AI foundations, Problem Solving and Search

• The foundations of AI do not include astrology.
• The Turing Test assesses a machine's ability to exhibit intelligent behavior equivalent to humans.
• In AI, a rational agent maximizes its performance measure based on percepts and knowledge.
• An unobservable environment is when an agent does not receive any percepts from the world.
• Medical diagnosis systems use expert systems for decision-making in AI applications.

Module 2: Problem Solving

• Informed search strategies use heuristics to guide the search, differentiating them from uninformed search.
• Breadth-first search expands the shallowest node first.
• A* search requires a heuristic function that is admissible and consistent.
• Bidirectional search proceeds from both the start and goal states simultaneously.
• A hill-climbing search algorithm is known to get stuck in local optima.

Module 3: Search in Complex Environments

• The Minimax algorithm assumes the opponent always makes optimal moves.
• Alpha-beta pruning improves Minimax by reducing unnecessary calculations.
• Backtracking in a Constraint Satisfaction Problem (CSP) is used to undo previous variable assignments when a dead end is reached.
• Arc consistency in CSPs ensures that each variable has at least one valid value.
• The most constrained variable heuristic in CSPs selects the variable with the fewest remaining legal values.

Module 4: Knowledge Representation and Reasoning

• Propositional Logic has the limitation of lacking expressiveness for complex relationships.
• Unification is an inference method that uses the process of substituting variables to make expressions identical.
• First-order logic differs from propositional logic by using variables, functions, and quantifiers.
• The existential quantifier (∃) in first-order logic indicates that a statement is true for at least one value.
• A disadvantage of backward chaining is that it requires a goal to start.

Module 5: Machine Learning

• In supervised learning, the training data consists of input-output pairs for learning a mapping function.
• Decision trees split the data based on the attribute that minimizes entropy or maximizes information gain.
• Overfitting occurs in machine learning when the model performs well on training data but poorly on new data.
• The goal in linear regression is to minimize the sum of squared residuals.
• A disadvantage of using logistic regression for classification is that it cannot model complex, nonlinear relationships without modifications.