ANN (Feedforward Neural Network)

Question 1

In a feedforward neural network, how does information flow?

Accepted Answer

Only in one direction, input toward output — Feedforward networks pass information in one direction only - from input layer through hidden layers to output layer, with no cycles or loops.

Question 2

What determines the output of a neuron?

Accepted Answer

A weighted sum of inputs plus bias, activated — A neuron computes the weighted sum of its inputs, adds a bias term, then applies an activation function to produce its output.

Question 3

A feedforward neural network is characterized by…

Accepted Answer

Information flowing in one single direction — In a feedforward net (MLP), data moves forward through layers with no feedback connections — unlike RNNs which have loops.

Question 4

The Multilayer Perceptron (MLP) is built from…

Accepted Answer

Fully connected — An MLP stacks dense layers, each applying weights, bias, and an activation — the classic feedforward architecture.

Question 5

Does a feedforward network have MEMORY of previous inputs?

Accepted Answer

No, since each input is processed independently — Feedforward nets are memoryless; modeling sequences/temporal dependencies requires RNNs, LSTMs, or transformers.

Question 6

The 'forward pass' in a feedforward network…

Accepted Answer

Computes predictions by propagating data from input to output — Layer by layer it applies W·x+b and activations until the output layer produces the prediction.

Question 7

Feedforward networks are trained with…

Accepted Answer

Backpropagation and gradient descent on the loss — The forward pass computes the loss; backprop computes gradients; an optimizer updates the weights.

Question 8

Hidden layers in a feedforward net provide…

Accepted Answer

Non-linear transforms enabling more complex representations — Each hidden layer learns a new representation; without them the model is just a linear classifier/regressor.

Question 9

A feedforward network with NO hidden layers and a sigmoid output is equivalent to…

Accepted Answer

Logistic regression — With no hidden layers it's a single linear+sigmoid unit — exactly logistic regression. Hidden layers add non-linear power.

Question 10

Feedforward networks can be used for…

Accepted Answer

Both classification and regression tasks on tabular or structured data — By choosing the output layer/loss, an MLP handles classification (softmax/sigmoid) or regression (linear output).

Animation

Input Values

Architecture

Network Summary

One forward pass

Test your knowledge

In a feedforward neural network, how does information flow?

Learn these first

Interview Questions

References & Further Reading

Love this visualization?