wayground logo

Free Printable Worksheets

NEW

Font size

S
M
L
XL
Worksheets

Perceptrons and Activation Functions

Total questions: 15

Worksheet time: 8mins

Name
Class
Date
1.

What is the primary role of a non-linear activation function in a perceptron?

a)

To speed up the training process

b)

To decrease the number of parameters in the network

c)

To reduce the computational complexity of the network

d)

To introduce non-linearities that allow the network to learn complex, non-linear relationships

2.

Which of the following is NOT a common activation function used in deep neural networks?

a)

Square function

b)

Rectified Linear Unit (ReLU)

c)

Hyperbolic tangent (tanh)

d)

Sigmoid function

3.

In a perceptron, the output ŷ is computed as ŷ = g(w0 + Σ(xi·Wi)). What does w0 represent?

a)

Bias term

b)

Output weight

c)

Activation function parameter

d)

Input weight

4.

Why are fully connected layers in neural networks called "Dense" layers?

a)

Because they compress the information densely

b)

Because they have high density of parameters

c)

Because all inputs are densely connected to all outputs

d)

Because they use dense matrix operations

5.

In binary classification with probabilistic outputs, which loss is most appropriate to minimize?

a)

Hinge loss for margin maximization

b)

Categorical cross-entropy with many classes

c)

Mean squared error on probabilities

d)

Binary cross-entropy over predicted probabilities

6.

Which statement best describes the role of the learning rate η in gradient descent?

a)

It scales the gradient to set update step size

b)

It measures validation accuracy each epoch

c)

It normalizes inputs to the network layers

d)

It counts the number of training epochs

7.

Backpropagation computes which quantity at each layer of a neural network?

a)

Forward activations for new inputs

b)

Regularization penalties only

c)

Gradients of loss via the chain rule

d)

Confusion matrix for predictions

8.

Compared with full-batch gradient descent, mini-batch SGD primarily offers which benefit?

a)

Guaranteed monotonic loss decrease

b)

Exact gradient using the entire dataset

c)

No need to tune the learning rate ever

d)

Faster, noisy updates using subsets of data

9.

What is the principal effect of dropout during training?

a)

Randomly zeroes activations to limit co-adaptation

b)

Replaces backpropagation with random search

c)

Permanently deletes neurons from a trained model

d)

Reduces model depth by merging adjacent layers

10.

Early stopping helps prevent overfitting by doing what when training proceeds?

a)

Halting when validation loss begins to rise

b)

Stopping immediately after training loss hits zero

c)

Continuing until training and validation losses match

d)

Halting after a fixed number of parameter updates

11.

In a convolution operation, what does the filter (kernel) do?

a)

Reduces the image size

b)

Computes element-wise multiplication of weights with patches and sums outputs to detect features

c)

Normalizes pixel values

d)

Applies non-linearity

12.

What are the two main benefits of pooling layers in CNNs?

a)

Increasing model complexity and improving training speed

b)

Reducing dimensionality and providing spatial invariance

c)

Improving accuracy and increasing memory usage

d)

Replacing activation functions and reducing bias

13.

In a typical CNN for classification, what is the correct sequence of operations?

a)

Flatten → Fully Connected → Convolution → Pooling → Softmax

b)

ReLU → Convolution → Pooling → Softmax → Fully Connected

c)

Convolution → ReLU → Pooling → Convolution → ReLU → Pooling → Flatten → Fully Connected → Softmax

d)

Convolution → Pooling → ReLU → Flatten → Fully Connected → Softmax

14.

As you progress through deeper layers in a CNN, what type of features are typically learned?

a)

Only low-level features like edges

b)

A hierarchy from low-level features (edges) to mid-level features (parts) to high-level features (objects)

c)

Random features that are not meaningful

d)

Identical features at all depths

15.

Why do convolutional layers scale better than fully connected layers for images?

a)

They eliminate the need for activation functions

b)

They operate faster by ignoring spatial relationships

c)

They exploit local connectivity and weight sharing to reduce parameters while capturing spatial structure

d)

They work only with color images