Our paper Correntropy-based density-preserving data sampling as an alternative to standard cross-validation has been accepted for an oral presentation in the main track of the 2010 IEEE World Congress on Computational Intelligence (WCCI 2010) and for publication in the conference proceedings.

Our paper Ridge regression ensemble for toxicity prediction has been accepted for an oral presentation in the main track of the International Conference on Computational Science (ICCS 2010) and for publication in the conference proceedings.

The Smart Technology Research Center at the School of Design, Engineering and Computing, Bournemouth University organizes seminars on various artificial intelligence and machine learning topics. The seminars start on 10/02/2010 and are run on a weekly basis. Click here for a detailed schedule including the topics.

My talk entitled "Correntropy-based Density-preserving data sampling" is scheduled for 30/03/2010. The abstract can be found here.

The Environmental Toxicity Prediction Challenge has been organized by ICANN'09: International Conference on Artificial Neural Networks, European Neural Network Society (ENNS) and CADASTER project. The goals of this study were to:

• Develop in silico models to predict environmental toxicity of molecules against T. pyriformis using data from.
• Estimate the prediction intervals for new compounds.

The criteria for success were as given below:

• Firstly, methods with Root Mean Squared Error (RMSE) non-significantly different to the method with lowest RMSE (according to the bootstrap test, p<0.05) will be identified as the First-pass winners.
• Secondly, the method providing the best likelihood criteria between estimated and observed confidences for the blind test set as described in will be identified amid the first-pass winners.

The results have been announced today and our model was one of the First-pass winners and has been ranked on a 6th place from over 500 submissions, based on the blind test set prediction error. Taking into account both blind and known test errors, our method is ranked as the 3rd best model.