Sunita Sarawagi, IIT Bombay
In many classification problems like node labelling in Social networks, a core hurdle to any prediction effort is that for many samples there is insufficient evidence for inferring a label, resulting in low overall accuracy. We propose that instead of providing high accuracy over all samples, we make those predictions which will be correct with high probability. ...
Brian Guenter, MSR Redmond
I developed a proprietary application specific neural network framework for an undisclosed use case. ...