Given basic elements of a neuron(as below) with a bias value: [![neuron structure][1]][1] [1]: https://i.sstatic.net/KPDLCRyG.png ------- I learnt that, a bias value allows you to shift the activation function(say **sigmoid function**) to the left or right, which may be **critical for successful learning**. Yet to understand, why? ----- Mathematically, when we change an argument of a function(say sigmoid function in this case) by adding or subtracting a constant, then the output values shift to left or right accordingly, as shown below: [![sigmoid function][2]][2] [2]: https://i.sstatic.net/wcaUpgY8.png ------ My understanding is that, binary classification problems(example - image recognition), use neural network based AI solutions. This is the reason, a neuron runs an activation function like sigmoid function that takes a summation of all inputs(with weights) and provide a output value within interval [0,1] ----- **Question:** 1) Given any argument(input) to a function(sigmoid), how shifting of output values(left or right) using bias **critical to** successful learning(training) in artificial neural network? 2) Mathematically, a function must have one domain and one range. For successful learning, why sigmoid function(single variable function) **use linear combiner**(Σ) for all the inputs(synapsesXparameters)? Because summation of synapsesXparameters may not provide uniqueness, if parameters(weights) have negative value