Neural Smithing Supervised Learning In Feedforward Artificial Neural Networks [EXCLUSIVE] Download

Neural Smithing Supervised Learning In Feedforward Artificial Neural Networks Download > __https://geags.com/2taRGj__

If we consider the training requirements of neural networks, however, there is still a potentially viable path to geologic applications. The reason is that, in many cases, a neural network must be trained using large amounts of input data before the network can be considered to have achieved an adequate level of performance. In contrast, the training requirements for statistical models are often greatly reduced. For example, the time required to train a neural network depends heavily on the quality of the input data. In principle, one could collect a large data set that is representative of the pattern of interest and then use an artificial neural network to determine the geologic parameters of interest. This will not produce the most accurate estimates but it would be a very inexpensive and rapid method of testing the hypothesis. Neural networks excel at this type of problem, because they can learn what is important and what is not without additional information, unlike statistical methods that require both the existence of a model and a data set. So neural networks may be preferred to statistical models if a large-scale data set is available and the desired geologic parameters are very intuitive. For example, the areas of the data set that contain the most data loss or noise may be left out of the training process. If the training is successful, the trained neural network could then be used to analyze the entire data set.

The reader may be wondering if artificial neural networks would be better suited to address geologic problems. The short answer is that although neural networks are conceptually attractive, they have not yet been widely applied to geologic problems. Among the reasons for this paucity of literature are the following. First, the computational resources required for successfully training a neural network are substantial. Even with today's powerful computing machines, these systems are still very limited in number and size. The hardware required to build and train neural networks is quite different from the hardware required for more traditional geologic methods. The second issue is that any of the standard neural network software is extremely complex and, to date, almost impossible to implement reliably. It is unlikely that any meaningful implementation can be built without a great deal of effort. Currently, there is no standard framework for building and training these neural networks.

While this book focuses on neural networks, there are other methods that can be used to solve a similar pattern matching problem. In particular, support vector machines (SVM) and K-nearest neighbors are discussed briefly as alternative approaches.

The availability of this diverse toolkit for tackling neural network models can help understand some of the limitations in your neural network approach to geology. In particular, neural networks often have limitations that require specific consideration when implementing the training of these models. In addition, neural networks can be easily over-trained, and an expert knowledge of the training procedure can assist in avoiding this training-induced behaviour. Finally, some problems are not efficiently solvable using a neural network. Very complex situations can be solved only through the use of rigorous machine-learning techniques. 827ec27edc