tailieunhanh - Fetal Electrocardiography – part 9

BPN do đó có thể thích nghi hoặc tìm hiểu các mô hình của mối quan hệ giữa đầu vào và giá trị sản lượng trong một tập dữ liệu lớn. Với một mạng lưới đủ lớn và đào tạo, mạng lưới có khả năng thích ứng với bất kỳ mô hình cung cấp các tập dữ liệu | 124 Fetal Electrocardiography compared with idealised results in the reference data set to produce corrections for input weightings and the processes are repeated until the outputs of the network are sufficiently close to that in the reference data set. The BPN is therefore able to adapt or learn the patterns of relationship between the input and output values in a large data set. Given a sufficiently large network and training the network is able to adapt to any pattern provided the dataset does not contain contradictory patterns. The method of adaptation is by iterative approximation and is not dependent on any assumption about the nature of the data. In analysis the BPN can be viewed as a form of non-parametric multiple regression but it is free from the assumptions of linearity normal distribution or regular relationship between measurements. Additionally trained BPNs are particularly useful as carriers of complex algorithms to transform sets of data from one domain to another. The SOM consists of neurones arranged in a regular pattern in one or more dimensions each containing the same number of weights as the expected input data. Input data are matched with every member of the map to find the best fit winner . The weights of the winner neurone and its neighbours are then adjusted towards the input values so that they are more likely to become winners when matched with similar patterns. This process is repeated so that clusters of neurones which are similar to each other but different from those in other clusters in the map appear as demonstrated in Fig. . Fig. . Schematic representation of a self-organising map SOM neural network. The different shadings represent clusters or families of neurones that are similar. The number of neurones within a family represents the variation within that family. In a SOM an input pattern is matched to all neurones in the map the neurone with the closet matching memorised pattern to the input being the winner. Information