The Korea Advanced Institute of Science and Technology (KAIST) is a national research university located in Daedeok Innopolis, Daejeon, South Korea. KAIST was established by the Korean government in 1971 as the nation's first public, research-oriented science and engineering institution. KAIST is considered to be one of the most prestigious universities in the nation. KAIST has been internationally accredited in business education, and hosting the Secretariat of the Association of Asia-Pacific Business Schools (AAPBS). KAIST has 10,504 full-time students and 1,342 faculty researchers (as of Fall 2019 Semester) and had a total budget of US$765 million in 2013, of which US$459 million was from research contracts.In 2007, KAIST partnered with international institutions and adopted dual degree programs for its students. Its partner institutions include the Technical University of Denmark, Carnegie Mellon University, the Georgia Institute of Technology, the Technical University of Berlin, and the Technical University of Munich.
Time-series data exists in every corner of real-world systems and services, ranging from satellites in the sky to wearable devices on human bodies. Learning representations by extracting and inferring valuable information from these time series is crucial for understanding the complex dynamics of particular phenomena and enabling informed decisions. With the learned representations, we can perform numerous downstream analyses more effectively. Among several approaches, deep learning has demonstrated remarkable performance in extracting hidden patterns and features from time-series data without manual feature engineering. This survey first presents a novel taxonomy based on three fundamental elements in designing state-of-the-art universal representation learning methods for time series. According to the proposed taxonomy, we comprehensively review existing studies and discuss their intuitions and insights into how these methods enhance the quality of learned representations. Finally, as a guideline for future studies, we summarize commonly used experimental setups and datasets and discuss several promising research directions.
Numerical investigations on prediction of flashback of a lean premixed hydrogen flame in an industrial gas turbine combustor were conducted. Three combustion models were tested for prediction capabilities: (i) the flamelet-generated manifold-finite rate (FGM-FR) model, (ii) the flamelet-generated manifold-turbulent flame speed (FGM-ST) model, and (iii) the eddy dissipation concept (EDC) model. Without an additional wall damping model, the three models are unable to predict the flame flashback limit correctly. A wall damping model was applied to modify the reaction rate near the wall. With certain values of the wall damping coefficient, the FGM-FR and FGM-ST models successfully predict the flame flashback limit.
Physics-based simulations are essential in science and engineering, yet creating them typically requires expert knowledge of numerical solvers and governing equations. Large language models (LLMs) offer new possibilities for natural language-based simulation, but they often fail when prompts are vague, incomplete, or multilingual. We present MCP-SIM (Memory-Coordinated Physics-Aware Simulation), a self-correcting multi-agent framework that transforms underspecified prompts into validated simulations and explanatory reports. The system integrates input clarification, code generation, error diagnosis, and multilingual explanation through structured agent collaboration and persistent memory. Rather than relying on one-shot code generation, MCP-SIM emulates expert-like reasoning via iterative plan–act–reflect–revise cycles. Tested on a twelve-task benchmark across diverse physics domains, MCP-SIM achieved 100% success, significantly outperforming baseline LLMs. In addition to numerical accuracy, the system produces interpretable, language-localized reports that explain each simulation’s physical logic. MCP-SIM represents a step toward general-purpose autonomous scientific assistants that simulate, adapt, and teach through natural language.
In this paper, a novel adaptive quality-based multi-fidelity (AQBMF) surrogate framework is introduced to maximize the utilization of low-fidelity (LF) data from various domains. The main goal of the proposed method is to adaptively select and combine LF data, by assessing its quality, to create the most accurate surrogate. The core idea lies in interpreting the quality levels of LF data sources as the relative importance of LF surrogates that serve as basis functions in a multi-fidelity (MF) surrogate. Based on this approach, the proposed AQBMF surrogate framework comprises four main stages. In the first stage, a newly defined augmented MF formulation is constructed, initially assuming equal importance for all LF data sources. In the second stage, LF surrogates are ranked by importance through the proposed MF basis screening method. In the third stage, promising candidate surrogates are systematically constructed based on the importance ranking of the LF surrogates. During this stage, both the selection and filtering of LF data, as well as the hierarchical and ensemble combination-based MF methods are considered. In the last stage, the best surrogate is selected from the candidates using the proposed algorithm. Various benchmark test results demonstrate the superior performance of the proposed framework. Finally, engineering application results show that the proposed AQBMF surrogate achieves higher accuracy than existing ones within the same computational budget.
Lithium–sulfur batteries (LSBs) are attracting significant attention as next‐generation energy storage systems due to their high theoretical specific capacity (1675 mAh g−1), high‐energy density (2600 Wh kg−1), and light weight. Despite these advantages, the practical application of Li–S batteries is impeded by several challenges, such as the low electrical conductivity of sulfur and lithium sulfide and the shuttle effect of lithium polysulfides (LiPSs), which lead to rapid capacity fading. Carbon materials have emerged as strong candidates for addressing these issues, owing to their excellent electrical conductivity, high porosity, and superior physical and chemical stability. Furthermore, the performance of carbon materials can be enhanced by doping with non‐metallic elements or incorporating metallic elements to take advantage of the large surface area. Such modifications enhance the chemical adsorption of LiPSs and improve the catalytic activity for electrochemical reactions, thereby significantly boosting the overall performance of LSBs. In this review, recent advances in carbon materials are systematically categorized according to their dimensional architecture—0D, 1D, 2D, and 3D. This classification reflects the distinct structure‐dependent characteristics of each dimensionality, including surface area, pore structure, electron/ion transport properties, and the accessibility of active sites. It enables comparative evaluation of their roles as sulfur hosts and interlayers and offers insights into the rational design of high‐performance carbon materials for Li–S batteries.
