Here, the neural network nodes compete with each other for the right to respond to a subset of the input data. The hidden layer is called the competitive layer. Every competitive neuron has its own weight and we calculate the similarity measure between the individual input vector and the neuron weight. For each input vector, the hidden neurons compete with each other to see which one is the most similar to the particular input vector:

The output neurons are said to be in competition for input patterns.

• During training, the output neuron that provides the highest activation to a given input pattern is declared the weights of the winner and is moved closer to the input pattern, whereas the rest of the neurons are left unchanged
• This strategy is also called winner-takes-all, since only the winning neuron is updated:

Let us see a simple competitive learning algorithm example to find three neurons within the given input data:

1. We will have three input neurons in the input layer. Each input to the neuron is a continuous variable and let the weight at each input neuron be a random number between 0.0 and 1.0. The output of each node is the product of the three weights and its input.
2. Each competitive layer neuron receives the sum of the product of weights and inputs.
3. The competitive layer node with the highest output is regarded as the winner. The input is then categorized as being within the cluster corresponding to that node.
4. The winner updates each of its weights, moving the weight from the connections that gave it weaker signals to the ones that gave it stronger signals.

Thus, as we receive more data, each node converges on the center of the cluster that it has come to represent. It activates more strongly for inputs belonging to this cluster and more weakly for inputs that belong to other clusters.

There are basically two stopping conditions of competitive learning:

• Predefined number of epochs: Only N epochs are run and this prevents the algorithm from running for a relatively long time without convergence
• Minimum value of weight update: The algorithm is run until we have a minimum value of weight update