Biography


I am now a Research Staff in Data Analysis and Machine Learning Group, Computer Science and Mathematics Division at Oak Ridge National Laboratory (ORNL). Prior to being a staff scientist, I was a postdoctoral fellow in the National Center for Computational Sciences at ORNL.

My current research interest is on Artificial Intelligence for Robust Engineering and Scientific Discovery (AIRESD), broadly involving scientific machine learning, uncertainty quantification, inverse problems, generative design and optimization with applications to computational mechanics, advanced materials design, and additive manufacturing systems.

Interests

  • Robust and trustworthy AI and ML
  • Uncertainty quantification
  • Probabilsitic generative models
  • Inverse problems
  • Physics-informed deep learning
  • Anomaly detection
  • Computational materials design
  • High-performance computing

Education

  • PhD in Johns Hopkins University, 2018

    Civil and Systems Engineering (Advisor - Prof. Michael Shields)

  • MSE in Johns Hopkins University, 2018

    Applied Mathematics & Statistics (Advisor - Prof. John Wierman)

  • MSE in Dalian University of Technology (China), 2014

    Computational Mechanics (Advisor - Prof. Gengdong Cheng)

  • BS in Dalian University of Technology (China), 2011

    Engineering Mechanics (Advisor - Prof. Bo Wang)

Recent News

  • [2021/10] Our paper titled On the Stochastic Stability of Deep Markov Models has been accepted by NeurIPS 2021.

  • [2021/10] Our paper titled Inverse design of two-dimensional materials with invertible neural networks has been accepted by npj Computational Materials - Nature

  • [2021/10] Our paper titled Self-Supervised Anomaly Detection via Neural Autoregressive Flows with Active Learning has been accepted by NeurIPS 2021 workshop on Deep Generative Models and Downstream Applications

  • [2021/9] Our LDRD proposal Reinforced Adversarial Learning for Graph Generation and Goal‑Oriented Design and Deep Reinforcement Learning for Inconsistently Sized Action‑Spaces are both funded by ORNL LDRD office and $250k for each one!

  • [2021/9] Our paper titled Byzantine-robust Federated Learning through Spatial-temporal Analysis of Local Model Updates has been accepted by the 27th IEEE International Conference on Parallel and Distributed Systems (ICPADS 2021)

  • [2021/8] I gave an invited talk titled Probabilistic self‑supervised anomaly detection with normalizing flows Department of Industrial Systems Engineering, The University of Tennessee, Knoxville

  • [2021/7] I gave an invited talk titled Uncertainty‑Aware Inverse Learning with Normalizing Flows to the Probabilistic Seminar Series, GE Global Research

  • [2021/7] I gave an invited keynote presentation titled Efficient Inverse Problem Learning with Precise Localization and Exploratory Sampling at the 16th U.S. National Congress on Computational Mechanics (USNCCM 16), Jul. 2021

  • [2021/6] I gave a talk titled Uncertainty-Aware Inverse Learning using Normalizing Flows to DOE ASCR Dn2CS project.

  • [2021/6] I am honored to give a one-hour talk about my recent work Uncertainty-Aware Inverse Learning using Generative Flows, invited by U.S. Association for Computational Mechanics USACM, Uncertainty Quantification (UQ) and Probabilistic Modeling Technical Thrust Aera USACM TT-UQ and the talk video is available at: https://lnkd.in/exbHm5n.

  • [2021/6] Our paper on transfer learning based variable-fidelity surrogate models was published in Composite Structures.

  • [2021/6] Our benchmarking paper on graph neural networks (GNN) for materials chemistry was published in npj Computational Materials.

  • [2021/5] I gave an invited talk titled " Uncertainty Quantification in Deep Learning: A Bayesian Approach and Normalizing Flow” to IDEAL Group led by Prof. Wei Chen in Department of Mechanical Engineering at Northwestern University.

  • [2021/5] I organized a minisymposium on “Machine Learning for Solving Inverse Problems in Computational Chemistry and Materials Science” at SIAM Conference on Mathematical Aspects of Materials Science (SIAM MS21)

  • [2021/5] Our paper on directional Gaussian smoothing for high-dimensional blackbox optimization was accepted by UAI 2021. The arxiv link is https://arxiv.org/abs/2002.03001

  • [2021/4] I presented our two papers: “Efficient inverse learning for materials design and discovery” and “Towards efficient uncertainty estimation in deep learning for robust energy prediction in crystal materials” at ICLR 2021 Workshop on Science and Engineering of Deep Learning, and Deep Learning for Simulation respectively.

Publications

Journal & Conference Papers

📝 AI & ML Conference Papers


  1. Ján Drgona, Sayak Mukherjee, Jiaxin Zhang, Frank Liu, Mahantesh Halappanavar. On the Stochastic Stability of Deep Markov Models. In Proc. 35th Conference on Neural Information Processing Systems (NeurIPS) (NeurIPS 2021). [PDF]

  2. Jiaxin Zhang, Kyle Saleeby, Thomas Feldhausen, Sirui Bi, Alex Plotkowski, David Womble. Self-Supervised Anomaly Detection via Neural Autoregressive Flows with Active Learning. In NeurIPS 2021 Workshop Deep Generative Models and Downstream Applications (NeurIPS 2021 Workshop). [PDF]

  3. Jiaxin Zhang, Hoang Tran, Dan Lu, Guannan Zhang. Enabling Long-range Exploration in Minimization of Multimodal Functions. In Proc. 37th Conference on Uncertainty in Artificial Intelligence (UAI 2021). [PDF]

  1. Jiaxin Zhang, Sirui Bi, Guannan Zhang. A Scalable Gradient Free Method for Bayesian Experimental Design with Implicit Models. In Proc. 24th International Conference on Artificial Intelligence and Statistics (AISTATS 2021). [PDF]

  2. Jiaxin Zhang, Jan Drgona, Sayak Mukherjee, Mahantesh Halappanavar, Frank Liu. Variational Generative Flows for Reconstruction Uncertainty Estimation. In ICML 2021 Workshop on Uncertainty & Robustness in Deep Learning (ICML 2021 Workshop). [PDF]

  3. Zhuohang Li, Luyang Liu, Jiaxin Zhang, Jian Liu. Byzantine-robust Federated Learning through Spatial-temporal Analysis of Local Model Updates. The 27th IEEE International Conference on Parallel and Distributed Systems (ICPADS 2021). [PDF]

  4. Jingbo Sun, Li Yang, Jiaxin Zhang, Frank Liu, Mahantesh Halappanavar, Deliang Fan, Yu Cao. Self-supervised Novelty Detection for Continual Learning: A Gradient-based Approach Boosted by Binary Classification. In IJCAI 2021 Workshop on Continual Semi-Supervised Learning (IJCAI 2021 Workshop). [PDF]

  5. Jiaxin Zhang, Victor Fung. Efficient Inverse Learning for Materials Design and Discovery. In ICLR 2021 Workshop on Science and Engineering of Deep Learning (ICLR 2021 Workshop). [PDF]

  6. Jiaxin Zhang, Sirui Bi, Victor Fung, Guannan Zhang. Towards Efficient Uncertainty Estimation in Deep Learning for Robust Energy Prediction in Crystal Materials. In ICLR 2021 Workshop on Deep Learning for Simulation (ICLR 2021 Workshop). [PDF]

  7. Jiaxin Zhang, Congjie Wei, Chenglin Wu. Thermodynamic Consistent Neural Networks for Learning Material Interfacial Mechanics. In NeurIPS 2020 Workshop on Interpretable Inductive Biases and Physically Structured Learning (NeurIPS 2020 Workshop). [PDF]

  8. Jiaxin Zhang, Sirui Bi, Guannan Zhang. A hybrid gradient method to designing Bayesian experiments for implicit models. In NeurIPS 2020 Workshop on Machine Learning and the Physical Sciences (NeurIPS 2020 Workshop). [PDF]

  9. Jiaxin Zhang, Sirui Bi, Guannan Zhang. Scalable deep-learning-accelerated topology optimization for additively manufactured materials. In NeurIPS 2020 Workshop on Machine Learning for Engineering Modeling, Simulation and Design (NeurIPS 2020 Workshop). [PDF]

  10. Jiaxin Zhang, Sirui Bi, Guannan Zhang. A nonlocal-gradient descent method for inverse design in nanophotonics. In NeurIPS 2020 Workshop on Machine Learning for Engineering Modeling, Simulation and Design (NeurIPS 2020 Workshop). [PDF]

  11. Guannan Zhang, Jiaxin Zhang, Jacob Hinkle. Learning nonlinear level sets for dimensionality reduction in function approximation. In Proc. 33rd Annual Conference on Neural Information Processing Systems (NeurIPS 2019). [PDF]

📝 Journal Papers


  1. Xianglin Liu, Jiaxin Zhang, Zongrui Pei. Machine Learning for High Entropy Alloys: Progress, Challenges and Opportunities. In Progress in Materials Science, 2021 (accepted proposal , full paper in review).

  2. Congjie Wei, Jiaxin Zhang, Kenneth M Liechti, Chenglin Wu. Deep-Green Inversion (DGI) to Extract Traction Separation Relationship at Material Interfaces. In International Journal of Solids and Structures, 2021 (in review).

  3. Congjie Wei, Jiaxin Zhang, Kenneth M Liechti, Chenglin Wu. Data Driven Modeling of Interfacial Traction Separation Relations using Thermodynamic Consistent Neural Network (TCNN). In Computer Methods in Applied Mechanics and Engineering, 2021 (in review).

  4. Sirui Bi, Benjamin Stump, Jiaxin Zhang, Yousub Lee, John Coleman, Matt Bement, Guannan Zhang. Blackbox Optimization for High-fidelity Heat Conduction Model Approximation in Metal Additive Manufacturing. In Materials Today Communications, 2021 (in review).

  5. Victor Fung, Jiaxin Zhang, Guoxiang Hu, P Ganesh, Bobby G Sumpter. Inverse design of two-dimensional materials with invertible neural networks. In npj Computational Materials, 2021 (accepted).

  6. Victor Fung, Jiaxin Zhang, Eric Juarez, Bobby Sumpter. Benchmarking graph neural networks for materials chemistry. In npj Computational Materials 7, 84, 2021. [PDF]

  7. Jiaxin Zhang, Hoang Tran, Guannan Zhang. Accelerating reinforcement learning with a Directional-Gaussian-Smoothing evolution strategy. In Electronic Research Archive, 2021. [PDF]

  8. Kuo Tian, Zengcong Li, Jiaxin Zhang, Lei Huang, Bo Wang. Transfer learning based variable-fidelity surrogate model for shell buckling prediction. In Composite Structures, 114285, 2021. [PDF]

  9. Jiaxin Zhang, Sirui Bi, Guannan Zhang. A directional Gaussian smoothing optimization method for computational inverse design in nanophotonics. In Materials & Design 197, 109213, 2021. [PDF]

  10. Xianglin Liu, Jiaxin Zhang, Junqi Yin, Sirui Bi, Markus Eisenbach, Yang Wang. Monte Carlo simulation of order-disorder transition in refractory high entropy alloys: a data-driven approach. In Computational Materials Science 187, 110135, 2021. [PDF]

  11. Jiaxin Zhang. Modern Monte Carlo methods for efficient uncertainty quantification and propagation: A survey. In Wiley Interdisciplinary Reviews: Computational Statistics e1539, 2020. [PDF]

  12. Jiaxin Zhang, Xianglin Liu, Sirui Bi, Junqi Yin, Guannan Zhang, Markus Eisenbach. Robust data-driven approach for predicting the configurational energy of high entropy alloys. In Materials & Design 185, 108247, 2020. [PDF]

  13. Jiaxin Zhang, Stephanie TerMaath, Michael Shields. Imprecise global sensitivity analysis using bayesian multimodel inference and importance sampling. In Mechanical Systems and Signal Processing 148, 107162, 2020. [PDF]

  14. Jiaxin Zhang, Michael Shields. On the quantification and efficient propagation of imprecise probabilities with copula dependence. In International Journal of Approximate Reasoning 122, 24-46, 2020. [PDF]

  15. Jiaxin Zhang, Michael Shields, Stephanie TerMaath. Probabilistic modeling and prediction of out-of-plane unidirectional composite lamina properties. In Mechanics of Advanced Materials and Structures 1-17, 2020. [PDF]

  16. Massimiliano Lupo Pasini, Ying Wai Li, Junqi Yin, Jiaxin Zhang, Kipton Barros, Markus Eisenbach. Fast and stable deep-learning predictions of material properties for solid solution alloys. In Journal of Physics: Condensed Matter 33, 084005, 2020. [PDF]

  17. Kuo Tian, Zengcong Li, Xiangtao Ma, Haixin Zhao, Jiaxin Zhang, Bo Wang. Toward the robust establishment of variable-fidelity surrogate models for hierarchical stiffened shells by two-step adaptive updating approach. In Structural and Multidisciplinary Optimization 61, 1515–1528, 2020. [PDF]

  18. Kuo Tian, Jiaxin Zhang, Xiangtao Ma, Yuwei Li, Yu Sun, Peng Hao. Buckling surrogate-based optimization framework for hierarchical stiffened composite shells by enhanced variance reduction method. In Journal of Reinforced Plastics and Composites 38, 959–973, 2019. [PDF]

  19. Jiaxin Zhang, Michael Shields. Efficient Monte Carlo resampling for probability measure changes from Bayesian updating. In Probabilistic Engineering Mechanics 55, 54–66, 2019. [PDF]

  20. Jiaxin Zhang, Michael Shields. The effect of prior probabilities on quantification and propagation of imprecise probabilities resulting from small datasets. In Computer Methods in Applied Mechanics and Engineering 334, 483–506, 2018. [PDF]

  21. Jiaxin Zhang, Michael Shields. On the quantification and efficient propagation of imprecise probabilities resulting from small datasets. In Mechanical Systems and Signal Processing 98, 465–483, 2018. [PDF]

  22. Kuo Tian, Bo Wang, Ke Zhang, Jiaxin Zhang, Peng Hao, Ying Wu. Tailoring the optimal load-carrying efficiency of hierarchical stiffened shells by competitive sampling. In Thin-Walled Structures 133, 216–225, 2018. [PDF]

  23. Michael Shields, Jiaxin Zhang. The generalization of Latin hypercube sampling. In Reliability Engineering & System Safety 148, 96–108, 2016. [PDF]

  24. Jiaxin Zhang, Bo Wang, Fei Niu, Gengdong Cheng. Design Optimization of Connection Section for Concentrated Force Diffusion. In Mechanics Based Design of Structures and Machines 43, 209-231, 2015. [PDF]

  25. Bo Wang, Peng Hao, Gang Li, Jiaxin Zhang, Kaifan Du, Kuo Tian, Xiaojun Wang, Xiaohan Tang. Optimum design of hierarchical stiffened shells for low imperfection sensitivity. In Acta Mechanica Sinica 30, 391-402, 2014. [PDF]

  26. Sirui Bi, Jiaxin Zhang, Qianjin Yue. Buckling control and energy absorption of corrugated web beam under axial compression. In Chinese Computer Aided Engineering 23, 79-85, 2014. [PDF]

  27. Jiaxin Zhang, Bo Wang, Fei Niu, Gengdong Cheng. Optimal design of concentrated force diffusion for short shell structure using hierarchical radial ribs. In Chinese J Comput Mech 31, 141-148, 2014. [PDF]

📝 Other Conference Papers


  1. Velkur Sundar, Jiaxin Zhang, Dimitris Giovanis and Michael Shields. Conditional sampling using affine invariantensemble MCMC: application to subset simulation. In Proc. IFIP WG-7.5 Reliability and Optimization of Structure Systems. [PDF]

  2. Bangalore Aakash, Jiaxin Zhang, Pawel Woelke and Michael Shield. Probabilistic calibration of material models from limited data and its influence on structural response. In Proc. 12th International Conference on StructuralSafety & Reliability (ICOSSAR 2017). [PDF]

  3. Jiaxin Zhang, Michael Shields. Efficient propagation of imprecise probabilities. In Proc. 7th International Workshop on Reliable Engineering Computing (REC 2016). [PDF]

📝 Ph.D. Thesis


  1. Jiaxin Zhang. Uncertainty Quantification From Small Data: A Multimodel Approach. Johns Hopkins University, 2018.

Funded Projects

Current Grants

Roles: Principal Investigator (PI)/Co-PI/Task Lead/Senior Personnel

Research

Research Overview and Highlights

Research Overview

The overall objective is to develop foundational methods for efficient and robust learning, design and decision-making in complex science and engineering problems.

drawing


Accelerated Inverse Learning with Invertible Neural Networks [2021-now]

We propose a novel approach leveraging recent advances in deep invertible models incorporated with a precise localization via gradient descent for efficiently and accurately solving inverse problems and apply to advanced materials design and discovery.

drawing

[1] Jiaxin Zhang, Victor Fung. Efficient Inverse Learning for Materials Design and Discovery. In ICLR 2021 Workshop on Science and Engineering of Deep Learning.

[2] Victor Fung, Jiaxin Zhang, Guoxiang Hu, P Ganesh, Bobby G Sumpter. Inverse design of two-dimensional materials with invertible neural networks. In Nature Communications, 2021 (in review).


Uncertainty-Aware Inverse Learning with Normalizing Flows [2021-now]

We propose a deep variational framework that leverages a generative flow to learn an approximate posterior distribution for UQ. To perform accurate uncertainty estimation, we propose a robust flow-based model where the stability is enhanced by adding bi-directional regularization and the flexibility is improved by using gradient boosting. We demonstrate our method on several benchmark tasks and two real-world applications (FastMRI and black hole image reconstruction) and show that it achieves a reliable and high-quality reconstruction with accurate uncertainty estimation.

drawing

[1] Jiaxin Zhang, Jan Drgona, Sayak Mukherjee, Mahantesh Halappanavar, Frank Liu. Variational Generative Flows for Reconstruction Uncertainty Estimation. In ICML 2021 Workshop on Uncertainty & Robustness in Deep Learning.


Graph Neural Networks for Materials Chemistry [2021-now]

We present a workflow and testing platform, MatDeepLearn Github repository, for quickly and reproducibly assessing and comparing GNNs and other machine learning models, and we use this platform to optimize and evaluate a selection of top performing GNNs on several representative datasets in computational materials chemistry.

drawing

[1] Victor Fung, Jiaxin Zhang, Eric Juarez, Bobby Sumpter. Benchmarking graph neural networks for materials chemistry. In npj Computational Materials 7, 84, 2021.


Self-Supervised Anomaly Detection [2021-now]

Anomaly Detection aims to automatically identify out-of-distribution (OOD) data, without any prior knowledge of them. It is a critical step in data monitoring, behavior analysis and other applications, helping enable continual learning in the field. Conventional methods of OOD detection perform multi-variate analysis on an ensemble of data or features, and usually resort to the supervision with OOD data to improve the accuracy. In reality, such supervision is impractical as one cannot anticipate the anomalous data.

We propose a novel, self-supervised approach that does not rely on any pre-defined OOD data, and achieves higher accuracy than previous supervised methods across all benchmarks. We perform a comprehensive evaluation of the proposed method across multiples datasets, namely, CIFAR-10, CIFAR-100, and ImageNet. The proposed approach consistently outperforms state-of-the-art supervised and unsupervised methods in the area under the receiver operating characteristic (AUROC) and area under the precision-recall curve (AUPR) metrics.

drawing

[1] Jingbo Sun, Li Yang, Jiaxin Zhang, Frank Liu, Mahantesh Halappanavar, Deliang Fan, Yu Cao. Self-supervised Novelty Detection for Continual Learning: A Gradient-based Approach Boosted by Binary Classification. In IJCAI 2021 Workshop on Continual Semi-Supervised Learning.

[2] Jingbo Sun, Li Yang, Jiaxin Zhang, Frank Liu, Mahantesh Halappanavar, Deliang Fan, Yu Cao. Gradient-based Novelty Detection Boosted bySelf-supervised Binary Classification. In NeurIPS 2021 (in review).


Stochastic Gradient Free Bayesian Experimental Design [2020-2021]

Bayesian experimental design (BED) is to answer the question that how to choose designs that maximize the information gathering. For implicit models, where the likelihood is intractable but sampling is possible, conventional BED methods have difficulties in efficiently estimating the posterior distribution and maximizing the mutual information (MI) between data and parameters. We propose a novel approach that leverages recent advances in stochastic approximate gradient ascent incorporated with a smoothed variational MI estimator for efficient and robust BED. Without the necessity of pathwise gradients, our approach allows the design process to be achieved through a unified procedure with an approximate gradient for implicit models. Several experiments show that our approach outperforms baseline methods, and significantly improves the scalability of BED in high-dimensional problems

drawing

[1] Jiaxin Zhang, Sirui Bi, Guannan Zhang. A Scalable Gradient Free Method for Bayesian Experimental Design with Implicit Models. In Proc. 24th International Conference on Artificial Intelligence and Statistics (AISTATS 2021).

[2] Jiaxin Zhang, Sirui Bi, Guannan Zhang. A hybrid gradient method to designing Bayesian experiments for implicit models. In NeurIPS 2020 Workshop on Machine Learning and the Physical Sciences.


Directional Gaussian Smoothing for High-Dimensional Optimization [2020-2021]

We develop a nonlocal gradient operator, Directional Gaussian smoothing (DGS), to skip small local optima and capture major structures of the loss’s landscape in black-box optimization, specifically high-dimensional cases (e.g., scale to 2000D). We have successfully applied DGS for a couple of scientific applications, including advanced materials design, structural optimization, heat conduction model calibration in additive manufacturing.

drawing

[1] Jiaxin Zhang, Hoang Tran, Dan Lu, Guannan Zhang. Enabling Long-range Exploration in Minimization of Multimodal Functions. In Proc. 37th Conference on Uncertainty in Artificial Intelligence (UAI 2021).

[2] Jiaxin Zhang, Sirui Bi, Guannan Zhang. A directional Gaussian smoothing optimization method for computational inverse design in nanophotonics. In Materials & Design 197, 109213, 2021.

[3] Jiaxin Zhang, Sirui Bi, Guannan Zhang. A nonlocal-gradient descent method for inverse design in nanophotonics. In NeurIPS 2020 Workshop on Machine Learning for Engineering Modeling, Simulation and Design.


Scalable Deep-Learning-Accelerated Topology Optimization [2019-2020]

Topology optimization (TO) is a popular and powerful computational approach for designing novel structures, materials, and devices. Two computational challenges have limited the applicability of TO to a variety of industrial applications. First, a TO problem often involves a large number of design variables to guarantee sufficient expressive power. Second, many TO problems require a large number of expensive physical model simulations, and those simulations cannot be parallelized.

To address these issues, we propose a general scalable deep-learning (DL) based TO framework, referred to as SDL-TO, which utilizes parallel schemes in high performance computing (HPC) to accelerate the TO process for designing additively manufactured (AM) materials. Unlike the existing studies of DL for TO, our framework accelerates TO by learning the iterative history data and simultaneously training on the mapping between the given design and its gradient. The surrogate gradient is learned by utilizing parallel computing on multiple CPUs incorporated with a distributed DL training on multiple GPUs. The learned TO gradient enables a fast online update scheme instead of an expensive update based on the physical simulator or solver. Using a local sampling strategy, we achieve to reduce the intrinsic high dimensionality of the design space and improve the training accuracy and the scalability of the SDL-TO framework.

drawing

[1] Jiaxin Zhang, Sirui Bi, Guannan Zhang. Scalable deep-learning-accelerated topology optimization for additively manufactured materials. In NeurIPS 2020 Workshop on Machine Learning for Engineering Modeling, Simulation and Design.


Thermodynamics of High Entropy Alloys: A Data-Driven Approach [2019-2020]

High entropy alloys (HEAs) are promising next-generation materials due to their various excellent properties. To understand these properties, it’s necessary to characterize the chemical ordering and identify order-disorder transitions through efficient simulation and modeling of thermodynamics. In this study, a robust data-driven framework based on Bayesian approaches is proposed for the accurate and efficient prediction of configurational energy of high entropy alloys. The recently proposed effective pair interaction (EPI) model with ensemble sampling is used to map the configuration and its corresponding energy. Given limited data calculated by first-principles calculations, Bayesian regularized regression not only offers an accurate and stable prediction but also effectively quantifies the uncertainties associated with EPI parameters.

We further introduce a data-driven approach to construct the effective Hamiltonian and study the thermodynamics of HEAs through canonical Monte Carlo simulation. The main characteristic of our method is to use pairwise interactions between atoms as features and systematically improve the representativeness of the dataset using samples from Monte Carlo simulation. We find this method produces highly robust and accurate effective Hamiltonians that give less than 0.1 mRy test error for all the three refractory HEAs: MoNbTaW, MoNbTaVW, and MoNbTaTiW. Using replica exchange to speed up the MC simulation, we calculated the specific heats and short-range order parameters in a wide range of temperatures.

drawing

[1] Jiaxin Zhang, Xianglin Liu, Sirui Bi, Junqi Yin, Guannan Zhang, Markus Eisenbach. Robust data-driven approach for predicting the configurational energy of high entropy alloys. In Materials & Design 185, 108247, 2020.

[2] Xianglin Liu, Jiaxin Zhang, Junqi Yin, Sirui Bi, Markus Eisenbach, Yang Wang. Monte Carlo simulation of order-disorder transition in refractory high entropy alloys: a data-driven approach. In Computational Materials Science 187, 110135, 2021.


Uncertainty Quantification From Small Data: A MultiModel Approach [2014-2018]

As a central area of computational science and engineering (CSE), uncertainty quantification (UQ) is playing an increasingly important role in computationally evaluating the performance of complex mathematical, physical and engineering systems. UQ includes the quantification, integration, and propagation of uncertainties that result from stochastic variations in the natural world as well as uncertainties created by lack of statistical data or knowledge and uncertainty in the form of mathematical models. A common situation in engineering practice is to have a limited cost or time budget for data collection and thus to end up with sparse datasets. This leads to epistemic uncertainty (lack of knowledge) along with aleatory uncertainty (inherent randomness), and a mix of these two sources of uncertainties (requiring imprecise probabilities) is a particularly challenging problem.

A novel methodology is proposed for quantifying and propagating uncertainties created by lack of data. The methodology utilizes the concepts of multimodel inference from both information-theoretic and Bayesian perspectives to identify a set of candidate probability models and associated model probabilities that are representative of the given small dataset. Both model-form uncertainty and model parameter uncertainty are identified and estimated within the proposed methodology. Unlike the conventional method that reduces the full probabilistic description to a single probability model, the proposed methodology fully retains and propagates the total uncertainties quantified from all candidate models and their model parameters. This is achieved by identifying an optimal importance sampling density that best represents the full set of models, propagating this density and reweighting the samples drawn from the each of candidate probability model using Monte Carlo sampling. As a result, a complete probabilistic description of both aleatory and epistemic uncertainty is achieved with several orders of magnitude reduction in Monte Carlo-based computational cost.

Along with the proposed new UQ methodology, an investigation is provided to study the effect of prior probabilities on quantification and propagation of imprecise probabilities resulting from small datasets. It is illustrated that prior probabilities have a significant influence on Bayesian multimodel UQ for small datasets and inappropriate priors may introduce biased probabilities as well as inaccurate estimators even for large datasets. When a multidimensional UQ problem is involved, a further study generalizes this novel UQ methodology to overcome the limitations of the independence assumption by modeling the dependence structure using copula theory. The generalized approach achieves estimates for imprecise probabilities with copula dependence modeling for a composite material problem. Finally, as applications of the proposed method, an imprecise global sensitivity analysis is performed to illustrate the efficiency and effectiveness of the developed novel multimodel UQ methodology given small datasets.

drawing

[1] Jiaxin Zhang. Modern Monte Carlo methods for efficient uncertainty quantification and propagation: A survey. In Wiley Interdisciplinary Reviews: Computational Statistics e1539, 2020.

[2] Jiaxin Zhang, Stephanie TerMaath, Michael Shields. Imprecise global sensitivity analysis using bayesian multimodel inference and importance sampling. In Mechanical Systems and Signal Processing 148, 107162, 2020.

[3] Jiaxin Zhang, Michael Shields. On the quantification and efficient propagation of imprecise probabilities with copula dependence. In International Journal of Approximate Reasoning 122, 24-46, 2020.

[4] Jiaxin Zhang, Michael Shields, Stephanie TerMaath. Probabilistic modeling and prediction of out-of-plane unidirectional composite lamina properties. In Mechanics of Advanced Materials and Structures 1-17, 2020.

[5] Jiaxin Zhang, Michael Shields. Efficient Monte Carlo resampling for probability measure changes from Bayesian updating. In Probabilistic Engineering Mechanics 55, 54–66, 2019.

[6] Jiaxin Zhang, Michael Shields. The effect of prior probabilities on quantification and propagation of imprecise probabilities resulting from small datasets. In Computer Methods in Applied Mechanics and Engineering 334, 483–506, 2018.

[7] Jiaxin Zhang, Michael Shields. On the quantification and efficient propagation of imprecise probabilities resulting from small datasets. In Mechanical Systems and Signal Processing 98, 465–483, 2018.

[8] Michael Shields, Jiaxin Zhang. The generalization of Latin hypercube sampling. In Reliability Engineering & System Safety 148, 96–108, 2016.

Software

drawing

UQpy: Github repository

UQpy (Uncertainty Quantification with python) is an open-source python package for modeling uncertainty in physical and mathematical systems. Capabilities include sampling, statistical inference, surrogate modeling, sensitivity analysis, dimension reduction, reliability and more. The Shields Uncertainty Research Group at the Department of Civil and Systems Engineering at Johns Hopkins has developed and maintain UQpy.

Reference - Audrey Olivier, Dimitris G. Giovanis, B.S. Aakash, Mohit Chauhan, Lohit Vandanapu, Michael D. Shields. UQpy: A general purpose Python package and development environment for uncertainty quantification. In Journal of Computational Science 47, 101204, 2020.


MatDeepLearn: Github repository

MatDeepLearn is a platform for testing and using graph neural networks (GNNs) and other machine learning (ML) models for materials chemistry applications. MatDeepLearn takes in data in the form of atomic structures and their target properties, processes the data into graphs, trains the ML model of choice (optionally with hyperparameter optimization), and provides predictions on unseen data. It allows for different GNNs to be benchmarked on diverse datasets drawn from materials repositories as well as conventional training/prediction tasks.

Reference - Victor Fung, Jiaxin Zhang, Eric Juarez, Bobby Sumpter. Benchmarking graph neural networks for materials chemistry. In npj Computational Materials 7, 84, 2021.


MatDesINNe: Github repository

The ability to readily design novel materials with chosen functional properties on-demand represents a next frontier in materials discovery. However, thoroughly and efficiently sampling the entire design space in a computationally tractable manner remains a highly challenging task. To tackle this problem, we propose an inverse design framework (MatDesINNe) utilizing invertible neural networks which can map both forward and reverse processes between the design space and target property.

Reference - Victor Fung, Jiaxin Zhang, Guoxiang Hu, P Ganesh, Bobby G Sumpter. Inverse design of two-dimensional materials with invertible neural networks. In npj Computational Materials, 2021 (accepted).

Awards and Honors

  • Promising Early-Career Researcher Award, CSMD, ORNL, 2020
  • NeurIPS Travel Award, 33th Conference on Neural Information Processing Systems, 2019
  • LANL Travel Scholarship, Machine Learning for Computational Fluid and Solid Dynamics, 2019
  • NSF Travel Award, UQ in Computational Solid and Structural Materials Modeling, 2019
  • List of Most Cited Articles (ranked 6/50, since 2016), Reliability Engineering & System Safety, 2019
  • Chinese Government Award for Outstanding Self-financed Students Abroad, 2018.
  • Acheson J. Duncan Graduate Research Award, Johns Hopkins University, 2018
  • Foundation of North America Travel Award, Symposium on Data Science & Statistics, 2018
  • SIAM Travel Award, SIAM Conference on Uncertainty Quantification, 2018
  • Outstanding Contribution in Reviewing, Mechanical System and Signal Processing, 2018
  • USACM Travel Award, U.S. National Congress on Computational Mechanics, 2017
  • USACM Travel Award, Uncertainty Quantification and Data-Driven Modeling Workshop, 2017
  • Hopkins Extreme Materials Institute Graduate Travel Grant, Johns Hopkins University, 2016
  • Whiting School of Engineering Dean′s Fellowship, Johns Hopkins University, 2014-2015
  • Hopkins Extreme Materials Institute Research Grant, Johns Hopkin University, 2014-2015
  • Qian Ling-Xi Mechanics Scholarship, Dalian, China, 2013
  • Chinese National Scholarship, Ministry of Education of China, 2019, 2013

Talks and Presentations

Upcoming talks

  • [2021/5] Uncertainty Quantification in Deep Learning: A Bayesian Approach and Normalizing Flow. Department of Mechanical Engineering at Northwestern University.

Services

AI/ML Conference Reviewer

  • NeurIPS (2020, 2021), ICLR (2021,2022), ICML(2020,2021), AISTATS(2021, 2022)

Journal Reviewer

  • Journal of Machine Learning Research
  • International Journal of Approximate Reasoning
  • Computer Methods in Applied Mechanics and Engineering
  • Journal of Computational Physics
  • npj Computational Materials - Nature
  • International Journal of Uncertainty Quantification
  • Mechanical Systems and Signal Processing
  • Structural and Multidisciplinary Optimization
  • International Journal of Approximate Reasoning
  • Scientific Report - Nature
  • Computational Materials Science
  • Journal of Alloys and Compounds
  • Proceedings of the Royal Society A: Mathematical, Physical and Engineering Science
  • IEEE Signal Processing Letters
  • many more …

Minisymposium Organizer

Professional Membership

  • SIAM - Society for Industrial and Applied Mathematics
  • USACM - U.S. Association for Computational Mechanics
  • ISSMO - International Society for Structural and Multidisciplinary Optimization

Contact

  • zhangj@ornl.gov
  • Oak Ridge National Laboratory, Oak Ridge, Tennessee 37830