Question 1

Supervised vs Unsupervised vs Reinforcement Learning?

Accepted Answer

1.  **Supervised Learning**: The model learns from **labeled data** (Input-Output pairs).
    *   *Goal*: Predict outcomes for new data.
    *   *Examples*: Spam filter, House price prediction.
2.  **Unsupervised Learning**: The model learns from **unlabeled data**.
    *   *Goal*: Find hidden patterns or structures.
    *   *Examples*: Customer segmentation, Anomaly detection.
3.  **Reinforcement Learning**: An agent learns to make decisions by performing actions and receiving **rewards or penalties**.
    *   *Goal*: Maximize cumulative reward.
    *   *Examples*: Game playing AI (AlphaGo), Robot navigation.

Question 2

Precision vs Recall - When to use which?

Accepted Answer

These metrics evaluate classification models, especially with imbalanced datasets.

*   **Precision**: Out of all predicted positives, how many were actually positive? (Quality)
    *   *Formula*: TP / (TP + FP)
    *   *Use*: When **False Positives** are costly (e.g., Email Spam detection - don't want to lose important email).
*   **Recall (Sensitivity)**: Out of all actual positives, how many did we find? (Quantity)
    *   *Formula*: TP / (TP + FN)
    *   *Use*: When **False Negatives** are dangerous (e.g., Cancer diagnosis - don't want to miss a case).

Question 3

What is Overfitting and how do you prevent it?

Accepted Answer

**Overfitting** occurs when a model learns the training data (and its noise) too well, resulting in poor performance on unseen data (high variance).

### Prevention Techniques:
1.  **Cross-Validation**: Use K-Fold cross-validation.
2.  **Regularization**: Add a penalty term to the loss function (L1/Lasso, L2/Ridge).
3.  **Pruning**: For decision trees.
4.  **Dropout**: For neural networks (randomly drop neurons during training).
5.  **Early Stopping**: Stop training when validation loss starts increasing.
6.  **More Data**: Increase the size of the training set.

Question 4

What is Gradient Descent?

Accepted Answer

**Gradient Descent** is an iterative optimization algorithm used to minimize a cost function (loss function).

### Concept:
Imagine you are on a mountain at night (high cost) and want to reach the valley (low cost).
1.  Check the slope (gradient) at your current position.
2.  Take a step in the opposite direction of the slope (downhill).
3.  Repeat until the slope is zero (minimum).

The size of the step is determined by the **Learning Rate**. Too large = overshoot; too small = slow convergence.

Question 5

What is the Bias-Variance Tradeoff?

Accepted Answer

It describes the problem of minimizing two sources of error that prevent supervised learning algorithms from generalizing beyond their training set.

*   **Bias**: Error due to overly simplistic assumptions (Underfitting). The model misses relevant relations.
*   **Variance**: Error due to excessive sensitivity to small fluctuations in the training set (Overfitting). The model models noise.

**Goal**: Find the "Sweet Spot" with low bias and low variance.

Question 6

What is a Confusion Matrix?

Accepted Answer

A **Confusion Matrix** is a table used to describe the performance of a classification model.

| | Predicted: No | Predicted: Yes |
|---|---|---|
| **Actual: No** | True Negative (TN) | False Positive (FP) |
| **Actual: Yes** | False Negative (FN) | True Positive (TP) |

It allows calculating Precision, Recall, Accuracy, and F1-Score.

Question 7

What is Transfer Learning?

Accepted Answer

**Transfer Learning** is a technique where a model developed for a task is reused as the starting point for a model on a second task.

*   *Example*: Using a pre-trained ImageNet model (like ResNet) that knows how to recognize edges and shapes, and fine-tuning it to recognize specific medical X-rays.
*   *Benefit*: Saves massive amounts of time and compute power; requires less training data.

Question 8

What are Activation Functions and why are they needed?

Accepted Answer

**Activation Functions** decide whether a neuron should be activated or not. They introduce **non-linearity** into the output of a neuron.

### Why needed?
Without non-linearity, a Neural Network would just be a big linear regression model, unable to learn complex patterns (like images or audio).

### Common Types:
1.  **Sigmoid**: 0 to 1. (Vanishing gradient problem).
2.  **ReLU (Rectified Linear Unit)**: 0 to infinity. Most common.
3.  **Softmax**: Probability distribution (output layer).

Question 9

What is NLP (Natural Language Processing)?

Accepted Answer

**NLP** is a branch of AI that gives computers the ability to understand, interpret, and manipulate human language.

### Common Tasks:
*   Sentiment Analysis.
*   Machine Translation (Google Translate).
*   Named Entity Recognition (NER).
*   Text Summarization.

### Modern Approaches:
Moved from Rule-based -> Statistical -> **Deep Learning (Transformers, BERT, GPT)**.

Question 10

Explain K-Means Clustering.

Accepted Answer

**K-Means** is an **unsupervised** learning algorithm used to partition observations into **k** clusters.

### Algorithm Steps:
1.  Choose **k** random centroids.
2.  Assign each data point to the nearest centroid.
3.  Recompute the centroid of each cluster (mean of points).
4.  Repeat steps 2-3 until centroids stop changing.

AI & Machine Learning

1Supervised vs Unsupervised vs Reinforcement Learning?

2Precision vs Recall - When to use which?

3What is Overfitting and how do you prevent it?

4What is Gradient Descent?

5What is the Bias-Variance Tradeoff?

6What is a Confusion Matrix?

7What is Transfer Learning?

8What are Activation Functions and why are they needed?

9What is NLP (Natural Language Processing)?

10Explain K-Means Clustering.