Shailesh Vaya

Assignment Techniques for Crowdsourcing Sensitive Tasks

Protecting the privacy of crowd workers has been an important topic in crowdsourcing, however, task privacy has largely been ignored despite the fact that many tasks, e.g., form digitization, live audio transcription or image tagging often contain sensitive information. Although assigning an entire job to a worker may leak private information, jobs can often be split into small components that individually do not. We study the problem of distributing such tasks to workers with the goal of maximizing task privacy using such an approach. We introduce information loss functions to formally measure the amount of private information leaked as a function of the task assignment. We then design assignment mechanisms for three different assignment settings: PUSH, PULL and a new setting Tug Of War (TOW), which is an intermediate approach that balances flexibility for both workers and requesters. Our assignment algorithms have zero privacy loss for PUSH, and tight theoretical guarantees for PULL. For TOW, our assignment algorithm provably outperforms PULL; importantly the privacy loss is independent of the number of tasks, even when workers collude. We further analyze the performance and privacy tradeoffs empirically on simulated and real-world collusion networks and find that our algorithms outperform the theoretical guarantees.

Distributed communication in bare-bones wireless networks

We consider wireless networks in which the effects of interference are determined by the SINR model. We address the question of structuring distributed communication when stations have very limited individual capabilities. In particular, nodes do not know their geographic coordinates, neighborhoods or even the size n of the network, nor can they sense collisions. Each node is equipped only with its unique name from a range {1, ..., N}. We study the following three settings and distributed algorithms for communication problems in each of them. In the uncoordinated-start case, when one node starts an execution and other nodes are awoken by receiving messages from already awoken nodes, we present a randomized broadcast algorithm which wakes up all the nodes in O(n log2 N) rounds with high probability. In the synchronized-start case, when all the nodes simultaneously start an execution, we give a randomized algorithm that computes a backbone of the network in O(Δ log7 N) rounds with high probability. Finally, in the partly-coordinated-start case, when a number of nodes start an execution together and other nodes are awoken by receiving messages from the already awoken nodes, we develop an algorithm that creates a backbone network in time O(n log2 N + Δ log7 N) with high probability.

Round complexity of leader election and gossiping in bidirectional radio networks

We consider the setting of ad hoc radio networks when the underlying network is bidirectional, the number of nodes in the network n is known and nodes can be assigned labels which are polynomially large in n. For this setting we present a protocol for deterministic gossiping which takes O(nlg^2nlglgn) rounds improving upon the previous best result by [L. Gasieniec, A. Pagourtizis, I. Potapov, Deterministic gossiping in radio networks with large labels, Algorithmica 47 (2007) 97-117], who give a complex protocol with the round complexity of O(nlg^3nlglgn), resolving an open problem posed by Gasieniec in the survey article [L. Gasieniec, On efficient gossiping in radio networks, in: Proceedings of Sixteenth International Colloquium on Structural Information and Communication Complexity, SIROCCO 2009, Sirince, Turkey, in: LNCS, vol. 5869, 2010, pp. 2-14]. The relationship between the asymptotic round complexity of deterministic Leader Election and Gossiping problem has not been known even for bidirectional networks. The gossiping protocol given by Gasieniec et al. invokes the leader election protocol O(lgn) times. Our gossiping protocol achieves this by a single invocation to the leader election protocol. We then use a known fact about the lower bound for leader election for bidirectional radio networks to conclude that the asymptotic round complexity of deterministic Leader Election and Gossiping is the same for bidirectional networks.

Information dissemination in unknown radio networks with large labels

We consider the problems of deterministic broadcasting and gossiping in completely unknown ad-hoc radio networks. It is assumed that nothing is known to the nodes of the network about the topology of the network, that is even the size of the network is not known, except that n>1. This lack of knowledge, about the value of n, is what distinguishes this setting from the vanilla model. For this setting, protocols for the vanilla model, may be executed with multiplicatively large estimates, say 2^i in the ith phase, on the upper bound on the size of the network n. When the respective protocol with estimate 2^i>=n on the size of the network is run, it will accomplish the task successfully. However, the problem that still remains is to determine when this process should terminate. Thus, to apply this design paradigm successful completion or incompletion of the process should be detected and this knowledge circulated in the network after appropriate number of rounds/phases of the protocol. In radio networks literature, this setting is known as the Acknowledged setting and broadcasting and gossiping problems for it are referred to as Acknowledged broadcasting and gossiping. An important feature of dynamic radio networks is that radio nodes can be dynamically introduced in the network from time to time and can be assigned labels in a much larger range, say polynomial in the size of the network, e.g. [1,...,n^c] for some constant c. It is easy to see that protocols can be designed for the acknowledged setting only when the underlying communication network is strongly connected. We present the following results for these networks: (a) A deterministic protocol for Acknowledged broadcasting which takes NRG(n,n^c) rounds, where NRG(n,n^c) is the round complexity of deterministic gossiping for vanilla model. (b) A deterministic protocol for acknowledged gossiping, which takes O(n^2lgn) rounds when collision detection mechanism is available. The schedule of the transmissions of nodes in the network, to enable them to infer collisions and discover existence of unknown in-neighborhood as a result, is abstracted as a family of sets of natural numbers which we call the Selecting-Colliding family. We prove the existence of Selecting-Colliding families using the probabilistic method and employ them to design protocol for acknowledged gossiping when no collision detection mechanism is available. Finally, we present a deterministic protocol for Acknowledged broadcasting for bidirectional networks, with a round complexity of O(nlgn) rounds.

Brief announcement: delay or deliver dilemma in organization networks

Organization networks are hierarchical trees and were proposed by Papadimitrious and Schreiber to model interactions in an organization. Packets arrive from the outside world at (possibly intermediate) nodes of this directed rooted tree and are to be forwarded to the root. A fixed delivery cost is charged every time a link is used and a delay cost is charged in proportion to the total time packets wait in the network before they reach the root. This is a natural setting. While the asynchronous distributed setting has been studied rigorously, it allows arbitrary large number of packets to arrive at nodes of the network. This is unrealistic. This work proposes a more sophisticated model, called the Bounded Bandwidth Model, which restricts the maximum number of packets that can be sent on a delivery to a constant M. For a complete binary tree of height k, we present a distributed online algorithm with a competitive ratio of O(min (√M, k)). We also give a constant competitive strategy for flat tree networks, a competitive strategy for arbitrary tree topologies and a lower bound for any oblivious distributed online algorithm for serial networks.

Brief Announcement: Multi-Broadcasting under the SINR Model

We study the multi-broadcast problem in multi-hop ad-hoc wireless networks, under the SINR model, deployed in the 2D Euclidean plane. In multi-broadcast, there are

Robust Reputation Mechanisms for Achieving Fair Compensation and Quality Assurance in Crowdcomputing

k

Method and system for providing access to crowdsourcing tasks

initial messages, potentially belonging to different nodes, that must be forwarded to all

Multi-Broadcasting under the SINR Model.

n

Deterministic Backbone Creation in an SINR Network without Knowledge of Location.

nodes of the network. We present deterministic algorithms for multi-broadcast for the different settings reflecting the different types of knowledge, about the topology of the network, available to the nodes: (i) the whole network topology (ii) their own coordinates and coordinates of their neighbors (iii) only their own physical coordinates (iv) only their own ids and the ids of their neighbors, but not actual physical co-ordinates. The last result is specially interesting, as it is the first deterministic protocol, for the SINR model, that does not require nodes to know their coordinates in the plane (a very specialized type of knowledge), but intricately exploits the fact that they lie on a two dimensional plane.