Generative AI

---

In supervised learning, the model is trained on labeled data. This means the training data includes input-output pairs, and the goal is to learn a mapping from inputs to outputs. Common algorithms include linear regression, logistic regression, and support vector machines.

Classification:  It is a type of supervised learning where the goal is to predict discrete labels (categories) for given inputs.

Example: Email Spam Detection

The model is trained on a labeled dataset of emails, where each email is marked as “spam” or “not spam.” The goal is to classify new emails into these categories based on learned patterns.

Regression:  It is a type of supervised learning where the goal is to predict continuous values for given inputs.

Example: House Price Prediction

The model is trained on a dataset of house features (e.g., size, number of bedrooms, location) and their corresponding prices. The goal is to predict the price of a house based on its features.

In unsupervised learning, the model is trained on unlabeled data. The system tries to learn the underlying structure of the data without specific input-output pairs. Common techniques include clustering (e.g., k-means, hierarchical clustering) and association (e.g., Apriori algorithm).

Clustering: Customer Segmentation

A retail company uses clustering to group customers based on purchasing behavior. This helps in identifying distinct customer segments for targeted marketing.

Association: Market Basket Analysis

Analyzing transaction data to find associations between products. For example, if customers often buy bread and butter together, the store might place these items close to each other.

Dimensionality Reduction: Principal Component Analysis (PCA) for Image Compression

PCA is used to reduce the number of features in an image dataset while preserving as much variance as possible, making it easier to store and process.

Reinforcement Learning (RL) involves training an agent to make a sequence of decisions by rewarding it for good actions and penalizing it for bad ones. The agent learns to achieve a goal by maximizing cumulative rewards. Key algorithms include Q-learning and deep Q-networks (DQN).

Q- Learning:  It is a value-based Reinforcement Learning algorithm that aims to find the optimal action-selection policy by learning the Q-values (quality values) for state-action pairs.

Deep Q-Networks (DQN): It extend Q-Learning by using deep neural networks to approximate the Q-values, making it feasible to handle large state spaces where maintaining a Q-table is impractical.

Example: Game Playing AI

A reinforcement learning agent is trained to play a game like chess or Go. It learns by playing many games, receiving rewards for winning and penalties for losing, ultimately improving its strategy over time.

Weights are the parameters within a neural network that transform input data within the network’s layers. They represent the strength of the connection between nodes (neurons).

Example: In a neural network predicting house prices, a weight might determine how much importance to give to the size of the house. If the weight is high, the size heavily influences the prediction.

Example: Health Care - Predicting Diabetes

In this neural network, a weight might determine how much importance to give to the glucose level when predicting diabetes. If the weight is high, glucose level heavily influences the prediction.

Biases are additional parameters in a neural network that allow the activation function to be shifted to the left or right. They help the model fit the data better by providing flexibility.

Example: In the same house price prediction network, a bias might be added to ensure that even with no input (e.g., a house of zero size, which is unrealistic), the model outputs a value that makes sense in the context of the problem.

Example: Health Care - Predicting Diabetes

A bias might be added to ensure that even with low or zero values for certain features, the model outputs a value that makes sense in the context of diabetes prediction.

Activation functions introduce non-linearity into the model, allowing it to learn more complex patterns. They determine the output of a neuron.

Common Activation Functions:

• Sigmoid: Outputs a value between 0 and 1.
• ReLU (Rectified Linear Unit): Outputs the input directly if it is positive; otherwise, it outputs zero.
• Tanh: Outputs a value between -1 and 1.

Example: Using ReLU in the house price model helps capture non-linear relationships, like sudden jumps in price due to certain features (e.g., having a pool).

Example: Health Care - Predicting Diabetes

Using the Sigmoid function in the output layer to predict the probability of diabetes.

 The loss function measures how well the neural network’s predictions match the actual data. It provides a way to quantify the error in predictions.

Common Loss Functions:

• Mean Squared Error (MSE): Used for regression tasks.
• Cross-Entropy Loss: Used for classification tasks.
• Binary Cross-Entropy Loss: Used for binary classification tasks like predicting diabetes (positive or negative).

Example: For house price prediction, Mean Squared Error might be used, calculated as the average of the squared differences between predicted and actual prices.

Example: Health Care - Predicting Diabetes

Binary Cross-Entropy might be used to quantify the difference between the predicted probability of diabetes and the actual outcome (0 for no diabetes, 1 for diabetes).

Gradient Descent is an optimization algorithm used to minimize the loss function by iteratively adjusting the weights and biases in the direction that reduces the loss.

Example: In our model, Gradient Descent would adjust the weights and biases to minimize the difference between predicted house prices and actual prices.

Example: Health Care - Predicting Diabetes 

Gradient Descent would adjust the weights and biases in the model to minimize the difference between the predicted probability of diabetes and the actual outcomes

Back Propagation is the process of computing the gradient of the loss function with respect to each weight by the chain rule, and it helps in updating the weights.

Example: In back propagation, the error from the output layer (difference between predicted and actual price) is propagated backward through the network to update the weights. This ensures that each weight adjustment reduces the overall error.

Example: Health Care - Predicting Diabetes 

In back propagation, the error from the output layer (difference between predicted and actual diabetes outcomes) is propagated backward through the network to update the weights and biases.