arXiv subjects

Debmalya Mandal

Publications and source records attributed to Debmalya Mandal.

39 records · Page 3Linked to original sources

arXiv2017

Peer Prediction with Heterogeneous Tasks

Peer prediction is a method to promote contributions of information by users in settings in which there is no way to verify the quality of responses. In multi-task peer prediction, the reports from users across multiple tasks are used to score contributions. This paper extends the correlated agreement (CA) multi-task peer prediction mechanism to allow the reports from users to be on heterogeneous tasks, each associated with different distributions on responses. The motivation comes from wanting to elicit user-generated content about places in a city, where tasks vary because places, and questions about places, vary. We prove that the generalized CA mechanism is informed truthful under weak conditions, meaning that it is strictly beneficial for a user to invest effort and acquire information, and that truthful reporting is the best strategy when investing effort, as well as an equilibrium. We demonstrate that the mechanism has good incentive properties when tested in simulation on distributions derived from user reports on Google Local Guides.

cs.GT↗

arXiv2017

Calibrated Fairness in Bandits

We study fairness within the stochastic, \emph{multi-armed bandit} (MAB) decision making framework. We adapt the fairness framework of "treating similar individuals similarly" to this setting. Here, an `individual' corresponds to an arm and two arms are `similar' if they have a similar quality distribution. First, we adopt a {\em smoothness constraint} that if two arms have a similar quality distribution then the probability of selecting each arm should be similar. In addition, we define the {\em fairness regret}, which corresponds to the degree to which an algorithm is not calibrated, where perfect calibration requires that the probability of selecting an arm is equal to the probability with which the arm has the best quality realization. We show that a variation on Thompson sampling satisfies smooth fairness for total variation distance, and give an $\tilde{O}((kT)^{2/3})$ bound on fairness regret. This complements prior work, which protects an on-average better arm from being less favored. We also explain how to extend our algorithm to the dueling bandit setting.

cs.LG↗

arXiv2015

Profit Maximizing Prior-free Multi-unit Procurement Auctions with Capacitated Sellers

In this paper, we derive bounds for profit maximizing prior-free procurement auctions where a buyer wishes to procure multiple units of a homogeneous item from n sellers who are strategic about their per unit valuation. The buyer earns the profit by reselling these units in an external consumer market. The paper looks at three scenarios of increasing complexity. First, we look at unit capacity sellers where per unit valuation is private information of each seller and the revenue curve is concave. For this setting, we define two benchmarks. We show that no randomized prior free auction can be constant competitive against any of these two benchmarks. However, for a lightly constrained benchmark we design a prior-free auction PEPA (Profit Extracting Procurement Auction) which is 4-competitive and we show this bound is tight. Second, we study a setting where the sellers have non-unit capacities that are common knowledge and derive similar results. In particular, we propose a prior free auction PEPAC (Profit Extracting Procurement Auction with Capacity) which is truthful for any concave revenue curve. Third, we obtain results in the inherently harder bi-dimensional case where per unit valuation as well as capacities are private information of the sellers. We show that PEPAC is truthful and constant competitive for the specific case of linear revenue curves. We believe that this paper represents the first set of results on single dimensional and bi-dimensional profit maximizing prior-free multi-unit procurement auctions.

cs.GT↗