I know two differences between a neuron and a perceptron

  1. Neuron employs non-linear activation function and perceptron employs only a threshold activation function.

  2. The output of a neuron is not necessarily a binary number and the output of a perceptron is always a binary number

I know no other difference between a perceptron and a neuron other than the above.

Are there any other differences between perceptron and neuron?


1 Answer 1


In addition to those mentioned differences, a perceptron can be thought of as a standalone model (which is trained with a specific algorithm, the perceptron algorithm), while the artificial neuron (sometimes only referred to as neuron, in a similar way that an artificial neuron network is commonly abbreviated to neural network) is the smallest computational unit of a neural network, so it's an abstraction for a relatively simple function (e.g. sigmoid) that will be composed with other simple functions to produce a more complicated function, which is typically non-linear.

Moreover, note that people often refer to "regular" neural networks as multi-layer perceptrons (abbreviated to MLPs) for one simple reason: you can think of such an MLP as the composition of multiple perceptrons, where, in this case, the perceptron would be a synonym for artificial neuron, so the smallest computational unit of a neural network, which performs, for example, a linear combination of its inputs followed by the application of an activation function, which can be the sigmoid, tanh, ReLU, identity, or any other function that is differentiable, if you plan to train the neural network with gradient descent.

So, sometimes, the term perceptron is a synonym for artificial neuron, so the perceptron (aka neuron), in this case, could have any activation function. However, the perceptron is often assumed to have the sign function as the activation function, which is not strictly differentiable, while, as you point out, artificial neurons are not limited to the sign function.

The original (photo)perceptron models, as described in this paper, were more complicated (e.g. the inputs were not directly connected to the outputs, or you could have feedback connections), so the definitions of these concepts or what these terms refer to have evolved or can still evolve. In the past, I have also seen people use the term perceptron to refer to an MLP, but this is probably because they were not aware of the model that we typically refer to as the perceptron, for example, as described in section 8.5.4 (p. 265) of the book Machine Learning: A Probabilistic Perspective by Kevin Murphy (you can find free pdfs of this book on the web).


You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .