What is a neural network (simplified) in AI for Everyone - Practice Questions & Exercises

Challenge - 5 Problems

🧠 Conceptual

intermediate

Understanding the basic structure of a neural network

Imagine a neural network as a group of friends passing notes to solve a problem together. Which part of the neural network acts like the friends passing notes?

A The data input only

B The computer hardware

C The final output only

D The layers of neurons

💡 Hint

Think about who is actively working together inside the network.

The layers of neurons are like friends passing notes, each layer processes information and passes it on to the next.

🔍 Analysis

intermediate

Output of a simple neural network forward pass

What is the output of this simple neural network forward pass code?

AI for Everyone

import numpy as np

def simple_forward(x, w, b):
    return np.dot(x, w) + b

x = np.array([1, 2])
w = np.array([0.5, -1])
b = 0.1
output = simple_forward(x, w, b)
print(output)

A 1.1

B 0.1

C -1.4

D -0.9

💡 Hint

Calculate dot product: (1*0.5) + (2*-1) + 0.1

The dot product is 0.5 + (-2) = -1.5, then add bias 0.1, total -1.4. Check carefully.

❓ Model Choice

advanced

Choosing the right neural network type for image recognition

You want to build a model to recognize objects in photos. Which neural network type is best suited for this task?

A Convolutional Neural Network (CNN)

B Recurrent Neural Network (RNN)

C Feedforward Neural Network without convolution

D Autoencoder

💡 Hint

Think about which network type is designed to handle images and spatial data.

CNNs are designed to process images by detecting patterns like edges and shapes using convolution layers.

❓ Hyperparameter

advanced

Effect of learning rate on training a neural network

What happens if the learning rate is set too high when training a neural network?

A The model may fail to learn and the loss might bounce around or increase

B The model will learn faster and reach perfect accuracy immediately

C The model will ignore the training data

D The model will always underfit the data

💡 Hint

Think about how big steps in learning affect stability.

A too high learning rate causes the model to take big steps, missing the best solution and causing unstable training.

❓ Metrics

expert

Interpreting confusion matrix metrics for a neural network classifier

A neural network classifier outputs the following confusion matrix:

True Positive (TP): 40
False Positive (FP): 10
True Negative (TN): 30
False Negative (FN): 20

What is the precision of this model?

A 0.66

B 0.80

C 0.67

D 0.75

💡 Hint

Precision = TP / (TP + FP)

Precision measures how many predicted positives are actually correct: 40 / (40 + 10) = 0.8

Practice

1. What is a neural network best compared to in simple terms?

easy

A. A single rule that always gives the same answer

B. A random number generator

C. A tiny brain made of connected nodes

D. A calculator that only adds numbers

Solution

1. Step 1: Understand the analogy

   A neural network is often described as a tiny brain because it has many connected parts that work together.

2. Step 2: Compare options to the description

   Only A tiny brain made of connected nodes describes a network of connected nodes like a brain, matching the simple explanation.

3. Final Answer:

   A tiny brain made of connected nodes -> Option C

4. Quick Check:

   Neural network = tiny brain [OK]

Hint: Think of neural networks as tiny brains with connections [OK]

Common Mistakes:

• Confusing neural networks with simple calculators
• Thinking neural networks are random
• Believing neural networks use fixed rules only

2. Which of the following is the correct way to describe how a neural network learns?

easy

A. It memorizes every input exactly as it is

B. It adjusts connections between nodes to recognize patterns

C. It guesses randomly without any changes

D. It deletes all connections after each use

Solution

1. Step 1: Recall how learning happens in neural networks

   Neural networks learn by changing the strength of connections between nodes to better recognize patterns.

2. Step 2: Match this to the options

   Only It adjusts connections between nodes to recognize patterns correctly describes this learning process.

3. Final Answer:

   It adjusts connections between nodes to recognize patterns -> Option B

4. Quick Check:

   Learning = adjusting connections [OK]

Hint: Learning means changing connections, not random guessing [OK]

Common Mistakes:

• Thinking neural networks memorize all data exactly
• Believing learning is random guessing
• Assuming connections are deleted after use

3. Consider this simple neural network code snippet:

inputs = [1, 0]
weights = [0.5, 0.5]
output = sum(i * w for i, w in zip(inputs, weights))
print(output)

What will be printed?

medium

A. 0.5

B. Error

C. 0.0

D. 1.0

Solution

1. Step 1: Calculate weighted sum

   Multiply inputs and weights element-wise: 1*0.5=0.5, 0*0.5=0.0

2. Step 2: Sum the results

   Sum is 0.5 + 0.0 = 0.5

3. Final Answer:

   0.5 -> Option A

4. Quick Check:

   Weighted sum = 0.5 [OK]

Hint: Multiply inputs by weights and add to find output [OK]

Common Mistakes:

• Adding inputs without weights
• Multiplying weights incorrectly
• Expecting an error from correct code

4. This code tries to compute a neural network output but has an error:

inputs = [1, 2]
weights = [0.3, 0.7]
output = 0
for i in range(len(inputs)):
    output += inputs[i] * weights[i]
print(output)

What is the error?

medium

A. Missing import statement

B. Weights list is too short

C. Output should start at 1, not 0

D. Using input values as indexes for weights causes an error

Solution

1. Step 1: Identify how weights are accessed

   The original code uses input values (1, 2) as indexes to access weights, but weights has only 2 elements at indexes 0 and 1.

2. Step 2: Understand the error cause

   Accessing weights[2] causes an index error because weights list has no index 2.

3. Final Answer:

   Using input values as indexes for weights causes an error -> Option D

4. Quick Check:

   Index error from input as index [OK]

Hint: Use indexes, not input values, to access weights [OK]

Common Mistakes:

• Thinking weights list length is the problem
• Starting output at wrong value
• Assuming imports are needed here

5. You want to build a simple neural network to recognize if an image is bright or dark. Which approach best fits this task?

hard

A. Use a network that adjusts connections based on pixel brightness patterns

B. Use a fixed rule that always says 'bright' regardless of image

C. Randomly guess bright or dark without learning

D. Manually check each pixel without any network

Solution

1. Step 1: Understand the task

   Recognizing brightness needs pattern detection from pixel data.

2. Step 2: Choose the correct method

   A neural network that learns by adjusting connections to pixel patterns can solve this task well.

3. Final Answer:

   Use a network that adjusts connections based on pixel brightness patterns -> Option A

4. Quick Check:

   Learning network fits pattern recognition [OK]

Hint: Choose learning networks for pattern tasks like brightness detection [OK]

Common Mistakes:

• Using fixed rules that ignore input
• Guessing randomly without learning
• Trying manual pixel checks without automation