2016 ANNUAL REVIEW Ministry of Science and Technology

are 36% and 23%, which compares favorably with the change in biological muscle molecules (27%). This study received support from MOST's Science Vanguard Research Program and National Taiwan University's Excellence Program, and a paper resulting from the project was published in the September 19, 2016 online edition of Nature- Chemistry ( Nature. Chem. 2016, DOI: 10.1038/ nchem.2608). (2) Variable selection and discovery of model structure in a semiparametric higher-dimensional big data model For more than a decade, thanks to the rapid development of data collection and access equipment, vast amounts of high-dimensional data in different formats have been produced as a result of applications and research in different areas. Data of this type includes genomic data, long-term tracking data, financial time series data, social networking data, and meteorological data. How to analyze huge bodies of complex data in order to obtain new knowledge and make decisions and forecasts, and be able to confirm the correctness and effectiveness of learning and applications, is an extremely challenging issue. As a result, there has been a great need for statistical learning theory and methods, which continue to make many groundbreaking contributions in this field. Long-term tracking data is commonly seen in large-scale medical, epidemiological, and quantitative economic research. In particular, the marginal varying-coefficient regression model is one of the standard analytical models for cases in which the values of the explanatory and dependent variables of a single entity are observed at different points in time. In extreme high dimensional situations, most explanatory variables are unimportant, and only a small number of explanatory variables have a non- zero variation coefficient function. In such a case, important explanatory variables can be distinguished as having either a constant coefficient of variation or a nonconstant coefficient of variation. In a paper published in Annals of Statistics in 2014, the research team in this project constructed a two-step method for obtaining the correct marginal varying-coefficient regression model, and this method A. General Specific-topic Research Projects 1. Natural Science Natural science research chiefly takes the form of basic scientific research, and encompasses the areas of mathematics, statistics, physics, chemistry, and earth science, etc. In order to strengthen interdisciplinary research involving the natural sciences and other areas of science and technology, while also taking into consideration international academic research trends, MOST has been actively planning and implementing focused research in relevant areas, and has sought to achieve the goals of promoting the long-term cultivation of natural science manpower and the pursuit of academic excellence in research. The following were among some of the most significant research results of the year: (1) Use of mechanically interlocked daisy chain-like structures as multidimensional artificial molecular muscles In recent years, large numbers of "smart materials" able to change their characteristics after receiving external stimuli or sensing environmental changes have begun appearing in people's everyday lives. As a consequence, the design and realization of such materials' molecular-level structures and behavior has become an extremely important part of the development of smart materials. Because the products of molecular daisy chain assembly are dimers, and dimers can contract and extend in cyclic fashion under the influence of external stimuli-which is similar to the behavior of the molecules in biological muscles-these products can serve as excellent artificial molecular muscle units. After successfully overcoming difficulties in achieving the self-assembly of higher-order ring daisy chains, a research team from the Dept. of Chemistry, National Taiwan University was able to complete the synthesis of ring daisy chain trimers and tetramers. The team also proved that the behavior of these materials can be controlled after performing structural interlocking, and they can serve as artificial molecular muscles simulating biological muscles in multidimensional spaces. The change in the length of the molecules in their extended and contracted states Research and Development Performance Ⅱ Support for Academic Research Ministry of Science and Technology 29