Fair Private and Distributed Evaluation of Crowd-sourced User Data
Rik Sarkar, Tariq Elahi, Hao Tsung Yang – University of Edinburgh
Our aim is to inform users of the value of their data. We envision a system where instead of simply asking for permission from a user to share data for analytics, our software computes an estimated a value for the actual user data in question, and informs the user of this value. The user can then make the decision to share the data, ask for compensation, etc.
Our approach will be to compute “value” as the contribution a user’s data makes to improving the machine learning or analytics used by the developer. The foundations of this approach spans machine learning theory and economics. We will develop a system that is robust to adversarial behaviour and can operate securely, while preserving privacy of user data. Our results will be based on theoretically sound concepts and have provable guarantees.
The impact of such a system will be fair compensation to users for their contributions. The compensation can be monetary, or in kind as use of the service. In either case, the transparency brought about by the system will enable a fair economy for both the user and the developer.