CSE455 Learning and Prediction II: MML Data Mining

Plan
The half-semester course is to follow on the heels of its main pre-requisite, CSE454. (CSE454 plan and CSE454's progress in 2002.)
(A plan for CSE454 and an initial plan for CSE455.)
Foundations of inductive inference from (algorithmic) information theory; intermediate to advanced Minimum Message Length (MML) inference; details of Fisher information and uncertainty regions; angular/circular models (von Mises, Wrapped Normal or trigonometric); Poisson distribution; MML of specific models such as decision graphs, hidden Markov models (or HMMs, also known as probabilistic finite state automata, or PFSAs), linear and polynomial regression, causal models, Bayesian nets, time series, sequences, segmentation, trends; probabilistic prediction and Kullback-Leibler distance. Statistical invariance, statistical consistency. Data mining. Additional models may include factor analysis and additional theory may include the Neyman-Scott problem (1948). Applications to be considered may include: models of protein folding and protein structure prediction, bushfire prediction, text and image analysis, DNA alignment and the human genome project, authorship identification for texts, etc. Further typical applications may be described.

Might also manage to fit in some or all of: polygon modelling; DNA pattern discovery and alignment, evolutionary trees; Lempel-Ziv text compression, C.S. Wallace improvement (1989, 1996), approximate repeats; HMMs (PFSAs) in mixture modelling; Markov fields, images.

Progress
The half-semester course followed on the heels of its main pre-requisite, CSE454, whose 7th and last lecture was on Tuesday 23rd April 2002.