Exam Questions: AI, ML, and Deep Learning — Top 10 Questions with Answers

 AI, ML, and Deep Learning — Top 10 Questions with Answers

 1. What is the difference between Artificial Intelligence, Machine Learning, and Deep Learning?

Answer:

Concept	Definition	Example
Artificial Intelligence (AI)	The broad field that enables machines to mimic human intelligence and decision-making.	Chatbots, game-playing agents
Machine Learning (ML)	A subset of AI that enables systems to learn from data and improve without explicit programming.	Spam email classification
Deep Learning (DL)	A subset of ML that uses neural networks with many layers to learn complex patterns.	Image recognition, speech translation

Relation:

Deep Learning ⊂ Machine Learning ⊂ Artificial Intelligence

2. What are the types of Machine Learning? Give examples.

Answer:

Type	Description	Example
Supervised Learning	Model learns from labeled data (input → output).	Email spam detection
Unsupervised Learning	Model finds patterns in unlabeled data.	Customer segmentation
Reinforcement Learning	Agent learns by interacting with environment via rewards/punishments.	Self-driving cars, game bots

---

3.What is Overfitting and Underfitting in ML?

Answer:

• Overfitting: Model performs well on training data but poorly on new/unseen data.
  Cause: Too complex model (e.g., deep decision tree).
  Solution: Regularization, pruning, dropout, or more data.

• Underfitting: Model is too simple to capture patterns.
  Cause: Too few features or linear model on non-linear data.
  Solution: Use complex model or feature engineering.

Visualization Tip:

• Training error ↓ but testing error ↑ → Overfitting

• Both errors high → Underfitting

4. Explain Bias-Variance Tradeoff.

Answer:

• Bias: Error due to wrong assumptions (underfitting).

• Variance: Error due to model sensitivity to small data changes (overfitting).

    Goal: Achieve a balance → low bias + low variance → good generalization.

    Tradeoff Curve:
As model complexity ↑ ⇒ bias ↓ but variance ↑.

5. What is the difference between Classification and Regression?

Criteria	Classification	Regression
Output Type	Categorical (discrete labels)	Continuous (numeric values)
Example	Email spam / not spam	Predicting house prices
Algorithms	Logistic Regression, SVM, Decision Tree	Linear Regression, Ridge, Lasso
Evaluation Metrics	Accuracy, F1-score	MSE, RMSE, R² score

6. Explain Gradient Descent and its role in ML/DL.

Answer:

• Gradient Descent is an optimization algorithm used to minimize loss function by iteratively updating parameters (weights).

Formula:
[
w = w - \eta \frac{\partial L}{\partial w}
]
where

• ( w ): weight

• ( \eta ): learning rate

• ( L ): loss function

Types:

• Batch Gradient Descent → Uses all data each step

• Stochastic Gradient Descent (SGD) → One sample at a time

• Mini-Batch Gradient Descent → Uses small data batches (best in practice)

7.  What are Activation Functions in Neural Networks?

Answer:
Activation functions introduce non-linearity in neural networks.

Function	Formula	Use
Sigmoid	( \frac{1}{1+e^{-x}} )	Binary classification
Tanh	( \frac{e^x - e^{-x}}{e^x + e^{-x}} )	Normalized outputs
ReLU	( f(x) = \max(0, x) )	Deep networks (fast convergence)
Softmax	( \frac{e^{x_i}}{\sum e^{x_j}} )	Multi-class classification

ReLU is the most commonly used activation function in deep networks.

8. What is a Confusion Matrix and its metrics?

Answer:   More👉

Confusion Matrix evaluates classification model performance.  More👉

9. What is the difference between CNN and RNN in Deep Learning?

Feature	CNN (Convolutional Neural Network)	RNN (Recurrent Neural Network)
Input Type	Spatial data (images)	Sequential data (text, speech)
Structure	Convolution + pooling layers	Loops with memory of previous outputs
Use Case	Image classification, object detection	Text prediction, translation
Example	ResNet, VGGNet	LSTM, GRU

---

10. What is Reinforcement Learning and what are its components?

Answer:
Reinforcement Learning (RL) is a learning paradigm where an agent learns by interacting with an environment to achieve a goal.

Main Components:

1. Agent → Learner/decision-maker

2. Environment → External system agent interacts with

3. State (S) → Current situation

4. Action (A) → Choices made by agent

5. Reward (R) → Feedback for actions

     Goal: Maximize cumulative rewards over time.

     Key Algorithm: Q-Learning
[
Q(s,a) = Q(s,a) + \alpha [R + \gamma \max Q(s',a') - Q(s,a)]
]

Summary 

No.	Topic	Key Concept
1	AI vs ML vs DL	Hierarchical relationship
2	Types of ML	Supervised, Unsupervised, RL
3	Overfitting/Underfitting	Model complexity balance
4	Bias-Variance Tradeoff	Generalization control
5	Classification vs Regression	Output nature
6	Gradient Descent	Optimization algorithm
7	Activation Functions	Non-linearity in NN
8	Confusion Matrix	Model evaluation
9	CNN vs RNN	Spatial vs Sequential data
10	Reinforcement Learning	Reward-based learning