| Autor: |
Xiru Zhang, Jacquelyn S. Fetrow, George Berg |
| Rok vydání: |
1994 |
| Předmět: |
|
| Popis: |
Publisher Summary This chapter describes the design of an auto-associative neural network with hidden layer activations used to reclassify local protein structures. Most artificial neural network models are trained to perform a specific task. This is done by presenting a series of input patterns to the network. For each pattern, the network calculates an output pattern that is compared to the desired output pattern for this input. The hidden unit values of a properly trained auto-associative neural network represent the important features of the networks inputs. If the hidden unit values are used as the canonical representation of the inputs, the input layer to hidden layer portion of the trained network can be used as an encoder for the representations and the hidden layer to output layer portion as a decoder. An auto-associative network should have the smallest hidden layer that allows the network to learn the auto-association task. This forces the hidden unit values vector to become a concise representation of the information in the inputs. |
| Databáze: |
OpenAIRE |
| Externí odkaz: |
|
