| MACHINE LEARNING (Full Module: CS5505) Overview - definitions of learning, history of machine learning, classification of machine learning methods. Simple concept learning - Winston's ARCH, candidate elimination (version space). AQ, induction of decision trees - ID3, noise handling, overfitting & pruning. Theory of induction - inductive bias, generalisation/specialisation operators, constructive induction. Unsupervised learning - conceptual clustering, probabilistic classification, UNIMEM, COBWEB, category utility. Instance-Based Learning - memory-based reasoning, k-nearest neighbour rule, prototypes. Bayesian learning - Bayes theorem, Naive Bayes (simple Bayesian classifier), Bayesian networks. Case Study: Learning models of text categorisation. Studying learning algorithms - empirical evaluation, learning curves, n-fold cross validation, characteristics of datasets. Connectionist Methods - Perceptron, linear threshold unit (LTU), linear separability, backpropagation and the generalized delta rule. Connectionist Methods - constructive learning procedures - cascade correlation, competitive learning - Kohonen nets. Case Study: Connectionist methods for face recognition. Genetic Algorithms - adaptive search techniques, simple genetic algorithm, notion of fitness, mating, genetic operators - crossover, mutation. Analytic/speedup learning - MACROPS, LEX, EBG, EBL, utility problem. Reinforcement Learning - Q-learning. Analogy - definition of analogical reasoning, target & source, analogy as access/mapping/evaluation/learning, structure mapping theory/structure mapping engine, derivational analogy. Case-Based Reasoning - generic CBR algorithm, indexing, memory structures, cases vs. rules, PROTOS, case-based planning - CHEF, CBR vs analogy. Multistrategy learning - KBANN, EITHER. First-order learning - first-order representations, limitations of attribute-based algorithms, generalisation/ specialisation operators. First-order learning - LINUS, FOIL, FOCL. Studying learning algorithms - computational learning theory, PAC analysis. Knowledge refinement - overview of refinement techniques, KRUST, REFINER, etc. | |
