Abstract
Increasing access to large, non-stationary face datasets and corresponding demands to process, analyze and learn from this data. This has lead to a new class of online/ incremental face recognition problems. While it is advantageous to build large scale learning systems when resources permit, a counter problem of learning with limited resources in presence of streaming data arises. We present a budgeted incremental support vector learning method suitable for online learning applications. Our system can process one sample at a time and is suitable when dealing with large streams of data. We discuss multiple budget maintenance strategies and investigate the problem of incremental unlearning. We propose a novel posterior probability estimation model based on Extreme Value Theory (EVT) and show its suitability for budgeted online learning applications (calibration with limited data). We perform thorough analysis of various probability calibration techniques with the help of methods inspired from meteorology. We test our methods on Labeled Faces in the Wild dataset and show suitability of the proposed approach for face verification/ recognition.