This paper describes a new method for classifying a dataset that partitions elements into different categories. It has relations with neural networks but a slightly different structure, requiring only a single pass through the classifier to generate the weight sets. A grid structure is required and a novel idea of converting a 1-D row of real values into a 2-D structure of value bands. Each cell in the band can then store a cell weight value and also a set of weights that represent its own importance to each of the output categories. For any input that needs to be categorised, all of the output weight lists for each relevant input cell can be retrieved and summed to produce a probability for what the correct output category is. So the relative importance of each input point to the output is distributed to each cell. The bands possibly work like hidden layers of neurons, but separate ones for each input variable. The construction process itself can simply be the reinforcement of the weight values, without requiring an iterative adjustment process, making it potentially much faster. For online or partial updating, it can possibly include a competitive process as well.
from cs.AI updates on arXiv.org http://ift.tt/1ENtzts
via IFTTT
No comments:
Post a Comment