SVM Technology Wins NSF-Sponsored Challenge at the World Congress for Computational Intelligence 2008

SVM Technology Wins NSF-Sponsored Challenge at the World Congress for Computational Intelligence 2008

Posted by Rebecca Martin on Thu, 12/06/2008 - 16:09

Health Discovery Corporation (OTCBB: HDVY), a leader in support vector machine (SVM) based molecular diagnostics development today announced that Yin Wen Chang, a student of Chih-Jen Lin from the National Taiwan University, distinguished herself in the first causality challenge organized for the World Congress on Computational Intelligence, WCCI 2008, which was held in Hong Kong, June 1-6, 2008.

The challenge, which is sponsored by the U.S. National Science Foundation, the European Network of Excellence PASCAL, and Microsoft Corporation, attracted over 50 participants. Yin Wen Chang ranked first on two tasks of the challenge and second and third on the two others, using SVM both for feature selection and for classification.

The four tasks proposed to the competitors were derived from real data in genomics, pharmacology and econometrics. The goal of the challenge was to uncover causes of a given outcome in order to make predictions of the result of future actions. For example, find genes to cure disease, find risk factors to control epidemics. Uncovering causes superficially resembles the problem of feature selection. But most feature selection algorithms emanating from machine learning like RFE-SVM do not seek to model mechanisms: they do not attempt to uncover cause-effect relationships between feature and target.

“We did not expect non-causal feature selection methods to do so well on these tasks,” explained Dr. Isabelle Guyon, co-organizer of the challenge and a member of HDC’s Science Team. “Causal discovery methods did very well at unraveling causal structure, and on average, we observed good correlation between the fraction of causally relevant features selected and the predictive power of learning machines on the tasks of the challenge. Yet, non-causal feature selection methods like RFE-SVM find feature subsets containing complementary features with high predictive power and SVMs are insensitive to the presence of false positive, so this is a combination that’s very hard to beat.”

“Solving problems of causality in order to predict the results of future actions is a critical component of identifying the right drug at the right dose for the right patient and is the cornerstone for the successful implementation of personalized medicine,” stated Stephen D. Barnhill, M.D., Chairman and CEO of Health Discovery Corporation. “We are thrilled that once again SVM technology has proven to be superior to other mathematical algorithms in solving these very difficult and unique problems. We congratulate Yin Wen Chang for her great accomplishment using SVMs to win the NSF-Sponsored Challenge at the World Congress for Computational Intelligence 2008.”

Dr. Barnhill continued, “With 32 issued patents around SVM technology and the only issued patents in the world on RFE-SVM, Health Discovery Corporation is in a unique position to capitalize on the proven success of these techniques to create and commercialize new diagnostic tests and play a significant role in bringing the promises of personalized medicine to reality.”

Further Infomation available at