Large Alphabet Sequence Compression and Text Modelling -&&- Combination of Probability and Logic
Wen Shao -&&- Hadi Afshar (with Pizza)
ARTIFICIAL INTELLIGENCE SEMINAR PhD MonitoringDATE: 2013-05-22
TIME: 11:30:00 - 12:30:00
LOCATION: RSISE Common LHS Room A203 ANU
CONTACT: JavaScript must be enabled to display this email address.
ABSTRACT:
Abstract1: Most raw data is not binary, but over some often large and structured 'alphabet' $mathcal{X}$. On one hand, data can easily be binarised and many data compression techniques are based on binarised data sequence; on the other hand, in practice data compression/modelling exploiting the original structure of the data can crucially improve performance. In this work, we perform a comparative study to investigate the relationship of binary data compression with modelling natural language (unbinarised, large alphabet), experimentally compare approaches and transfer ideas between them.
Abstract2: First/Higher order logics provide expressive knowledge representation formalisms that can be used to exploit structure but they cannot deal with uncertainty. On the other hand, classic probabilistic approaches use less expressive and less flexible knowledge representations (e.g. propositional) but they can handle uncertainty. Hutter et al. 2012 is a theory in a line of research that seeks to integrate logical and probabilistic models of reasoning.
This Particular theory has interesting properties such as the ability to confirm universally quantified hypotheses (which approaches such as Bayes/Laplace rule or Carnap's confirmation theory fail to address). I seek to develop this theory by trying to address the following: 1- Develop approximation schemes for the currently incomputable aspects of the general theory. 2- Get optimal priors (maybe based on Solomonoff induction). 3- Develop a formal (incomplete, approximate) reasoning calculus. 4- Study the relation between the concepts mentioned in this paper with other concepts such as omega-consistency.
BIO:
http://people.cecs.anu.edu.au/user/4416
http://people.cecs.anu.edu.au/user/4691
