Open-Set Support Vector Machines

Often, when dealing with real-world recognition problems, we do not need, and often cannot have, knowledge of the entire set possible classes that might appear during operational testing. In such cases, need to think robust classification methods able deal "unknown" properly reject samples belonging never seen training. Notwithstanding, existing classifiers date were mostly developed for closed-set scenario, i.e., setup in which it is assumed all test belong one classifier was trained. open-set however, a sample can none known must by classifying as unknown. this work, extend upon well-known Support Vector Machines (SVM) introduce Open-Set (OSSVM), suitable setups. OSSVM balances empirical risk unknown ensures region feature space would be classified (one classes) always bounded, ensuring finite also highlight properties SVM related provide necessary sufficient conditions an RBF have bounded open-space risk.