About Me
I am a member of the Machine Learning Group in the University of Waikato. I am currently comlpeting a PhD. My thesis is titled: "scalable multi-label classification"; my supervisors are: Bernhard Pfahringer and Geoff Holmes.
My CV (July 2010) -- Post-doctoral position candidate.
Contact Information
Jesse READ
| E-mail: | jmr30@cs.waikato.ac.nz |
| Phone: | +64 7 838 4466 ext. 8766 |
| Fax: | +64 7 838 5095 |
| Address: |
Machine Learning Lab (G.2.11) Department of Computer Science The University of Waikato Private Bag 3105 Hamilton 3240 New Zealand |
Research
Multi-label Classification
Multi-label classification is the supervised classification context where each data point may belong to multiple labels, as opposed to a single class label like in the traditional multi-class problem. Multi-label classification is quite natural to many domains, such as text categorisation, scene and video classification, medical diagnosis, and biological applications (protein function classification and genomics).
One of the main issues involved in multi-label classification is the importance of detecting and incorporating correlations between labels into the learning process. A second and related issue is the additional complexity involved in multi-label learning, as compared to single-label learning. I am interested in developing general multi-label methods which are both high-performing and can scale up to large datasets.
Data Streams
Many multi-label applications are found in the context of data streams, where data points arrive rapidly and continuously in a theoretically-infinite stream. Examples of data sources include sensor data, network traffic, social networking posts, forum and newsgroup posts, news articles, RSS feeds, and large deployments of e-mail. In the data stream context, the standard pool-based train/test scenario becomes inappropriate, and incremental methods become necessary to learn and make predictions in real time. Data streams are almost always inevitable subject to concept drift: methods must adapt to changes in the data over time.
Publications
Jesse Read, Albert Bifet, Geoff Holmes, Bernhard Pfahringer. Efficient Multi-label Classification for Evolving Data Streams. Technical Report 2010/04. University of Waikato. New Zealand. March 2010.
Jesse Read, Bernhard Pfahringer, Geoff Holmes, Eibe Frank. Classifier Chains for Multi-label Classification. In Proc. of 20th European Conference on Machine Learning (ECML 2009). Bled, Slovenia, September 2009. [slides] [software]
Jesse Read, Bernhard Pfahringer, Geoff Holmes. Generating Synthetic Multi-label Data Streams. In Proc. of ECML/PKKD 2009 Workshop on Learning from Multi-label Data (MLD'09). Bled, Slovenia, September 2009. [poster] [software]
Jesse Read, Bernhard Pfahringer, Geoff Holmes. Multi-label Classification using Ensembles of Pruned Sets. Proc. of IEEE International Conference on Data Mining (ICDM 2008). Pisa, Italy, December 2008. (long version) [slides] [software]
Jesse Read. A Pruned Problem Transformation Method for Multi-label Classification. In Proc. of the NZ Computer Science Research Student Conference. Christchurch, New Zealand (2008). [slides]
Daniel Kuen Seong Su, Victoria Siew Yen Yee, and Jesse Read. Exploring Text-based and Graphical-based Usable Interfaces for Mobile Chat Systems. eMinds International Journal on Human-Computer Interaction. Vol. I(3), (2007).
Jesse Read. Filtering Spam with Machine Learning. Honours Thesis, University of Waikato, Hamilton, New Zealand. (2005)
Presentations
"Efficient Multi-label Classification". Internal Doctoral Conference. University of Waikato, November 2009
"Methods for On-line Multi-label Classification". Internal Doctoral Conference. University of Waikato, December 2008
"On-line Multi-label Classification". Universitat Politècnica de Catalunya, Departament de Llenguatges i Sistemes Informàtics, Barcelona, October 2008
"Ensembles of Nested Dichotomies for Multi-label Classification". Machine Learning Group, Department of Computer Science, University of Waikato, Hamilton, New Zealand, July 2008
"Online Hierarchical Multi-label Classification". Mixed Reality Lab, Department of Computer Science, University of Nottingham, U.K., September 2007
Software
I have moved the MEKA software to sourceforge.net where it is being further developed.
Datasets
Datasets are now at sourceforge.net