SearcharxivSearch

arXiv subjects

Ujjwal Kumar

Publications and source records attributed to Ujjwal Kumar.

4 recordsLinked to original sources

Strategy-Proof Probabilistic Social Choice Correspondences under Conditional Expected Utility

We study unanimous and strategy-proof probabilistic social choice correspondences (PSCCs), where the selected set of alternatives is interpreted as an interim outcome, and agents evaluate sets using conditional expected utility. We analyze two preference domains introduced by Barbera et al. (2001): the conditionally expected utility consistent (CEUC) domain and the conditionally expected utility consistent with equal probabilities (CEUCEP) domain. Our results characterize all unanimous and strategy-proof PSCCs on these domains and identify cases when randomization enlarges the class of admissible rules. On the CEUC domain, every unanimous and strategy-proof PSCC is a random dictatorship, showing that randomization over sets yields no additional flexibility. In contrast, the CEUCEP domain admits a richer family of unanimous and strategy-proof PSCCs. For at most three agents, these rules are precisely the random bi-dictatorial rules, which are convex combinations of bi-dictatorial rules introduced in Feldman (1980). For four or more agents, the characterization depends on the number of alternatives. When there are exactly three alternatives, the class expands to the larger family of coalition-weighted rules. Thus, randomization enlarges the class of strategy-proof correspondences in the three-alternative case, producing rules that are not convex combinations of deterministic strategy-proof correspondences. However, for four or more alternatives, the class of unanimous and strategy-proof probabilistic correspondences again collapse to random bi-dictatorships.

econ.TH

On the existence of EFX allocations for goods

We consider a set $E$ of $m$ indivisible goods and a set $N$ consisting of $n \geq 2$ agents. The paper shows that if two agents have \textit{arbitrary} set monotonic valuation functions and the remaining agents have size monotonic valuation functions, then EFX allocations always exist.

econ.TH

DMM: Distributed Matrix Mechanism for Differentially-Private Federated Learning Based on Constant-Overhead Linear Secret Resharing

Federated Learning (FL) solutions with central Differential Privacy (DP) have seen large improvements in their utility in recent years arising from the matrix mechanism, while FL solutions with distributed (more private) DP have lagged behind. In this work, we introduce the distributed matrix mechanism to achieve the best-of-both-worlds; better privacy of distributed DP and better utility from the matrix mechanism. We accomplish this using a novel cryptographic protocol that securely transfers sensitive values across client committees of different training iterations with constant communication overhead. This protocol accommodates the dynamic participation of users required by FL, including those that may drop out from the computation. We provide experiments which show that our mechanism indeed significantly improves the utility of FL models compared to previous distributed DP mechanisms, with little added overhead.

cs.CR

Tracking dispersion measure variations of timing array pulsars with the GMRT

We present the results from nearly three years of monitoring of the variations in dispersion measure (DM) along the line-of-sight to 11 millisecond pulsars using the Giant Metrewave Radio Telescope (GMRT). These results demonstrate accuracies of single epoch DM estimates of the order of 5x10^(-4) cm^(-3) pc. A preliminary comparison with the Parkes Pulsar Timing Array (PPTA) data shows that the measured DM fluctuations are comparable. We show effects of DM variations due to the solar wind and solar corona and compare with the existing models.

astro-ph.HE