Chinthalapati, V L Raju (2012) Learning from noisy data and Markovian processes. Submitted. (Unpublished)

Preview

PDF (Preprint) 13366_Chinthalapati_Learning_from_noisy_data_(submitted_version)_2012.pdf - Submitted Version Download (198kB)

Abstract

We discuss more realistic models of computational learning. We extend the existing literature on the Probably Approximately Correct (PAC) framework to finite Markov chains in two directions by considering: (1) the presence of classification noise (specifically assuming that the training data has currupted labelled examples), and (2) real valued function learning. In both cases we address the key issue of determining how many training examples must be presented to the learner in the learning phase for the learning to be successful under the PAC paradigm.

Item Type:	Article
Uncontrolled Keywords:	PAC Learning, Noisy Data, VC dimension, Classification Noise, Markovian Process, Real-valued and Boolean-valued Function Learning.
Subjects:	H Social Sciences > HG Finance
Faculty / School / Research Centre / Research Group:	Faculty of Business Faculty of Business > Department of Accounting & Finance
Last Modified:	12 Nov 2019 11:18
URI:	http://gala.gre.ac.uk/id/eprint/13366

Actions (login required)

View Item

Downloads