Latest news Latest news

Girls with top science GCSEs 'deterred from study at higher level'

Below is an interesting report  about why girls don't take STEM subjects at higher levels.(Spoiler: too many boys, lack of confidence) Visit: Girls top science gcse deterred study higher...

Use of Big Health and Actuarial Data Impact Workshop

Prof. Elena Kulinskaya is the PI on the ARC programme on Use of Big Health and Actuarial Data for understanding Longevity and Morbidity risks. The research programme held a one day Impact...

Smarter maths generates better value insurance products

Lens on Research and Innovation  (LORI) was selected by EPSRC for their showcase of impact.  Visit: www.epscr.ac.uk/lori  

Congratulations to CMP Head of School Professor Gerard Parr on his MBE

Prof Gerard Parr, Head of the School of Computing Sciences, has been made an MBE in the New Year Honours, for his services to developing telecoms infrastructure in Northern Ireland (NIR)....

CMP and Earlham Institute researchers develop new phylogenetics software package

Together with researchers in the Earlham Institute, members of Phylogenetics group in CMP have developed SPECTRE, a new open-source software package, that simplifies the complex business of...

Latest Events and Seminars Latest Events and Seminars

Back

Empirical Evaluation of Semi-Supervised Naive Bayes for Active Learning

Date and time: 11th October 13:00-14:00

Location: Queens 1.03

Title: Empirical Evaluation of Semi-Supervised Naive Bayes for Active Learning

Speaker: Awat Saeed

Institution: School of Computing Sciences, UEA

Organiser: Dr. Michal Mackiewicz

 

Abstract

Active learning aims to minimise the amount of labelled data required by using the model to direct labelling of the most informative unlabelled examples to give the greatest improvement in the subsequence generation of model, but the key difficulty with active learning is that the initial model often gives a poor direction for labelling the unlabelled data in the early stages. However, using both labelled and unlabelled data with semi-supervised learning might result in a better initial model. In this talk, we describe an empirical evaluation of semi-supervised and active learning individually, and in combination for the naive Bayes classifier. For this purpose, a suite of benchmark datasets are used and the learning curves for experiments to compare the performance of each approach are presented. First, we show that the semi-supervised naive Bayes does not improve the performance of the naive Bayes classifier. Subsequently, a down-weighting technique is used to control the influence of the unlabelled data, but again this does not improve performance. In the next experiment, a novel algorithm is proposed by using a sigmoid transformation to recalibrate the overly confident prediction of the naive Bayes classifier. This algorithm does not significantly improve on the naive Bayes classifier, but at least does improve the semi-supervised naive Bayes classifier somewhat. In the final experiment we investigate the effectiveness of the combination of active and semi-supervised learning and empirically illustrate when the combination does work, and when does not.