About

Dimitris Giovanis is an assistant research professor in the Department of Civil and Systems Engineering at Johns Hopkins University, with a secondary appointment in the Department of Applied Mathematics and Statistics. His research focuses on the development of advanced computational methodologies and tools at the intersection of probabilistic modeling, data science, and physics-informed machine learning. His work aims to accelerate and optimize simulation and analysis for scientific discovery and data-informed decision making in science and engineering. A central objective of his work is the creation of digital twins—virtual representations that integrate data and physics-based models—for a variety of complex systems. Giovanis' research is applied across diverse domains including materials science, natural hazards, health and biomedicine, aerospace engineering, and astrophysics. His methods support the development of models for structural ceramics, energetic materials, carbon-based composites, and amorphous solids, as well as performance-based earthquake and wind engineering, regional hazard modeling, and post-wildfire debris flow analysis. In health and biomedicine, his work involves traumatic brain injury, digital twins of the human heart, and epidemic modeling. His aerospace research includes structural and aeroelastic systems, while in astrophysics, he contributes to space weather modeling. His research is supported by agencies such as the NSF, DOE, and DARPA, including a DARPA INTACT Grant awarded in 2025.

Research topics

• Machine Learning
• Artificial Intelligence
• Computer Science
• Mathematics
• Algorithm
• Programming language
• Statistics
• Applied mathematics
• Theoretical computer science
• Mathematical analysis
• Mathematical optimization
• Computational science
• Pure mathematics

Selected publications

• From data to process insights: Hybrid modeling strategies for chemical vapor deposition processes

  International Journal of Refractory Metals and Hard Materials · 2026-03-07

  articleOpen access

  The optimization of complex manufacturing processes, such as Chemical Vapor Deposition, requires integrated approaches that combine physical modeling with advanced data-driven methodologies. This review synthesizes recent advances in hybrid modeling frameworks that merge equation-based computational fluid dynamics, machine learning, and natural language processing models to enhance process understanding, prediction, optimization and control. In particular, natural language processing techniques are leveraged to generate embedding-based predictors that inform learning tasks. The proposed framework integrates data acquisition, dimensionality reduction, and feature engineering with contextual language processing embeddings, surrogate modeling, and sensitivity analysis. This results in improved forecasting accuracy and interpretability. Key applications include coating thickness prediction, process regime classification, and critical parameter identification using SHAP analysis and Sobol’ indices. Nevertheless, significant challenges remain, including limitations in sensor infrastructure, assessment of dataset sufficiency for specific industrial objectives, and restricted generalizability across reactor designs. This work highlights how hybrid frameworks, together with natural language processing models applied to industrial process datasets, can bridge the gap between data availability in industrial environments and the actionable insights required for practical implementation, while identifying necessary future directions for robust, scalable, and interpretable modeling systems in advanced manufacturing. • Hybrid framework that unifies CFD, machine learning, and NLP for industrial CVD. • Dimensionality reduction and surrogate models accelerate predictive CVD analytics. • Clustering and feature-importance analysis reveal dominant process mechanisms. • NLP-based encoding enhances predictive modeling of categorical industrial variables.

  PublisherDOI

• Generative learning of densities on manifolds

  Computer Methods in Applied Mechanics and Engineering · 2025-07-30 · 2 citations

  articleOpen access1st authorCorresponding

  A generative modeling framework is proposed that combines diffusion models and manifold learning to efficiently sample data densities on manifolds. The approach utilizes Diffusion Maps to uncover possible low-dimensional underlying (latent) spaces in the high-dimensional data (ambient) space. Two approaches for sampling from the latent data density are described. The first is a score-based diffusion model, which is trained to map a standard normal distribution to the latent data distribution using a neural network. The second one involves solving an Itô stochastic differential equation in the latent space. Additional realizations of the data are generated by lifting the samples back to the ambient space using Double Diffusion Maps , a recently introduced technique typically employed in studying dynamical system reduction; here the focus lies in sampling densities rather than system dynamics. The proposed approaches enable sampling high dimensional data densities restricted to low-dimensional, a priori unknown manifolds. The efficacy of the proposed framework is demonstrated through a benchmark problem and a material with multiscale structure.

  PublisherDOI

• Implementing NLP in industrial process modeling: Addressing categorical variables

  Computers & Chemical Engineering · 2025-04-21 · 2 citations

  articleOpen access

  Important variables of processes are often categorical, i.e. names or labels representing, e.g. categories of inputs, or types of reactors or a sequence of steps. In this work, we use Natural Language Processing Models to derive embeddings of such inputs that represent their actual meaning, or reflect the “distances” between categories, i.e. how similar or dissimilar they are. This is a marked difference from the current standard practice of using binary, or one-hot encoding to replace categorical variables with sequences of ones and zeros. Combined with dimensionality reduction techniques , either linear such as Principal Component Analysis, or nonlinear such as Uniform Manifold Approximation and Projection, the proposed approach leads to a meaningful , low-dimensional feature space. The significance of obtaining meaningful embeddings is illustrated in the context of an industrial coating process for cutting tools that includes both numerical and categorical inputs. In this industrial process, subject matter expertise suggests that the categorical inputs are critical for determining the final outcome but this cannot be taken into account with the current state-of-the-art. The proposed approach enables feature importance which is a marked improvement compared to the current state-of-the-art in the encoding of categorical variables. The proposed approach is not limited to the case-study presented here and is suitable for applications with similar mix of categorical and numerical critical inputs. • Inputs represented by categorical values are embedded by language models. • Dimensionality reduction is used to reduce size of resulting inputs. • The similarity between categories is realistically represented. • Accurate predictive models are trained using new input embeddings. • Realistic feature importance is enabled using Shapley analysis.

  PublisherDOI

• UQpy Version 4.2: Uncertainty quantification with Python

  SoftwareX · 2025-09-25

  articleOpen access

  We introduce a new module for the UQpy software package which extends its capabilities into the field of Scientific Machine Learning. This module builds on PyTorch to create a flexible and robust platform for uncertainty quantification in machine learning. The scientific machine learning module of UQpy introduces custom layers, neural networks, and neural network trainers that are compatible with torch version 2.2.2 and allow for “plug and play” integration into existing torch code.

  PublisherDOI

• Guided Wave-Based Structural Awareness Under Varying Operating States via Manifold Representations

  ArXiv.org · 2025-04-15

  preprintOpen access

  Guided wave-based structural health monitoring (SHM) remains a powerful strategy for identifying early-stage defects and safeguarding vital aerospace structures. Yet, its practical use is often hindered by the enormous, high-dimensional data streams produced by sensor arrays operating at megahertz sampling rates, coupled with the added complexity of shifts in environmental and operational conditions (EOCs). Studies have explored various data-compression approaches that retain critical diagnostic details in a lower-dimensional latent space. While conventional techniques can streamline dimensionality to some extent, they do not always capture the nonlinear interactions typical of guided waves. Manifold learning, as illustrated by Diffusion Maps, tackles these nonlinearities by deriving low-dimensional embeddings directly from wave signals, minimizing the need for manual feature extraction. In parallel, developments in deep learning -- particularly autoencoders -- provide an encoder-decoder model for both data compression and reconstruction. Convolutional autoencoders (CAEs) and variational autoencoders (VAEs) have been particularly effective for guided wave applications. However, current methods can still struggle to maintain accurate state estimation under changing EOCs, and they are often limited to a single task. In response, the proposed framework adopts a two-fold strategy: it compresses high-dimensional signals into lower-dimensional representations and then leverages those representations to both estimate structural states and reconstruct the original data, even as conditions vary. Applied to two real-world SHM use-cases, this integrated method has proven its ability to preserve and retrieve key damage signatures under noise, shifting operational parameters, and other complicating factors.

  PublisherOA PDFDOI

• Neural operators for stochastic modeling of nonlinear structural system response to natural hazards

  Engineering Structures · 2025-09-26 · 4 citations

  article
  PublisherDOI

• Unified Framework for Probabilistic Modeling and Uncertainty Quantification of Aerospace Structures via Stochastic Latent Space Representations

  2025-01-03

  article

  Structural Health Monitoring (SHM) is a critical area for ensuring the reliability of engineering systems, with ultrasonic waves emerging as a preferred tool due to their ability to propagate over long distances and their sensitivity to structural changes. However, the inherent complexities of guided wave propagation, including frequency-dependent velocities, multimodal dispersion, and overlapping reflections, necessitate advanced feature extraction techniques to enable accurate damage detection. Conventional methods, such as Damage Index (DI), are effective for identifying anomalies but struggle to maintain sensitivity and stability under varying environmental and operational conditions (EOCs). Recent advancements in computational power have significantly advanced the adoption of machine learning techniques, particularly convolutional neural networks (CNNs) and autoencoders (AEs), in Structural Health Monitoring (SHM) applications. Convolutional autoencoders (CAEs), which leverage the strengths of both CNNs and AEs, are particularly effective for efficiently compressing and reconstructing data in system identification tasks. Alternatively, nonlinear data compression techniques, such as diffusion maps (DMaps), can also be utilized within state estimation schemes to address nonlinearity and high-dimensionality challenges. This study proposes an innovative framework that integrates these two distinct approaches - CAEs and DMaps - for data compression and reconstruction in SHM, particularly under varying EOCs. Stochastic signals are compressed into a low-dimensional latent space, significantly reducing computational costs. The first approach leverages diffusion maps for nonlinear dimensionality reduction, followed by polynomial chaos expansion (PCE) and Laplacian pyramids for signal reconstruction. The second approach employs an encoder-decoder architecture for signal processing. By comparing the performance and robustness of these methods across two distinct datasets, this work demonstrates their potential for achieving reliable state prediction and signal reconstruction, thereby advancing SHM capabilities under complex operational conditions.

  PublisherDOI

• Generative Learning of Densities on Manifolds

  ArXiv.org · 2025-03-05

  preprintOpen access1st authorCorresponding

  A generative modeling framework is proposed that combines diffusion models and manifold learning to efficiently sample data densities on manifolds. The approach utilizes Diffusion Maps to uncover possible low-dimensional underlying (latent) spaces in the high-dimensional data (ambient) space. Two approaches for sampling from the latent data density are described. The first is a score-based diffusion model, which is trained to map a standard normal distribution to the latent data distribution using a neural network. The second one involves solving an Itô stochastic differential equation in the latent space. Additional realizations of the data are generated by lifting the samples back to the ambient space using Double Diffusion Maps, a recently introduced technique typically employed in studying dynamical system reduction; here the focus lies in sampling densities rather than system dynamics. The proposed approaches enable sampling high dimensional data densities restricted to low-dimensional, a priori unknown manifolds. The efficacy of the proposed framework is demonstrated through a benchmark problem and a material with multiscale structure.

  PublisherOA PDFDOI

• Neural Operators for Stochastic Modeling of Nonlinear Structural System Response to Natural Hazards

  arXiv (Cornell University) · 2025-02-16

  preprintOpen access

  Traditionally, neural networks have been employed to learn the mapping between finite-dimensional Euclidean spaces. However, recent research has opened up new horizons, focusing on the utilization of deep neural networks to learn operators capable of mapping infinite-dimensional function spaces. In this work, we employ two state-of-the-art neural operators, the deep operator network (DeepONet) and the Fourier neural operator (FNO) for the prediction of the nonlinear time history response of structural systems exposed to natural hazards, such as earthquakes and wind. Specifically, we propose two architectures, a self-adaptive FNO and a Fast Fourier Transform-based DeepONet (DeepFNOnet), where we employ a FNO beyond the DeepONet to learn the discrepancy between the ground truth and the solution predicted by the DeepONet. To demonstrate the efficiency and applicability of the architectures, two problems are considered. In the first, we use the proposed model to predict the seismic nonlinear dynamic response of a six-story shear building subject to stochastic ground motions. In the second problem, we employ the operators to predict the wind-induced nonlinear dynamic response of a high-rise building while explicitly accounting for the stochastic nature of the wind excitation. In both cases, the trained metamodels achieve high accuracy while being orders of magnitude faster than their corresponding high-fidelity models.

  PublisherOA PDFDOI

• Data-driven inverse uncertainty quantification: application to the Chemical Vapor Deposition Reactor Modeling

  ArXiv.org · 2025-12-15

  preprintOpen access

  This study presents a Bayesian framework for (inverse) uncertainty quantification and parameter estimation in a two-step Chemical Vapor Deposition coating process using production data. We develop an XGBoost surrogate model that maps reactor setup parameters to coating thickness measurements, enabling efficient Bayesian analysis while reducing sampling costs. The methodology handles a mixture of data including continuous, discrete integer, binary, and encoded categorical variables. We establish parameter prior distributions through Bayesian Model Selection and perform Inverse Uncertainty Quantification via weighted Approximate Bayesian Computation with summary statistics, providing robust parameter credible intervals while filtering measurement noise across multiple reactor locations. Furthermore, we employ clustering methods guided by geometry embeddings to focus analysis within homogeneous production groups. This integrated approach provides a validated tool for improving industrial process control under uncertainty.

  PublisherOA PDFDOI

Frequent coauthors

• Michael D. Shields

  23 shared

• Vissarion Papadopoulos

  National Technical University of Athens

  20 shared

• Dimitrios Loukrezis

  9 shared

• Ioannis G. Kevrekidis

  Johns Hopkins University

  8 shared

• Katiana Kontolati

  Johns Hopkins University

  6 shared

• Manolis Papadrakakis

  National Technical University of Athens

  6 shared

• Paris Papavasileiou

  6 shared

• George Stavroulakis

  National Technical University of Athens

  5 shared

Education

• Phd, Civil Engineering

  National Technical University of Athens

Awards & honors

• DARPA INTACT Grant (2025)
• Richard J. Carroll Memorial Lecture
• Ross B.. Corotis Lecture