CSE 592 Topics of Study

Applications

The class will be motivated by the following sampling of real-world applications which involved successful AI techniques.

• Autonomous control of complex embedded systems (e.g. the NASA space probe Deep Space One, to be launched later this year).
• Software agents, information integration systems and declarative web site specification.
• Decision Support Systems (e.g. medical diagnosis, software help systems)
• Datamining (e.g. prediction of credit card fraud, analysis of manufacturing log data)

Core Algorthms and Ideas

In the course of underatnding the applications described above, we'll cover a wide range of ideas, datastructures and algorithms that form the core foundation of AI techniques. The emphasis will be on proven algorithms that have been shown to work.

• Search
  • Brute force
    Depth first, breadth first, iterative deepening, iterative broadening
  • Heuristic
    best first, beam, hill climbing, limited discrepancy
  • Optimizing
    Branch & bound, A*, IDA*, SMA*
  • Constraint Satisfaction
    As search, preprocessing, backjumping, forward checking, dynamic variable ordering
• Knowledge Representation & Reasoning
  • Propositional Logic
    Syntax, inference: modus ponens, clausal form, resolution, unit propagation & linear resolution, davis putnam (DPLL), stochastic methods, GSAT, WalkSAT; truth maintenance, hard SAT problems and phase transitions
  • Predicate Calculus Syntax, inference: unification, resolution, compilation to propositional form
  • Modal Logic
    Representing action and change, diagnosis, planning
  • Bayesian Belief Networks
    Joint probability distribution, indepenence, Bayes rule, inference and diagnosis: clustering methods, cutset conditioning methods, stochastic simulation; utility functions, decision networks
• Learning
  • Supervised Learning
    Methodology: N-fold cross-validation; induction of decision trees, entropy, information gain, ID3, C4.5, overfitting, ensembles of classifiers, bagging, boosting, scaleup, complete classification vs nuggets, feature selection: LOOCV, racing, schemas; inductive logic programming, FOIL, minimum description length, Grendel
  • Reinforcement Learning
    Cummulative discounted reward, dynamic programming, policy iteration, value iteration, temporal-difference learning
  • Bayesian Learning

  ---