Artificial Intelligence and Machine Learning Course

Explore the foundations of artificial intelligence, covering the course outline, representation and search techniques (state space search and heuristic search), machine learning, logical reasoning, and rule-based programming for decision making.

Define artificial intelligence as science and engineering of making intelligent machines, per John McCarthy, focusing on intelligent agents that perceive environments and act to maximize success, including Turing test.

Explore how intelligent agents perceive environments, map percept sequence to action via agent function and program, and realize rational behavior in simple and complex decision making.

Explore state space search for problem solving in artificial intelligence, covering states, state space, representation schemas, and search methods like depth-first, breadth-first, backtracking, with goal driven and data driven approaches.

Explore how state space search uses graph theory, illustrated by the Königsberg bridges, converting locations to nodes and bridges to labeled arcs, with and/or operators decompose subproblems.

Explore state space search by modeling problems with nodes, arcs, and start and goal states, illustrated by the eight puzzle and traveling salesman problem.

explores finite state machines, states and inputs, and their transition graphs; introduces data-driven and goal-driven search, and compares depth-first and breadth-first search using an illustrative bfs algorithm.

Perform breadth-first search on a graph by expanding the leftmost open node level by level, enqueuing its children, and closing explored nodes until the goal state is found.

Demonstrate depth first search by using a stack for the open list, pushing children to the left, moving explored nodes to the closed list, and stopping at the goal.

Trace depth first search on a graph and apply iterative deepening with depth bounds to guarantee the shortest path, comparing to breadth first search.

Explore backtracking algorithms that systematically explore state space paths from a start state to a goal, using depth-first backtracking, handling dead ends, and pruning the path.

Trace the backtracking algorithm on a graph by applying the pseudocode to a start node, manage sl, nsl, and cs, and backtrack from dead ends to explore siblings.

Explore trace backtracking on a graph, managing current state, NSL, SL, and dead ends. Follow goal checks, children generation, and updates to reveal the path g c.

Trace the backtracking algorithm in state space search, comparing current state to the goal, generating children, and backtracking from dead ends to explore the search space.

Explore heuristic search methods for problem solving, including hill climbing, simulated annealing, best-first search, and min-max with alpha-beta pruning, plus evaluating heuristics in reducing state space.

Apply tic-tac-toe heuristics to reduce state space and guide the search toward the best solution, using f(n) = g(n) + h(n) where h is the heuristic value guiding progress.

Explain three heuristics for the eight puzzle: tiles out of place, sum of distances, and direct tile reversals, and combine them with g(n) to compute f(n) for state expansion.

Explore hill climbing as a heuristic search that evaluates a node's children to select the best path, and understand limits like local maxima, plateaus, and alternatives such as simulated annealing.

Apply best first search by ordering the open list with a heuristic h, maintain a closed set, and use a priority queue to reach the goal via the shortest path.

Trace the best-first search on a graph, expanding nodes by heuristic values, maintaining open and closed lists, until reaching the goal P.

This lecture explains best first search on the eight puzzle, detailing open and closed states, frontier, and the f(n)=g(n)+h(n) heuristic guiding state expansion.

Explore best-first search using f(n)=g(n)+h(n), where h(n) counts tiles out of place, and select the lowest f to expand, yielding the admissible, optimal a star path.

Learn the minimax algorithm for game playing, including forming a game graph, labeling levels as min and max, and propagating leaf heuristics up to select optimal moves.

The lecture explains heuristic measuring conflict in tic tac toe, where e(n) equals the difference between my and the opponent's possible wins, and applies two-ply minimax to the opening move.

Apply alpha beta pruning to the min max search to reduce exploration by using alpha for max nodes and beta for min nodes, with depth-first evaluation.

Explore the fundamentals of artificial intelligence with machine learning, covering neural networks, unsupervised learning, gradient descent, and k-nearest neighbor learning, plus logic functions via McCulloch-Pitts models.

Explore how the perceptron extends the McCulloch-pitts neuron with weights and a threshold, applying supervised learning and a delta-w update to minimize error until coverage.

Examine perceptrons for linearly separable data by computing f(net) from x1, x2 and weights to draw a separating line; XOR proves single-layer perceptrons cannot solve nonlinearly separable data.

Backpropagation with multilayer networks propagates errors backward through hidden layers, updating weights via the delta rule and gradient descent using a sigmoid activation and mean squared error.

Learn to compute delta w for hidden nodes using back propagation, deriving error terms for hidden and output units, and update weights and biases via a 16-step algorithm.

Initialize weights, propagate inputs forward through the network, compute errors, and update weights and biases via backpropagation with logistic activation and learning rate.

Demonstrates backpropagation steps for computing I5 and I6 from multiple previous nodes, and derives hidden-layer outputs using a sigmoid activation.

Explains backpropagation calculations for hidden and output layers, computing i4, i5, i6 and o4, o5, o6, then deriving output and hidden layer errors using weights like w46 and w56.

Update weights and bias using the delta rule with learning rate 0.9, and outline supervised learning as classification and unsupervised learning as clustering, including clustering goals and k-means.

Explore k-means clustering and key machine learning concepts, from unsupervised and supervised learning to instance-based and reinforcement learning, and their applications in biology, medicine, finance, and robotics.

Explore logics and reasoning, including propositional calculus and predicate calculus, and learn how logical inference derives conclusions from truthful premises using a knowledge base and an inference engine.

Explore propositional calculus symbols, atomic propositions p, q, r, s, and truth values true and false, with rules for negation, conjunction, disjunction, implication, and equivalence, and truth tables.

Convert English sentences to propositional calculus by assigning propositions and forming implications. Apply negation, contrapositive, and De Morgan's laws, and verify proofs via truth tables and state-space reasoning.

Delve into predicate calculus, extending propositional logic with predicates, variables, and quantifiers. Learn the symbolic alphabet, atomic sentences, and how assignments and scope define interpretations.

Explore the first order predicate calculus and how quantified variables refer to objects in the domain of discourse. Map English sentences to calculus using implies, negation, and universal/existential forms.

Represent family relationships using predicate calculus, defining parents and siblings, and assess satisfiability, consistency, and validity. Apply modus ponens, modus tollens, and universal instantiation to derive conclusions.

Design and implement a reasoning system with an inference engine and knowledge base, using unification to match predicate expressions and apply resolution for deduction.

Learn how to prove by resolution by converting statements to clause form, negating the theorem, and resolving to an empty clause; includes complementary literals and clause form logic.

Transform English statements into predicate form, convert to clause form, apply negation to implications, and use resolution refutation to derive a contradiction, as demonstrated by the John example proving happiness.

Convert English statements into predicate and clause forms, then apply resolution refutation to determine whether anyone can have an exciting life.

Demonstrate how to apply resolution refutation to a predicate logic problem about an exciting life, transforming statements into predicate, clause, and class formats, and deriving a contradiction to prove existence.

This lecture demonstrates unification substitution and skolemization within resolution refutation to derive where Fido is and that John has a grandparent, illustrating clause formatting, negation, and answer extraction.

Explore Prolog as a logic programming language and its predicate calculus, including horn clauses and resolution. Examine deductive, inductive, abductive, analogical, and non-monotonic reasoning.

Explore model-based reasoning with theoretical device models and simulations, and case-based reasoning using a problem-solution database to retrieve, modify, apply the transform case, and record outcomes.

Explore rule-based programming and production systems that model human problem solving with facts in short-term memory and rules in long-term memory, enabling inference and pattern-driven problem solving.

Trace a simple production system by applying the rules ba -> ab, ca -> ac, and cb -> bc to a starting memory, iterating substitutions and resolving conflicts.

Explore the three by three knight problem on a 3x3 chessboard, derive moves and production rules, and solve the path from 1 to 2 via 1-8-3-4-9-2.

Explore rule matching and conflict resolution in data driven and goal driven expert systems, refraction, recency, specificity, forward and backward chaining, production rules, and working memory in CLIPS and Prolog.

Explore data driven and goal driven production systems for diagnostics, using working memory and rule firing to diagnose engine problems such as gas presence, turning over, or spark plugs.

Compare data driven and goal driven approaches, detailing how branching factor affects search space, and illustrate with clips rule-based programming and production system principles.

Learn rule based programming with Clips: define rules and facts, watch and reset data, run and save clips, declare variables, and implement hypotenuse function, exploring inference engines for expert systems.

Learn how to install and launch CLIPS on Windows or Linux, access the user manual, and begin working with facts, rules, and commands like assert, clear, reset, and exit.

Master clips basics by managing facts database: assert color green or red, use facts command, retract or clear facts, and explore grouped assertions like (A B C) with f0 indexing.

Learn to assert facts in CLIPS with case sensitive entries and spaces, then retract, watch memory with watch commands, and reset to initial facts.

Explore clips basics by building if-then rules from facts, using diff rule syntax, and asserting actions to fire rules and produce a sound like quack when the animal is duck.

Explore deferral and diff rule commands in CLIPS, inserting facts and defining if-then rules. Learn how two conditions trigger a duck assertion and troubleshoot syntax errors.

Learn how CLIPS uses facts, insert and assert rules, and chaining to infer sound quack from animal duck, plus loading rules from .clp files.

Learn to save and load a CLP file by creating and testing simple facts like animal is duck and sound is quack, then load, run, and verify results.

Learn how clips handles facts across resets by reasserting facts, retracting, and using patterns with wildcards and variables to match and print results.

Explore CLIPS tutorial 8: define and assert facts, apply a diff rule with multiple patterns to identify the mammal by matching animal, warm-blooded, and egg-laying traits.

Explore pattern matching with variables in rules to infer mammal facts from animal data, including dog as a mammal and puppy via child of relationships.

Learn how to retract facts in a CLIPS diff rule by binding to a fact variable, identify mammal facts, and execute retractions with arithmetic evaluation.

Apply multi-field wildcards to match 'member of' facts, capture band and members, and print 'a band called' lines based on eclipse facts.

Explore wildcard matching in a rule-based database by extracting and printing band members and band names using dollar question mark wildcards for single and multiple members.

Examine variables in rule-based processing using the bind function to create temporary values. Practice asserting and retracting facts, tracing number additions, and exploring global variables, templates, and conditions.

Demonstrates using def template and dev templates to enroll two people's data, with single slots for name, age, weight, height, and multi-slots for blood pressure ranges.

Leverage deffacts and deftemplates to assert and modify personal data, updating slots like name, age, weight, height, and blood pressure, with retraction for truth maintenance.

Learn to write rules with conditional elements, using and/or logic to evaluate weights and print the name, age, and weight of matching facts.

Apply the not operator in the Clips programming language to negate a predicate, using personal data (name and weight) to check birthday and print when not today.

Use the test element to check conditions on a rule's left-hand side, including exists, and use assert to add facts and print results from the personal data.

Explore how to use exist and for all in a diff rule to query a fact database, print results, and troubleshoot syntax for accurate outputs.

Create a multi-slot personal data template, assert facts, and manage date of birth slots to explore truth and control and truth maintenance with logical operators.

Explore diff rules and diff templates, troubleshoot why facts are not created in the defect statement context, and practice personal data conditions and assertions for cardiac risk and smoker data.

Explore how intelligent agents perceive environments via sensors, form percept sequences, and map inputs to actions using agent functions and programs, with simple and complex decision making shaping rational behavior.

Explore how a simple reflex agent uses current perceptions and condition-action rules to select actions, with examples from vacuum cleaning and car braking, and its limits in partial observability.

Explore how a model based agent uses an internal state to track percept history, handle partial observability, and update state through perception, model, and rules to choose actions.

Explore goal based agents that reason with current state and explicit goals, contrasting with reflex and model based agents, and examine search, planning, and flexible decision making.

Learn how utility-based agents maximize expected utility via probability and utilities, bridging decision theory with Bayesian networks and decision networks for rational behavior under uncertainty.

Explore the basics of utility theory, including utility functions and expected utility, and how rational agents maximize outcomes under uncertainty via six axioms: orderability, transitivity, continuity, substantiality, monotonicity, decomposability.

Compute expected utility of actions across states using probabilities and utilities to select the maximum, illustrated by one- and two-action examples; discuss the value of information and utility axioms.

Explore decision theory by combining probability and utility to form rational agents, and represent decision problems with decision networks that integrate chance, decision, and utility nodes for the expected outcome.

Explore reinforcement learning where an agent learns from rewards to maximize outcomes. The lecture covers Markov decision processes, partially observable MDPs, dynamic decision networks, and sequential decision making.

Explore Markov decision processes, policies, and optimal policy concepts; compare fully observable MDPs with partially observable MDPs and introduce a dynamic decision network and game theory for multi-agent decision making.

Explore probabilistic reasoning under uncertainty and compare it with logic-based approaches, including inductive, analogical, abductive, probabilistic, temporal, and fuzzy reasoning, plus rule-based locality, detachment, and truth functionality.

Explore basics of set theory, including union and intersection, inclusion-exclusion for three sets, Cartesian product, the multiplication principle, sample space and elementary events, probability axioms, and counting rules.

Explore random variables and probability distributions, including boolean, discrete, and continuous types; apply binomial, Poisson, and normal models, and use joint, conditional, and independence rules.

Learn the bayesian rule for conditional probability, including joint and conditional probabilities, bayes theorem, and how these tools apply to scenarios like marbles and coin tosses.

Apply Bayes theorem to real-world examples, compute conditional probabilities like P(D2|A1) from dealer data, and use Bayes classification to infer gender from names.