Machine learning - Kohonen Neural Network WTA vs WTM learning - what's the difference?

Winner Takes All (WTA Learning)

In that learning, always only weights of winning neuron are updated per iteration.

Winner Takes Most (WTM Learning)

In that learning, weights of winning neuron and neurons around are updated per iteration. That learning uses radius to indicate update scope on neurons around. When radius is equal to zero, WTM learning uses WTA learning.

We have different distribution functions to calculate updates for neurons.

e.g. standard normal distribution where winning neuron is in the center (Gausian Distribution):

xxxxxxxxxx

function calculateGausianDistribution(radius, winnerX, winnerY, neuronX, neuronY) {

    var dx = winnerX - neuronX;

    var dy = winnerY - neuronY;

    return Math.exp(-0.5 * (dx * dx + dy * dy) / (radius * radius));

}

e.g. for radius = 3 distibution function looks like:

See also

1. JavaScript - Kohonen Neural Network (WTA Learning)

2. JavaScript - Kohonen Neural Network (WTM Learning)

References

• Winner-take-all (computing)