Cathy H. Xia
Research Subjects
Cathy H. Xia has a total of 19 co-authors. Here are the top co-authors by number of shared publications:
- Zhen Liu
IBM T.J. Watson Research Center, Hawthorne, NY 10532, USA
4 shared publications George Michailidis
Department of Statistics, University of Michigan, Ann Arbor, MI
2 shared publicationsLi Zhang
IBM T.J. Watson Research Center
2 shared publications- Li Zhang
IBM T.J. Watson Research Center
2 shared publications Mark S. Squillante
IBM T.J. Watson Research Center
2 shared publications- Parijat Dube
IBM Research
2 shared publications
Latest Publications
- 17 September 2008
- Published by Springer Nature in Annals of Operations Research
- Vol. 170 (1) , 161-182
A decentralized control mechanism for stream processing networks
- 17 September 2008
- Published by Springer Nature in Annals of Operations Research
- Vol. 170 (1) , 161-182
A decentralized control mechanism for stream processing networks
- 1 December 2007
- Published by Elsevier in Computers & Operations Research
- Vol. 34 (12) , 3541-3559
Competitive equilibrium in e-commerce: Pricing and outsourcing
- 1 December 2007
- Published by Elsevier in Computers & Operations Research
- Vol. 34 (12) , 3541-3559
Competitive equilibrium in e-commerce: Pricing and outsourcing
- 1 November 2007
- Published by SAGE Publications in Production and Operations Management
- Vol. 16 (6) , 701-712
Dynamic Pricing in e‐Services under Demand Uncertainty
High volatility of the e‐services market, due to increasing competition, low life cycle of products, and easy availability of information about competing service offerings to customers, makes the demand for service offerings quite uncertain. Revenue management in such markets calls for real‐time techniques to learn the demand and its dependence on both the price and the service level associated with the service offering. We assume firms reply on exploratory approaches for demand estimation, in which firms experiment with different service offerings in order to simultaneously learn the demand while doing business. Such exploration and learning process can be costly without supervision. As reported by Rothschild (Journal of Economic Theory, 9 185‐202, 1974), traditional Bayesian dynamic control approaches may conclude with suboptimal offerings. We propose a novel demand learning approach that is guaranteed to converge to the optimal offering. The approach combines simulated annealing algorithm with Bayesian learning. We further present intelligent techniques that adaptively reduce the effort of exploration on suboptimal service offerings so as to improve the long‐run average profit.
- 1 November 2007
- Published by SAGE Publications in Production and Operations Management
- Vol. 16 (6) , 701-712
Dynamic Pricing in e‐Services under Demand Uncertainty
High volatility of the e‐services market, due to increasing competition, low life cycle of products, and easy availability of information about competing service offerings to customers, makes the demand for service offerings quite uncertain. Revenue management in such markets calls for real‐time techniques to learn the demand and its dependence on both the price and the service level associated with the service offering. We assume firms reply on exploratory approaches for demand estimation, in which firms experiment with different service offerings in order to simultaneously learn the demand while doing business. Such exploration and learning process can be costly without supervision. As reported by Rothschild (Journal of Economic Theory, 9 185‐202, 1974), traditional Bayesian dynamic control approaches may conclude with suboptimal offerings. We propose a novel demand learning approach that is guaranteed to converge to the optimal offering. The approach combines simulated annealing algorithm with Bayesian learning. We further present intelligent techniques that adaptively reduce the effort of exploration on suboptimal service offerings so as to improve the long‐run average profit.
- Open Access
Distributed source coding in dense sensor networks
We study the problem of the reconstruction of a Gaussian field defined in [0,1] using N sensors deployed at regular intervals. The goal is to quantify the total data rate required for the reconstruction of the field with a given mean square distortion. We consider a class of two-stage mechanisms which a) send information to allow the reconstruction of the sensor's samples within sufficient accuracy, and then b) use these reconstructions to estimate the entire field. To implement the first stage, the heavy correlation between the sensor samples suggests the use of distributed coding schemes to reduce the total rate. We demonstrate the existence of a distributed block coding scheme that achieves, for a given fidelity criterion for the reconstruction of the field, a total information rate that is bounded by a constant, independent of the number $N$ of sensors. The constant in general depends on the autocorrelation function of the field and the desired distortion criterion for the sensor samples. We then describe a scheme which can be implemented using only scalar quantizers at the sensors, without any use of distributed source coding, and which also achieves a total information rate that is a constant, independent of the number of sensors. While this scheme operates at a rate that is greater than the rate achievable through distributed coding and entails greater delay in reconstruction, its simplicity makes it attractive for implementation in sensor networks.Open AccessDistributed source coding in dense sensor networks
We study the problem of the reconstruction of a Gaussian field defined in [0,1] using N sensors deployed at regular intervals. The goal is to quantify the total data rate required for the reconstruction of the field with a given mean square distortion. We consider a class of two-stage mechanisms which a) send information to allow the reconstruction of the sensor's samples within sufficient accuracy, and then b) use these reconstructions to estimate the entire field. To implement the first stage, the heavy correlation between the sensor samples suggests the use of distributed coding schemes to reduce the total rate. We demonstrate the existence of a distributed block coding scheme that achieves, for a given fidelity criterion for the reconstruction of the field, a total information rate that is bounded by a constant, independent of the number $N$ of sensors. The constant in general depends on the autocorrelation function of the field and the desired distortion criterion for the sensor samples. We then describe a scheme which can be implemented using only scalar quantizers at the sensors, without any use of distributed source coding, and which also achieves a total information rate that is a constant, independent of the number of sensors. While this scheme operates at a rate that is greater than the rate achievable through distributed coding and entails greater delay in reconstruction, its simplicity makes it attractive for implementation in sensor networks. - 1 January 2005
- Published by Elsevier in Performance Evaluation
- Vol. 61 (2-3) , 181-201
Web traffic modeling at finer time scales and performance implications
- 1 January 2005
- Published by Elsevier in Performance Evaluation
- Vol. 61 (2-3) , 181-201
Web traffic modeling at finer time scales and performance implications
- 1 November 2004
- Published by Springer Nature in Queueing Systems
- Vol. 48 (3-4) , 263-307
Dynamic Scheduling of a Multiclass Fluid Model with Transient Overload
- 1 November 2004
- Published by Springer Nature in Queueing Systems
- Vol. 48 (3-4) , 263-307
Dynamic Scheduling of a Multiclass Fluid Model with Transient Overload
- 22 January 2004
- Published by Cambridge University Press (CUP) in Probability in the Engineering and Informational Sciences
- Vol. 18 (1) , 87-101
LOWER BOUNDS FOR LRD/GI/1 QUEUES WITH SUBEXPONENTIAL SERVICE TIMES
We investigate the tail distribution of the virtual waiting times in a LRD/GI/1 queue where the arrival process is long-range dependent (LRD) and the service times are independent and identically distributed (i.i.d.) random variables. We present two lower bounds on the stationary waiting time tail asymptotics, which illustrate the different dominating components that influence server performance under various conditions. In particular, we show that the tail distribution of the stationary waiting time is bounded below by that of the associated LRD/D/1 queues resulting from replacing all random service times by the mean. This shows the performance impact purely due to the long-range dependency of the arrival process. On the other hand, when the service times are subexponential, we show that the tail distribution of the stationary waiting time is bounded below by that of the corresponding D/GI/1 queue by replacing the dependent arrival process with its associated independent version. This shows the minimum performance impact due to the tail distribution of the service times. The above two lower bounds indicate that the performance of LRD/GI/1 queues will be dominated by the heavier tail of the corresponding LRD/D/1 and D/GI/1 queues. These features are further illustrated and quantified through examples and via numerous simulation experiments.
- 22 January 2004
- Published by Cambridge University Press (CUP) in Probability in the Engineering and Informational Sciences
- Vol. 18 (1) , 87-101
LOWER BOUNDS FOR LRD/GI/1 QUEUES WITH SUBEXPONENTIAL SERVICE TIMES
We investigate the tail distribution of the virtual waiting times in a LRD/GI/1 queue where the arrival process is long-range dependent (LRD) and the service times are independent and identically distributed (i.i.d.) random variables. We present two lower bounds on the stationary waiting time tail asymptotics, which illustrate the different dominating components that influence server performance under various conditions. In particular, we show that the tail distribution of the stationary waiting time is bounded below by that of the associated LRD/D/1 queues resulting from replacing all random service times by the mean. This shows the performance impact purely due to the long-range dependency of the arrival process. On the other hand, when the service times are subexponential, we show that the tail distribution of the stationary waiting time is bounded below by that of the corresponding D/GI/1 queue by replacing the dependent arrival process with its associated independent version. This shows the minimum performance impact due to the tail distribution of the service times. The above two lower bounds indicate that the performance of LRD/GI/1 queues will be dominated by the heavier tail of the corresponding LRD/D/1 and D/GI/1 queues. These features are further illustrated and quantified through examples and via numerous simulation experiments.
- 22 May 2002
- Published by Cambridge University Press (CUP) in Probability in the Engineering and Informational Sciences
- Vol. 16 (3) , 289-307
OPTIMAL CONTROL OF PARALLEL QUEUES WITH BATCH SERVICE
We consider the problem of dynamic allocation of a single server with batch processing capability to a set of parallel queues. Jobs from different classes cannot be processed together in the same batch. The arrival processes are mutually independent Poisson flows with equal rates. Batches have independent and identically distributed exponentially distributed service times, independent of the batch size and the arrival processes. It is shown that for the case of infinite buffers, allocating the server to the longest queue, stochastically maximizes the aggregate throughput of the system. For the case of equal-size finite buffers the same policy stochastically minimizes the loss of jobs due to buffer overflows. Finally, for the case of unequal-size buffers, a threshold-type policy is identified through an extensive simulation study and shown to consistently outperform other conventional policies. The good performance of the proposed threshold policy is confirmed in the heavy-traffic regime using a fluid model.
- 22 May 2002
- Published by Cambridge University Press (CUP) in Probability in the Engineering and Informational Sciences
- Vol. 16 (3) , 289-307
OPTIMAL CONTROL OF PARALLEL QUEUES WITH BATCH SERVICE
We consider the problem of dynamic allocation of a single server with batch processing capability to a set of parallel queues. Jobs from different classes cannot be processed together in the same batch. The arrival processes are mutually independent Poisson flows with equal rates. Batches have independent and identically distributed exponentially distributed service times, independent of the batch size and the arrival processes. It is shown that for the case of infinite buffers, allocating the server to the longest queue, stochastically maximizes the aggregate throughput of the system. For the case of equal-size finite buffers the same policy stochastically minimizes the loss of jobs due to buffer overflows. Finally, for the case of unequal-size buffers, a threshold-type policy is identified through an extensive simulation study and shown to consistently outperform other conventional policies. The good performance of the proposed threshold policy is confirmed in the heavy-traffic regime using a fluid model.
- 17 August 2001
- Published by Elsevier in Performance Evaluation
- Vol. 46 (2-3) , 219-233
Dynamic on-line task scheduling on parallel processors
- 17 August 2001
- Published by Elsevier in Performance Evaluation
- Vol. 46 (2-3) , 219-233
Dynamic on-line task scheduling on parallel processors
- 2 February 2001
- Published by SPIE-Intl Soc Optical Eng
- Vol. 4211, 113-122
Optimal scheduling in queueing network models of high-volume commercial web sites
The optimal control of various performance-based measures in high-volume commercial web sites requires a fundamental understanding of the interactions between the diverse set of Internet services that support customer needs and the different importance levels of these services to both the customer and the e-commerce merchant. We present here a study of the control policy for each server in a multiclass queueing network that maximizes a particular function of profit, or minimizes a particular function of cost, across the different classes of Internet services.
- 2 February 2001
- Published by SPIE-Intl Soc Optical Eng
- Vol. 4211, 113-122
Optimal scheduling in queueing network models of high-volume commercial web sites
The optimal control of various performance-based measures in high-volume commercial web sites requires a fundamental understanding of the interactions between the diverse set of Internet services that support customer needs and the different importance levels of these services to both the customer and the e-commerce merchant. We present here a study of the control policy for each server in a multiclass queueing network that maximizes a particular function of profit, or minimizes a particular function of cost, across the different classes of Internet services.