{"type":"ScholarlyArticle","title":"Transductive Optimization of Top k Precision","authors":[{"name":"Li-Ping Liu","slug":"li-ping-liu","given":"Li-Ping","family":"Liu","affiliation":""},{"name":"Thomas G. Dietterich","slug":"thomas-g-dietterich","given":"Thomas G.","family":"Dietterich","affiliation":""},{"name":"Nan Li","slug":"nan-li-2","given":"Nan","family":"Li","affiliation":""},{"name":"Zhi-Hua Zhou","slug":"zhi-hua-zhou","given":"Zhi-Hua","family":"Zhou","affiliation":""}],"abstract":"Consider a binary classification problem in which the learner is given a labeled training set, an unlabeled test set, and is restricted to choosing exactly k test points to output as positive predictions. Problems of this kind — transductive precision@k — arise in many applications. Previous methods solve these problems in two separate steps, learning the model and selecting k test instances by thresholding their scores. In this way, model training is not aware of the constraint of choosing k test instances as positive in the test phase. This paper shows the importance of incorporating the knowledge of k into the learning process and introduces a new approach, Transductive Top K (TTK), that seeks to minimize the hinge loss over all training instances under the constraint that exactly k test instances are predicted as positive. The paper presents two optimization methods for this challenging problem. Experiments and analysis confirm the benefit of incoporating k in the learning process. In our experimental evaluations, the performance of TTK matches or exceeds existing state-of-the-art methods on 7 benchmark datasets for binary classification and 3 reserve design problem instances.","doi":null,"url":"https://ijcai.org/proceedings/2016/255","pdf":"https://ijcai.org/proceedings/2016/0255.pdf","markdown":"https://ijcai.org/proceedings/2016/255.md","bibtex":"https://ijcai.org/proceedings/2016/bibtex/255","volume":{"slug":"2016","title":"Proceedings of the Twenty-Fifth International Joint Conference on Artificial Intelligence","short_title":"IJCAI 2016","url":"https://ijcai.org/proceedings/2016/","year":2016,"location":"New York, NY, USA","dates":"9–15 July 2016","editors":"Subbarao Kambhampati","isbn":"978-1-57735-770-4; 978-1-57735-771-1","publisher":"AAAI Press / International Joint Conferences on Artificial Intelligence","doi_prefix":""},"section":null,"track":"","number":255,"pages":{"first":1781,"last":1786},"keywords":[],"videos":[],"references":[],"license_note":"Copyright International Joint Conferences on Artificial Intelligence."}