Deep Learning 1.1 Quiz

What is the input to the neural network designed to recognize handwritten digits?

• A 10-neuron layer representing digits 0-9.

• A 28x28 pixel image, which translates to 784 input neurons.

• Two hidden layers, each with 16 neurons.

• A single neuron representing the overall digit.

In the context of this neural network, what does a "neuron" primarily represent?

• A complex biological cell that processes information.

• A thing that holds a number between 0 and 1, called its activation.

• A connection between different layers of the network.

• A specific handwritten digit from 0 to 9.

What is the general principle behind how a neural network recognizes complex patterns like digits?

• It directly matches the entire image to a stored template.

• It breaks down patterns into simpler sub-components recognized by different layers.

• It uses a single, complex algorithm to identify all features simultaneously.

• It relies solely on the input pixels without any intermediate processing.

Are you enjoying the video lesson?

• Yes

• No

What do weights and biases represent in a neural network neuron?

• Weights determine the input image's resolution, and biases control the network's speed.

• Weights define the specific pixel pattern a neuron is looking for, and biases set the activation threshold for that pattern.

• Weights are random numbers, and biases are fixed constants.

• Weights control the output layer, and biases control the input layer.

Why is the Rectified Linear Unit (ReLU) activation function often preferred over the Sigmoid function in modern deep neural networks?

• ReLU is computationally more complex, leading to more accurate models.

• Sigmoid functions are better at handling negative inputs, which ReLU struggles with.

• ReLU functions are generally easier and faster to train in deep networks compared to Sigmoid functions.

• Sigmoid functions are primarily used for output layers, while ReLU is for hidden layers.