A short description
Active learning is a promising strategy to actively querying experts and obtaining the actual labels for
particular examples. Its goal is decreasing the number of labels needed to achieve high level of accuracy. In
this paper, we present a novel active learning algorithm for regression problems when training samples are
connected by a set of edges to form a network. These edges are often characterized by a form of correlation,
so that the labels of linked nodes are (auto-)correlated. In this situation, collective inference is a powerful
paradigm to learn how to exploit label correlations in the network and to improve the performance of
predictive models on network data. However, it needs access to the labels of linked nodes. Since labels
are typically scarce throughout a network, it is rarely the case that labels of linked nodes are known. We
introduce a novel active learning regression algorithm, named CoNeRa, to handle the problem of missing labels and provide
the collective inference with sufficient supervision for learning label correlations. It uses collective inference,
in order to accommodate label correlations in the active selection of network nodes for labeling.
The distribution package
The CoNeRa algorithm and its competitors (Random, DAL, PAL (Pasolli et al.2012, Douaka et al. 2012) and AlfNet (Biljic et al 2010)) have been implemented in Java systems.
They have been evaluated with regression problems formulated on eleven social and spatial data networks.
| file |
Description |
| README |
A description file |
| Conera,AlfNet,Random,DAL,PAL |
This rar bundle contains jar files and example of executions |
| Data |
This rar bundle contains 11 social and spatial datasets of the comaprative analysis |
| Data | This rar bundle contains forest fires dataset |
Warning:
The CoNeRa Learner (as well as the implementation of its competitors) is free for evaluation, research and teaching purposes, but not for commercial
purposes.
Please Acknowledge
Top of
this page
Related publications
Appice, A., Loglisci, C., Malerba, D.
Active learning via collective inference in network regression problems (2018) Information Sciences, 460-461, pp. 293-317.
M. Bilgic, L. Mihalkova, L. Getoor, Active learning for networked data, in: J. Furnkranz, T. Joachims (Eds.), Proceedings
of the 27th International Conference on Machine Learning, ICML 2010, Omnipress, 2010, pp. 79-86.
E. Pasolli, F. Melgani, N. Alajlan, Y. Bazi, Active learning methods for biophysical parameter estimation, IEEE Transactions
on Geoscience and Remote Sensing 50 (2012) 4071-4084.
F. Douaka, F. Melgania, N. Alajlanc, E. Pasollia, Y. Bazic, N. Benoudjitb, Active learning for spectroscopic data regression,
Journal of Chemometrics 26 (2012) 374-383.
Top of this page
Project team
Dr. Annalisa APPICE
Dr. Corrado LOGLISCI
Prof. Donato MALERBA
Contact
| Name |
Email address |
Tel. number |
Fax |
| Annalisa Appice |
annalisa.appice@uniba.it |
+39 080 5443262 |
+39 080 5443262 |
Top of this page