SCS Undergraduate Thesis Topics
|Woo Tae Kim||Pragnesh Modi||Classification of Examples by Multiple Agents with Private Features|
We consider the problem of classification where relevant features are distributed among a set of agents and cannot be centralized, for example due to privacy restrictions. Accurate prediction of the output class is difficult for an isolated single agent because the target concept may involve features to which the agent does not have access. To increase prediction accuracy, a learning algorithm is required in which agents collaborate to classify new examples, while preserving the privacy of their local features. We formalize this problem as the distributed classification task. We introduce a novel distributed decision-tree inspired algorithm for such tasks named DDT. One of the key ideas in DDT is that agents can communicate the information gain of a private feature without revealing the semantics of the feature or its actual value. We present empirical results in a calendar management domain where software assistant agents classify new meetings as .likely to be difficult to schedule. using private features such as each attendee.s willingness to attend the meeting. We show empirically that our approach outperforms a single agent learner and performs as good as a centralized learner with hypothetical access to all the features.