“Unpaired Image-to-Image Translation using Cycle-Consistent Adversarial Networks”の学習報告

Uncertainty quantification (UQ) plays a pivotal role in reduction of uncertainties during both optimization and decision making processes. It can be applied to solve a variety of real-world applications in science and engineering. Bayesian approximation and ensemble learning techniques are two most widely-used UQ methods in the literature. In this regard, researchers have proposed different UQ methods and examined their performance in a variety of applications such as computer vision (e.g., self-driving cars and object detection), image processing (e.g., image restoration), medical image analysis (e.g., medical image classification and segmentation), natural language processing (e.g., text classification, social media texts and recidivism risk-scoring), bioinformatics, etc. This study reviews recent advances in UQ methods used in deep learning. Moreover, we also investigate the application of these methods in reinforcement learning (RL). Then, we outline a few important applications of UQ methods. Finally, we briefly highlight the fundamental research challenges faced by UQ methods and discuss the future research directions in this field.

A Review of Uncertainty Quantification in Deep Learning: Techniques, Applications and Challenges

Computational science is paramount to the understanding of underlying processes in internal combustion engines of the future that will utilize non-petroleum-based alternative fuels, including carbon-neutral biofuels, and burn in new combustion regimes that will attain high efficiency while minimizing emissions of particulates and nitrogen oxides. Next-generation engines will likely operate at higher pressures, with greater amounts of dilution and utilize alternative fuels that exhibit a wide range of chemical and physical properties. Therefore, there is a significant role for high-fidelity simulations, direct numerical simulations (DNS), specifically designed to capture key turbulence-chemistry interactions in these relatively uncharted combustion regimes, and in particular, that can discriminate the effects of differences in fuel properties. In DNS, all of the relevant turbulence and flame scales are resolved numerically using high-order accurate numerical algorithms. As a consequence terascale DNS are computationally intensive, require massive amounts of computing power and generate tens of terabytes of data. Recent results from terascale DNS of turbulent flames are presented here, illustrating its role in elucidating flame stabilization mechanisms in a lifted turbulent hydrogen/air jet flame in a hot air coflow, and the flame structure of a fuel-lean turbulent premixed jet flame. Computing at this scale requires close collaborations between computer and combustion scientists to provide optimized scaleable algorithms and software for terascale simulations, efficient collective parallel I/O, tools for volume visualization of multiscale, multivariate data and automating the combustion workflow. The enabling computer science, applied to combustion science, is also required in many other terascale physics and engineering simulations. In particular, performance monitoring is used to identify the performance of key kernels in the DNS code, S3D and especially memory intensive loops in the code. Through the careful application of loop transformations, data reuse in cache is exploited thereby reducing memory bandwidth needs, and hence, improving S3D's nodal performance. To enhance collective parallel I/O in S3D, an MPI-I/O caching design is used to construct a two-stage write-behind method for improving the performance of write-only operations. The simulations generate tens of terabytes of data requiring analysis. Interactive exploration of the simulation data is enabled by multivariate time-varying volume visualization. The visualization highlights spatial and temporal correlations between multiple reactive scalar fields using an intuitive user interface based on parallel coordinates and time histogram. Finally, an automated combustion workflow is designed using Kepler to manage large-scale data movement, data morphing, and archival and to provide a graphical display of run-time diagnostics.

https://iopscience.iop.org/article/10.1088/1749-4699/2/1/015001/pdf

Terascale direct numerical simulations of turbulent combustion using S3D

Computational science is paramount to the understanding of underlying processes in internal combustion engines of the future that will utilize non-petroleum-based alternative fuels, including carbon-neutral biofuels, and burn in new combustion regimes that will attain high efficiency while minimizing emissions of particulates and nitrogen oxides. Next-generation engines will likely operate at higher pressures, with greater amounts of dilution and utilize alternative fuels that exhibit a wide range of chemical and physical properties. Therefore, there is a significant role for high-fidelity simulations, direct numerical simulations (DNS), specifically designed to capture key turbulence-chemistry interactions in these relatively uncharted combustion regimes, and in particular, that can discriminate the effects of differences in fuel properties. In DNS, all of the relevant turbulence and flame scales are resolved numerically using high-order accurate numerical algorithms. As a consequence terascale DNS are computationally intensive, require massive amounts of computing power and generate tens of terabytes of data. Recent results from terascale DNS of turbulent flames are presented here, illustrating its role in elucidating flame stabilization mechanisms in a lifted turbulent hydrogen/air jet flame in a hot air co-flow, and the flame structure of a fuel-lean turbulent premixed jet flame. Computing at this scale requires close collaborations betweenmore » computer and combustion scientists to provide optimized scaleable algorithms and software for terascale simulations, efficient collective parallel I/O, tools for volume visualization of multiscale, multivariate data and automating the combustion workflow. The enabling computer science, applied to combustion science, is also required in many other terascale physics and engineering simulations. In particular, performance monitoring is used to identify the performance of key kernels in the DNS code, S3D and especially memory intensive loops in the code. Through the careful application of loop transformations, data reuse in cache is exploited thereby reducing memory bandwidth needs, and hence, improving S3D's nodal performance. To enhance collective parallel I/O in S3D, an MPI-I/O caching design is used to construct a two-stage write-behind method for improving the performance of write-only operations. The simulations generate tens of terabytes of data requiring analysis. Interactive exploration of the simulation data is enabled by multivariate time-varying volume visualization. The visualization highlights spatial and temporal correlations between multiple reactive scalar fields using an intuitive user interface based on parallel coordinates and time histogram. Finally, an automated combustion workflow is designed using Kepler to manage large-scale data movement, data morphing, and archival and to provide a graphical display of run-time diagnostics.« less

Terascale direct numerical simulations of turbulent combustion using S3D.

Embracing imperfect datasets: A review of deep learning solutions for medical image segmentation

As scientific supercomputing moves toward petascale and exascale levels, in situ visualization stands out as a scalable way for scientists to view the data their simulations generate. This full picture is crucial particularly for capturing and understanding highly intermittent transient phenomena, such as ignition and extinction events in turbulent combustion.

In Situ Visualization for Large-Scale Combustion Simulations

The ability to identify and present the most essential aspects of time-varying data is critically important in many areas of science and engineering. This paper introduces an importance-driven approach to time-varying volume data visualization for enhancing that ability. By conducting a block-wise analysis of the data in the joint feature-temporal space, we derive an importance curve for each data block based on the formulation of conditional entropy from information theory. Each curve characterizes the local temporal behavior of the respective block, and clustering the importance curves of all the volume blocks effectively classifies the underlying data. Based on different temporal trends exhibited by importance curves and their clustering results, we suggest several interesting and effective visualization techniques to reveal the important aspects of time-varying data.

/pdf/importance-driven-time-varying-data-visualization-5bb2wmvsq0.pdf

Importance-Driven Time-Varying Data Visualization

In recent years, there is an emerging direction that leverages information theory to solve many challenging problems in scientific data analysis and visualization. In this article, we review the key concepts in information theory, discuss how the principles of information theory can be useful for visualization, and provide specific examples to draw connections between data communication and data visualization in terms of how information can be measured quantitatively. As the amount of digital data available to us increases at an astounding speed, the goal of this article is to introduce the interested readers to this new direction of data analysis research, and to inspire them to identify new applications and seek solutions using information theory.

https://www.mdpi.com/1099-4300/13/1/254/pdf

Information Theory in Scientific Visualization

We present an alternative method for viewing time-varying volumetric data. We consider such data as a four-dimensional data field, rather than considering space and time as separate entities. If we treat the data in this manner, we can apply high dimensional slicing and projection techniques to generate an image hyperplane. The user is provided with an intuitive user interface to specify arbitrary hyperplanes in 4D, which can be displayed with standard volume rendering techniques. From the volume specification, we are able to extract arbitrary hyperslices, combine slices together into a hyperprojection volume, or apply a 4D raycasting method to generate the same results. In combination with appropriate integration operators and transfer functions, we are able to extract and present different space-time features to the user.

/pdf/high-dimensional-direct-rendering-of-time-varying-volumetric-2ri29txzr0.pdf

High dimensional direct rendering of time-varying volumetric data

The growing power of parallel supercomputers gives scientists the ability to simulate more complex problems at higher fidelity, leading to many high-impact scientific advances. To maximize the utilization of the vast amount of data generated by these simulations, scientists also need scalable solutions for studying their data to different extents and at different abstraction levels. As we move into peta- and exa-scale computing, simply dumping as much raw simulation data as the storage capacity allows for post-processing analysis and visualization is no longer a viable approach. A common practice is to use a separate parallel computer to prepare data for subsequent analysis and visualization. A naive realization of this strategy not only limits the amount of data that can be saved, but also turns I/O into a performance bottleneck when using a large parallel system. We conjecture that the most plausible solution for the peta- and exa-scale data problem is to reduce or transform the data in-situ as it is being generated, so the amount of data that must be transferred over the network is kept to a minimum. In this paper, we discuss different approaches to in-situ processing and visualization as well as the results of our preliminary study using large-scale simulation codes on massively parallel supercomputers.

/pdf/in-situ-processing-and-visualization-for-ultrascale-n6h9v00zq3.pdf

Chaoli Wang

Papers

In Situ Visualization for Large-Scale Combustion Simulations

Importance-Driven Time-Varying Data Visualization

Information Theory in Scientific Visualization

High dimensional direct rendering of time-varying volumetric data

In-situ processing and visualization for ultrascale simulations