Publication Plan

Thesis

AAAI ROC Curves as Dataset Signatures

• Using ROC Curves? as dataset signatures. Predicting the dataset based on ROC curves from other models. The overall procedure is:
  • Train several models as in Cross Validation. Compute the ROCDataset?. Use it as a feature.
  • Compare to a vector of standard performance metrics: Accuracy?, True Positive, False Positive, Precision, Recall, Area Under ROCCurve?.

• Results for using performance data only are not very good, ROCClustering Results?.
• Get better prediction results using ROCClassification Results?
• Paper: AAAI07 will discuss this and the possibility for doing regression on ROCDataset?s, as in the current experiments in MOBLRegression Procedures? and Regression Error Analysis

Mixture SVM

ICML, Mixture SVM

• Implemented a custom SVM using mixture model.
• Plan is to complete experiements by 1-18-2006.

Omni Tree

ECML, Omni Tree

• Use Rademacher Complexity to decide which model to use for the tree.

Cougar^2

PKDD

• Design document and comparison to Weka.