Neural network architecture in ML Python - Practice Problems & Coding Challenges

Challenge - 5 Problems

❓ Model Choice

intermediate

Choosing the right layer for image feature extraction

You want to build a neural network to recognize objects in photos. Which layer type is best to start with for extracting features like edges and shapes?

A A dropout layer

B A convolutional layer

C A fully connected layer

D A recurrent layer

💡 Hint

Think about which layer type is designed to scan images for patterns.

Convolutional layers scan images with small filters to detect edges and shapes, making them ideal for image tasks.

❓ Metrics

intermediate

Understanding output shape of a dense layer

You have a dense (fully connected) layer with 128 neurons. If the input to this layer has shape (batch_size, 64), what will be the output shape?

A (batch_size, 64)

B (64, 128)

C (128, 64)

D (batch_size, 128)

💡 Hint

Each neuron produces one output per input example.

A dense layer outputs one value per neuron for each input example, so output shape is (batch_size, number_of_neurons).

❓ Predict Output

advanced

Output shape after convolution and pooling

What is the output shape of the tensor after applying the following layers in order?

Input shape: (32, 64, 64, 3)  # batch_size=32, 64x64 RGB images
Conv2D(filters=16, kernel_size=3, padding='same')
MaxPooling2D(pool_size=2)

A (32, 32, 32, 16)

B (32, 62, 62, 16)

C (32, 64, 64, 16)

D (32, 31, 31, 16)

💡 Hint

Remember 'same' padding keeps width and height, pooling halves them.

The Conv2D with 'same' padding keeps the spatial size at 64x64 and changes channels to 16. MaxPooling with pool size 2 halves width and height to 32x32.

🔧 Debug

advanced

Identifying error in model layer connection

Consider this simple neural network code snippet:

model = Sequential()
model.add(Dense(64, input_shape=(100,)))
model.add(Conv2D(32, kernel_size=3))
model.add(Flatten())
model.add(Dense(10, activation='softmax'))

What error will this code raise when run?

A No error, model compiles successfully

B TypeError because Dense layer cannot be followed by Conv2D

C ValueError due to Conv2D expecting 4D input but receiving 2D

D RuntimeError due to missing activation function in Conv2D

💡 Hint

Check the input shape expected by Conv2D layers.

Conv2D layers expect 4D input (batch, height, width, channels), but Dense outputs 2D (batch, features), causing a ValueError.

🧠 Conceptual

expert

Effect of increasing depth on model capacity

What is the main effect of increasing the number of layers (depth) in a neural network architecture?

A It increases the model's ability to learn complex features by creating hierarchical representations

B It decreases training time by reducing parameters

C It always reduces overfitting by simplifying the model

D It makes the model less flexible and harder to train

💡 Hint

Think about how deeper networks learn features step-by-step.

More layers allow the model to learn complex, hierarchical features, improving its capacity to understand data patterns.