Human-in-the-loop learning (HIL): Analysis and Mechanisms

Mobile users in existing information systems face limited resources such as data plan, transmission bandwidth, computation capacity, and location-based information. Given that some users may not be utilizing fully their resources, human-in-the-loop resource sharing is a promising approach to exploit users' diversity in resource use and for pooling their resources or information. In this seminar, I first review that traditional resource allocation solutions are mostly centralized without considering users' local connectivity constraints, becoming not scalable for large-scale sharing. In addition, there may be sharing failure since selfish users will not truthfully report their actual valuations and quantities for buying or selling resources. Accordingly, I present a fast-greedy algorithm based on maximum weighted matching to achieve guaranteed average allocative efficiency. Then, we combine it with a fully distributed pricing mechanism that adjusts the final trading prices for buying and selling resources in a way that buyers and sellers are incentivized to truthfully report their valuations and available resource quantities. 

Besides human-in-the-loop resource sharing mechanisms, I also present human-in-the-loop mechanisms to regulate information sharing among mobile users in congestion/routing games. In Google Maps and Waze, selfish users use myopic routing to jam each other, without learning time-varying traffic conditions of the non-shortest paths for future users. Our study extends the traditional congestion games fundamentally to create positive information learning generated by users themselves. I show that the myopic routing leads to arbitrarily large efficiency loss and propose a selective information disclosure mechanism to bound the loss well.