Advanced Machine Vision Paradigms for Medical Image Analysis

3. Extreme learning machines

Artificial neural networks are among the most successful techniques applied to classification problems [26]. In most neural networks, it is necessary to tune network parameters to achieve optimal performance [30]. In most neural network algorithms, the tuning process is critical to avoid local minima and reduce computational complexity (i.e., memory use and execution time). Computational complexity tends to be high in architectures such as MLPs.

ELMs are known for their ability to provide reduced training and execution time compared with other connectionist methods [25]. ELM networks usually consist of nonrecurrent intermediate layers composed of neurons with random weights [5,25]. Because we define the output weights ...