Foundations of AI: From Problem-Solving to Machine Learning

Learn how problems become search spaces and how initial states are explored using breadth-first, depth-first, and informed or uninformed search methods, including genetic algorithms, hill climbing, and simulated annealing.

Discover how heuristic functions guide branch selection in search trees within artificial intelligence. Learn Euclidean and Manhattan distance methods, eight-puzzle tile swaps, misplaced tiles, and tic-tac-toe win heuristics.

Analyze tic tac toe as a two-player game represented by a search tree, using a heuristic function (x's wins minus o's) to guide min-max decisions toward winning or drawing paths.

Learn about informed and uninformed searching algorithms in artificial intelligence, illustrated with the water and tic tac toe problems, and how heuristic functions guide informed search vs blind search.

Explore the breadth first search algorithm, a queue-based, uninformed method that traverses a search tree level by level to find the shortest path.

Depth first search uses a stack to explore deepest paths, backtracks when needed, and differs from breadth first search; iterative deepening and depth-limited search mitigate infinite loops.

Explore the water jug problem using bfs and dfs, building a search tree from initial states and applying fill, empty, and pour operations to reach a 2-liter goal.

Learn uniform cost search, a blind search that expands the lowest root-to-node path cost using a priority queue, a type of breadth-first search without heuristic guidance.

Enforce a fixed depth in depth-limited search to prevent infinite loops in depth-first search. Learn how a deepening algorithm balances exploration within a predefined depth.

Explore iterative deepening depth limited search, which starts at depth zero and increments toward infinity to reach the goal state, balancing depth limits and computational cost.

Explore bidirectional search in artificial intelligence, starting from both ends with two queues to meet at an intermediate node, improving efficiency over traditional breadth-first or depth-first searches.

Explore the generate and test searching algorithm with informed depth-first search, using heuristics to guide backtracking toward valid solutions such as the traveling salesman problem.

Learn best first search, an informed search using a heuristic to prioritize nodes in an open list, with a closed list to trace paths to a goal.

Best first search for the eight puzzle problem is explained, using a heuristic based on misplaced tiles to guide a search tree from the initial state toward the goal state.

Explore the A* searching algorithm as an informed search using path cost g(n) and heuristic h(n) to compute f(n) and guide best-first search, with eight-puzzle and graph examples.

Use the A* algorithm on the eight puzzle, applying heuristic functions like misplaced tiles or Manhattan distance with path cost to find the goal state via a search tree.

Explore the AO* searching algorithm for and-or graphs, using water jug and football examples to illustrate calculating path costs and heuristics, with backward recomputation of costs toward the goal.

Learn how genetic algorithms encode problems into chromosome structures, using binary or permutation encodings, and apply selection, crossover, mutation, and a fitness function to optimize solutions.

Explore a simple genetic algorithm to maximize f(x)=x^2+2x for x in 0–31, using binary encoding, population of four, roulette wheel selection, single-point crossover, and mutation considerations.

Explore hill climbing, a local search method that starts from a random point and climbs to nearby higher values, with variants to escape plateaus and local or global maxima.

Learn how simulated annealing works as a local search that heats to a peak and cools slowly, accepting worse moves with Boltzmann probability to escape local optima via energy function.

Local beam search expands k best successors at each level, guiding a search toward a goal state. It uses a fixed k to explore multiple states, unlike BFS or DFS.

Explain the minimax algorithm for two-player games, where max and min players recursively search a game tree using leaf values and infinity bounds to determine the root outcome.

Explore tic tac toe as a search problem with an initial empty board, two players x and o, winning lines, a min-max tree, and a heuristic guiding moves.

Explore alpha-beta pruning, an optimized variant of minimax that prunes branches using alpha and beta values during depth-first search to improve efficiency.

Explore the Wumpus world, a grid-based single-agent cave where breeze, stench, glitter, bump, and scream guide logical reasoning to grab gold, fire an arrow, and exit safely.

Learn how propositional logic builds a knowledge base from declarative statements using boolean values, negation, conjunction, disjunction, implies, truth tables, and parse trees.

Interpretation of propositional logic teaches assigning truth values to statements, evaluating them via truth tables, and identifying models and semantic entailment. Explore negation, conjunction, and implication via parse trees.

Learn to model agent reasoning in propositional logic using sensors, rules, and inference patterns such as modus ponens, modus tollens, and and-elimination to identify safe moves and apply resolution.

Apply the resolution algorithm to a knowledge base by converting rules to cnf and adding the negated goal, then derive a nullified state.

Learn predicate logic as first-order logic, with universal and existential quantifiers, predicates, and how to construct statements via examples like every student and all birds can fly.

Explore predicate logic statements using alphabets, terms, connectives, constants, and variables; form CNF formulas and distinguish binding from free variables with parse trees and quantifiers.

Define a model in predicate logic by fixing a universal set, predicates, and functions, then verify formulas with examples like state transition diagrams and natural numbers.

Explore the unification algorithm for first-order logic, using substitutions to convert statements to propositional form and applying lifting and resolution to derive conclusions.

Explore resolution algorithm, predicate logic, and CNF through concrete cricket score facts for Delhi vs Mumbai. Use unification and universal quantifiers to derive the winner.

Explore forward chaining in ai, a rule-based inference method that derives truth from a knowledge base using modus ponens and first-order logic.

Explore how inference derives new statements from a knowledge base using deduction, induction, and abduction, and how reasoning derives conclusions via common sense, hypothetical, and analogical methods.

Explore knowledge representation with semantic nets, a graphical structure of nodes and edges that capture relationships such as is a, has a, and owned by, revealing hierarchical connections among objects.

Explore knowledge representation using semantic nets, a graphical based representation with nodes and edges that encode relationships like dog is a mammal and bird can fly.

The lecture explains bayesian belief networks, showing how a directed acyclic graph models independent and dependent variables to infer rain, grass wetness, and related outcomes using joint and conditional probabilities.

Explore Dempster-Shafer theory, a belief and evidence framework for AI, handling uncertain information with mass, belief, and plausibility in sensor fusion and diagnostics.

Explore planning in artificial intelligence, defining the domain model, initial and goal states, and actions with preconditions and effects to reach a goal efficiently.

Learn how a simple planning agent uses a sensor to perceive the environment, choose actions with preconditions and effects, and execute a goal-directed plan. Note the four core assumptions.

The block world problem as a classical planning task in robotics, using the Strips language to model initial and goal states and actions like pick up, stack, and put down.

Apply mean end analysis, a goal-driven AI planning strategy, to reduce the gap between current and goal states. Choose the action that moves closest to the goal and apply preconditions.

Explore planning languages that express states, goals, and actions for automated reasoning with AI planners. Learn how STRIPS, ADL, and PDDL support preconditions, effects, quantifiers, and add/delete representations.

Learn how artificial neural networks imitate the biological brain, detailing neurons, weights, activation function, and synaptic gap to perform pattern recognition and classification.

Explore backpropagation networks as multilayer, fully connected, feedforward perceptron models, learn via supervised training, propagate errors, and update weights using the generalized delta rule.

Explore the backpropagation neural network, covering feedforward input, error backpropagation, and weight adjustment using the generalized delta rule. Learn how z, activation, targets, and weight updates drive learning.

Explore the backpropagation network algorithm, including activation functions—sigmoid, bipolar sigmoid, and hyperbolic tanh—along with feed-forward, error backpropagation, and weight updates in supervised learning.

Demonstrates a backpropagation network example with bipolar sigmoid activation, covering feedforward, backpropagation, error calculation, and weight updates across input, hidden, and output layers.