Is the bias neuron of a neural network equivalent to a y-intercept of a linear regression?

Question

Is the above statement true? bias and \beta_0 are both weights that are independent of the input so the allow the model to add some constant value.

Consider a case where the network trains equally well either with or without an output bias neuron. In such a case, the answer is no even if the bias neuron output is not zero. — James Phillips, Aug 02 '19 at 02:08

score 1 · Accepted Answer · answered Aug 04 '19 at 16:53

1

Firstly, a linear regression tries to estimate a function while a single neuron divides the input space into two sub-spaces, so they do essentially different tasks.

Having said that, the \beta_0 in neuron and the y-interception in linear regression are both biases: constants that are applied to the final output regardless of the output (just for clarification: they depend on the input in the sense that models learns them from input data).

answered Aug 04 '19 at 16:53

Bahman Rouhani

1,139
2
14
33

thanks for the clarification, that does make sense to me. – Quastiat Aug 17 '19 at 11:18

Is the bias neuron of a neural network equivalent to a y-intercept of a linear regression?

1 Answers1