Multi-Output Prediction: Theory and Practice

Page 1. Multi-Output Prediction: Theory and Practice Inderjit S. Dhillon UT Austin & Amazon PAKDD 2020 Singapore May 13, 2020 Inderjit S. Dhillon UT Austin & Amazon Multi-Output Prediction: Theory and Practice Page 2. Outline 1 Multi-Output Prediction Modern Prediction Problems Linear Prediction Inductive Matrix Completion (IMC) Positive-Unlabeled Learning 2 Prediction for Enormous and Correlated Output Spaces (PECOS) PECOS: Three-Stage Framework Deep Learned Neural Matcher Experimental Results 3 Conclusions and Future Work Inderjit S. Dhillon UT Austin & Amazon Multi-Output Prediction: Theory and Practice Page 3. Modern Prediction Problems Spam detection Character Recognition Regression Analysis Inderjit S. Dhillon UT Austin & Amazon Multi-Output Prediction: Theory and Practice Page 4. Single-Output Regression …