Geek Answers Handbook

Choosing the Right Neural Network Type

I have a controlled learning problem where my algorithm will be provided with a set of training examples to examine whether a figure is a circle in a square. I was wondering which type of ANN would be the best. I know that you can choose a perceptron if the data is linearly separable. Can I easily have a hyperplane that divides my squares and circles up? So, is the perceptron a good enough choice? However, are multilayer direct transmission networks used more often? What is the natural choice and why?

The following figure shows the training data given to the system. NN, it is necessary to classify the two-dimensional data A = [a1, a2] into squares and circles.

enter image description here

Thanks.

+7
source share
2 answers

The dataset you provided is not linearly shared in the space covered by a1 and a2, so the perceptron will not do. You need a multilayer perceptron (MLP). In general, MLPs are used more often because they can do everything a single-layer perceptron can do (look for a universal approximation theorem). The radial base function will also do the job. Noli hinted at something interesting, but more complicated - the data set becomes linearly separable with a high probability if projected onto a very very multidimensional space (cover theorem). This is the motivation for using vector support machines.

Thus, there is no natural choice; this is a completely specific problem. Experiment. My lecturer said "cross valid is your friend"

+3
source

Why are you installing NN for a specific reason? bored? if not ... do not watch LibSVM

http://www.csie.ntu.edu.tw/~cjlin/libsvm/

+2
source

More articles:


All Articles