As computing tasks become increasingly complex, the search for advanced methods to improve and enhance automization becomes more and more pressing. A recent day-long seminar at the IBM Haifa Labs examined the current research and applications of machine learning. This is a field in which data-driven techniques are developed to enable computers to learn and enhance performance automatically through experience.
Organized by the Machine Learning Group in the Haifa research facility, the seminar attracted 130 representatives from universities, start-ups, global high-tech companies, and government institutions. Seminar presentations focused on innovation that matters, presenting a blend of machine learning core technologies and applications in such fields as verification, search, life sciences, and ecommerce.
"Israeli firms and universities are applying a vast range of machine learning techniques," noted Shai Fine, leader of machine learning activities in the IBM Haifa Labs, and a co-organizer of the seminar, together with Yishay Mansour of Tel Aviv University.
"Israeli scientists are studying the use of machine learning in diverse fields," noted Shai. "In addition to bioinformatics applications, a popular domain of exploitation for machine learning in academia and industry, researchers are examining the use of machine learning technology in such areas as information retrieval, Web services, search engines, text analysis, natural language analysis, and most recently simulation-based hardware verification. Several speakers at the seminar presented the work they are doing in these areas."
A highlight of the seminar was the keynote address delivered by Vladimir Vapnik, one of the founding fathers of the machine learning field. Considered by many to be the most important figure in this area, Vapnik reviewed the main innovations and breakthroughs of machine learning since the early 60's, highlighting new trends and technologies developed in the last decade. Currently at NEC Laboratories and Columbia University, Vapnik believes the field is moving from an era of deductive reasoning, in which data was used to build models for predicting further data, to an era of transductive reasoning. In transductive reasoning, direct conclusions are made about future data based on present data, without constructing a model as a middle step. This developing trend will allow machine learning to develop more accurate applications for a wide variety of fields.
Although geared for Israeli researchers and IT professionals, the seminar garnered a good deal of international attention. Most notable was the official external recognition of the event by PASCAL, the European network-of-excellence that focuses on pattern analysis, statistical modeling and computational learning.
To review the abstracts or see the presentations from the seminar, go to the seminar Web site at http://www.haifa.il.ibm.com/Workshops/ml2005/index.html.