UncertainSCI and the effectiveness of the architecture and implementation in achieving our design goals. UnceratainSCI differs from similar software, notably UQLab, Uncertainpy, and Simnibs, in that it can be efficiently and non-invasively used with external simulation software, specifically with high resolution 3D simulations often used in Bioelectric field simulations. Figure 1 illustrates the use of UncertainSCI in computing UQ with modeling pipelines for bioelectricity simulations

Separating algorithms from their computation schedule has become a de facto solution to tackle the challenges of developing high performance code on modern heterogeneous architectures. Common approaches include Domain-specific languages (DSLs) which provide familiar APIs to domain experts, code generation frameworks that automate the generation of fast and portable code, and runtime systems that manage threads for concurrency and parallelism. In this paper, we present the Halide code generation framework for Phylanx distributed array processing platform. This extension enables compile-time optimization of Phylanx primitives for target architectures. To accomplish this, (1) we implemented new Phylanx primitives using Halide, and (2) partially exported Halide's thread pool API to carry out parallelism on HPX (Phylanx's runtime) threads. (3) showcased HPX performance analysis tools made available to Halide applications. The evaluation of the work has been done in two steps. First, we compare the performance of Halide applications running on its native runtime with that of the new HPX backend to verify there is no cost associated with using HPX threads. Next, we compare performances of a number of original implementations of Phylanx primitives against the new ones in Halide to verify performance and portability benefits of Halide in the context of Phylanx.