Despite the increasing importance of in silico experiments to the scientific discovery process, state-of-the-art software engineering practices are rarely adopted in computational science. To understand the underlying causes for this situation and to identify ways to improve it, the authors conducted a literature survey on software engineering practices in computational science. They identified 13 recurring key characteristics of scientific software development that are the result of the nature of scientific challenges, the limitations of computers, and the cultural environment of scientific software development. Their findings allow them to point out shortcomings of existing approaches for bridging the gap between software engineering and computational science and to provide an outlook on promising research directions that could contribute to improving the current situation.
Despite the effectiveness of convolutional neural networks (CNNs), especially for image classification tasks, the effect of convolution features on learned representations is still limited, mainly focusing on an images salient object but ignoring the variation information from clutter and local objects. The authors propose a multiple vector of locally aggregated descriptors (VLAD) encoding method with CNN features for image classification. To improve the VLAD coding methods performance, they explore the multiplicity of VLAD encoding with the extension of three encoding algorithms. Moreover, they equip the spatial pyramid patch (SPM) on VLAD encoding to add spatial information to CNN features. The addition of SPM, in particular, allows their proposed framework to yield better performance compared to the traditional method.
Interval sets and hesitant fuzzy sets capture two types of hesitant situations and provide novel expression formats for decision makers. The authors propose a new hesitant fuzzy interval set and explore its information aggregation for multicriteria decision-making. An example is presented to elaborate on the performance of the approach.
The reliability of services offered by intelligent transportation systems is attributed to the accuracy and timely availability of road-network traffic information. However, in the present era of big data, compliance with anticipated service quality requirements mandates consistent real-time processing of big spatiotemporal traffic data. Thus, development of low-dimensional models is a crucial challenge in traffic data processing. The authors developed such representations using data graph framework and graph Fourier transform (GFT) approaches. Experimental results on California daily network traffic data showed that, even with a 15:1 compression ratio, GFT-based models offered less than 6 percent reconstruction error (RE), instigating a less than 2 percent increase in mean absolute percentage error of corresponding predictions. The authors also proposed a 3D graph framework, which reduced RE by almost 2 percent compared to its 2D counterpart.
Distortions such as shape changes, ghosting, seam lines, and missing edges often occur in multiperspective image stitching. The authors describe a content-preserving image stitching and completion method based on hybrid structure warping to restrain these distortions. Experiments showed that the proposed approach significantly reduced ghosting and distortion, and when compared to existing methods, displayed a superior visual effect.
The study of intelligent information processing seeks to establish theories, algorithms, and systematic methods and technology for dealing with complex system information and its uncertainty. This issue highlights four aspects of intelligent information processing: multicriteria decision-making, image processing and classification, intelligent transportation, and energy management.
We deeply appreciate the efforts of everyone who reviewed the many articles submitted to CiSE last year-the peer review process helps maintain the magazines revered quality. All of us in the computational science community owe gratitude to people who participate in this crucial service. Readers who would like to contribute as reviewers can visit www.computer.org/web/peer-review/magazines to find out how they can get involved.
Use of the Python language in scientific computing has always been characterized by the coexistence of interpreted Python code and compiled native code, written in languages like C or Fortran. This column takes a fresh look at the problem and introduces Pythran, a new optimization tool designed to efficiently handle unmodified Python code.
Multiagent-Based Coordination Consensus Algorithm for State-of-Charge Balance of Energy Storage Unit
A multiagent-based coordination consensus algorithm was designed to simultaneously meet state-of-charge consensus and DC bus voltage stability requirements. A distributed energy storage system model with four batteries was built in Matlab/Simulink, and local droop control was employed to achieve the proposed algorithm. A multiagent system and agent weak communication network were constructed in the JADE platform, and the effectiveness of the proposed algorithm was verified by Simulink and JADE interactive simulation. The effect of communication delay on the system was also analyzed.
Clay Shirkys influential book Here Comes Everybody: The Power of Organizing Without Organizations explored the potential for Internet-based social networking to change society by making it easier for people to come together. This column evaluates that claim, concluding that Shirky might have overestimated how much people, campaigning aside, really do want to come together.
The process of getting research results accepted to an established magazine or journal can be stressful. CiSE Editor in Chief Jim Chen shares his experiences with the manuscript review process, with the goal of increasing readers and potential authors understanding.
It’s not a stretch to compare a soldier’s orders to the program that governs today’s autonomous drones. In those circumstances, the programmer and the person who authorized the program are responsible and culpable in the event of a bug that causes a war crime.
How Do We Create More Equitable, Diverse, and Inclusive Organizations, and Why Does it Matter? A White Male’s Perspective
Diversity and inclusion must be personally meaningful so that they continue to be on our minds in everything we do long after diversity training has ended. When we create deep connections, we build community and inspire others to follow suit, gently but boldly leading a wave of change throughout our organizations. This is how we make computational science more equitable, diverse, and inclusive.
Former CiSE EICs reflect on the magazine’s 20th anniversary.
Transforming generic and powerful visual means into a visual interface for navigating and exploring scientific data sets requires a fully integrated pipeline of data transformation, representation, and visual mapping. This article presents TransGraph for time-varying data visualization and FlowGraph for flow field exploration. Both graph designs are hierarchical, enabling level-of-detail exploration of large scientific data in an adaptive manner.
If you have been following developments in software engineering in recent years, you have probably noticed that the term DSL (domain-specific language) has become a minor buzzword in that field. You may have concluded that this is a hot new idea that is certainly not ready for application in real life. But, as I will show in this article, computational scientists (and others) have been using DSLs for decades. What is new is not DSLs per se, but the name and the attention given to them.
A new initiative in data science is using the Mira and Theta supercomputers at the US Department of Energy’s Argonne National Laboratory to discover new dye materials suitable for dye-sensitized solar cells. This project aims to develop a new material-by-design methodology by using natural language processing, machine learning, and data mining in conjunction with large-scale simulation and experiments. This synergistic computational and experimental science approach will enable the discovery of new light-absorbing dye molecules, which are needed for the development of solar-powered windows that have the potential to power buildings in an entirely energy-sustainable fashion.
This book addressed the science and use of machine learning. Liberal use of examples helps the reader connect with concepts and techniques in machine learning, which constitutes roughly two-thirds of the text. Discussions of the digital age and computer infrastructure diverge from the main narrative of novel computation, but might prove valuable for those who have less experience with computers. The author introduces questions of ethical data stewardship, which is important given how well this new paradigm has performed and the promise it holds. As a non-mathematical survey of ideas and approaches in machine learning that animate data into solutions, this book is largely a success. I recommend it for readers early in their acquaintance with the new AI.
Physics, medicine, astronomy — these and other hard sciences share a common need for efficient algorithms, system software, and computer architecture to address large computational problems. And yet, useful advances in computational techniques that could benefit many researchers are rarely shared. To meet that need, Computing in Science & Engineering (CiSE) presents scientific and computational contributions in a clear and accessible format.Subscribe to IEEE NEWS feed