![]() PAVLIS N K, HOLMES S A, KENYON S C, et al. The New Generation of Global Mean Sea Surface Height Model Based on Multi-altimetric Data. JIN Taoyong, LI Jiancheng, JIANG Weiping, et al. International Association of Geodesy Symposia. ![]() The DTU13 MSS (Mean Sea Surface) and MDT (Mean Dynamic Topography) from 20 Years of Satellite Altimetry//RIZOS C. DNSC08 Mean Sea Surface and Mean Dynamic Topography Models. Strategies for Solving the Vertical Datum Problem Using Terrestrial and Satellite Geodetic Data//SüNKEL H, BAKER T. ![]() Height Datum Definition, Height Datum Connection and the Role of the Geodetic Boundary Value Problem. Proceedings of the 3rd Meeting of the International Gravity and Geoid Commission. Defining the Geoid by W=W 0≡U 0:Theory and Practice of a Modern Height System//TZIAVOS I N. Key words: geoid, global vertical datum, GPS/leveling, 1985 national height datum, vertical shift The final results demonstrate that the 1985 national height datum is 0.298 0 and 0.464 2 m above the mean sea level and the global geoid, respectively. The vertical shift value is improved by weighting, and utilizing two methods to verify the rationality and correctness after weighting. Then, 649 GPS/leveling data,distributed evenly over the mainland of China,are selected to calculate the 1985 national height datum geopotential and the vertical shift using three methods, combined with the global gravity model EGM2008. Finally, the paper summarizes representative applications of disentangled representation learning in the field of remote sensing and discusses its future development.The mean geopotential value of the global mean sea surface, 62 636 856.550 7 m 2s -2, is determined based on the global gravity model(EIGEN-6C4、EGM2008) and global mean sea surface height model(DNSC08、DTU10、DTU13).The geoidal potential (62 636 858.179 0 m 2s -2) can be obtained by subtracting the mean sea surface topography from the mean sea surface height. Subsequently, the loss functions and objective evaluation metrics commonly used in existing work on disentangled representation are classified. Then, disentangled representation learning algorithms are classified into four categories and outlined in terms of both mathematical description and applicability. In this paper, we first introduce and analyze the current status of research on disentangled representation and its causal mechanisms and summarize three crucial properties of disentangled representation. Disentangled representation learning can capture information about a single change factor and control it by the corresponding potential subspace, providing a robust representation for complex changes in the data. Disentangled representation learning aims to learn a low-dimensional interpretable abstract representation that can identify and isolate different potential variables hidden in the high-dimensional observations. ![]() However, the current representations are usually highly entangled, i.e., all information components of the input data are encoded into the same feature space, thus affecting each other and making it difficult to distinguish. The transition of input representations for machine learning algorithms from handcraft features, which dominated in the past, to the potential representations learned through deep neural networks nowadays has led to tremendous improvements in algorithm performance. Representation learning is one of the core problems in machine learning research.
0 Comments
Leave a Reply. |
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |