以機器學習方法進行互動驗證

Abstract

驗證問題是一個有很多應用且需要使用人力的問題。機器學習可以減少花費在驗證問題上的人力。透過結合驗證問題中的學習和驗證兩個階段，我們提出一個稱做``互動驗證'的新問題。這個新問題可以藉著自由分配學習和驗證來更有效的運用人力。我們提出使用情境式拉霸問題 (Contextual Bandit Problem) 中的上信賴界 (Upper Confidence Bound) 方法來解決互動驗證問題。在真實世界資料上的實驗結果證實了上信賴界可以有效的解決互動驗證問題

The verification problem comes with many applications and requires human efforts. Machine learning can help reduce human efforts spent on verification. By combining the learning and verification stages in a verification problem, we formalize the needs as a new problem called interactive verification. The problem allows an algorithm to flexibly use the limited human resource on learning and verification together. We propose to adopt upper confidence bound (UCB) algorithm, which has been widely used for the contextual bandit, to solve the interactive verification problem. Experiment results demonstrate that UCB has superior performance on interactive verification on many real-world datasets.