JAWAHARLAL NEHRU TECHNOLOGICAL UNIVERSITY HYDERABAD B.Tech III MACHINE LEARNING Previous Question Paper (August - 2022 - R18)

Designing a machine learning system involves addressing several fundamental challenges to ensure the model learns effectively from data. The core design issues include:

1. Choosing the Training Experience: Determining the nature of the data (direct or indirect feedback) and whether the learner controls the sequence of training examples. For instance, a game-playing agent might learn from its own moves (direct) or from the final win/loss outcome (indirect).
2. Choosing the Target Function: Defining exactly what is being learned. In a checkers program, the target function could be the board state evaluation function, which assigns a numerical value to any given board position.
3. Choosing a Representation for the Target Function: Selecting how to represent the learned function (e.g., linear polynomials, neural networks, or decision trees). A linear combination of board features is a common representation for checkers.
4. Choosing an Approximation Algorithm: Deciding how to adjust the parameters of the representation to best fit the training data. This often involves optimization techniques like gradient descent to minimize the error between predicted and actual outcomes.

Approaches to machine learning are generally categorized based on the nature of the learning signal:

• Supervised Learning: The model is trained on labeled data where the correct output is known. Example: Predicting house prices based on historical sales data (regression) or classifying emails as spam or not spam (classification).
• Unsupervised Learning: The model identifies hidden patterns or structures in unlabeled data. Example: Customer segmentation based on purchasing behavior where no predefined categories exist.
• Reinforcement Learning: The agent learns by interacting with an environment, receiving rewards or penalties for its actions. Example: Training a robot to navigate a maze or an AI to play video games by maximizing a cumulative reward signal.