In a multi-layer perceptron (MLP) architecture, input features are passed through several intermediate layers of neurons where they undergo non-linear transformations before reaching the final layer that outputs the model's predictions. What are these intermediate layers called?
Select an answer to reveal the explanation.
Short Explanation and Infographic
Okay, let's dive in. When you look at a diagram of a neural network, you've got your inputs on the far left (like pixel values or house features) and your outputs on the right (like a classification label or price). Everything in the middle is what we call a hidden layer. Why 'hidden'? Simple: because they don't interact directly with the outside world. Their job is to extract patterns and features from the input and pass them along. As you add more hidden layers, the network can learn more complex relationships. It's called deep learning because we're stacking these hidden layers! They are definitely trainable (in fact, that's where most of the learning happens), they don't just contain biases, and they don't make the final prediction—that's the output layer's job. Remember, if it's in the middle, it's a hidden layer.
Full explanation below image
Full Explanation
The correct answer is Hidden layers (Option A). An artificial neural network consists of three types of layers: the input layer, one or more hidden layers, and the output layer. The input layer receives the raw features, and the output layer generates the final prediction. The hidden layers are located between the input and output layers. They are called "hidden" because their intermediate activations are not directly visible as inputs or outputs of the system. In these layers, neurons compute weighted sums of their inputs, add a bias, and apply a non-linear activation function. This allows the network to learn complex, non-linear relationships and represent abstract features at different levels of hierarchy. Option B (Bias layers) is incorrect because biases are parameters added to individual neurons within layers, not separate layers themselves. Option C (Prediction layers) is incorrect because the final predictions are generated by the output layer, not intermediate layers. Option D (Static layers) is incorrect because the weights and biases within the hidden layers are dynamic and updated continuously during training using optimization techniques like gradient descent.