Puneet Sharma - Academics

Current Research

My current research involves developing scalable combinatorial optimization algorithms for solving static and dynamic resource allocation problems. Most of these problems are intimately related to a class of combinatorial resource allocation problems that have been studied extensively in various formulations such as the minimum distortion problem in data compression, facility location assignments, optimal quadrature rules and discretization of partial differential equations, pattern recognition, drug discovery, neural networks, and clustering analysis. In contrast, these problems are relatively recent in the control literature, having arisen in coarse quantization, coverage control, mobile sensing networks, and motion coordination algorithms.

Computationally, these problems are typically complex and time intensive, if not intractable. Complexity is further complicated by their inherent non-convex nature, and the curse of dimensionality. Additionally, the huge size of the underlying datasets, and distributed implementation calls for algorithms that are scalable.

We have used tools from information theory, graph theory, statistical physics and control theory to design such algorithms. We have implemented a Maximum Entropy based framework, and augmented it with additional constraints for addressing the issues of scalability and distributed implementation. We have successfully implemented these scalable algorithms for solving the library selection problems in combinatorial drug-discovery [1], by simultaneously addressing the key issues of diversity, representativeness, inclusion/exclusion and scalabilty. Similar framework is also used for solving the state-space discretization problem in quantized control.

Presently, we are working on the coverage and tracking problem with mobile sites and resources. We have developed a Maximum Entropy based algorithm for solving this dynamic locational optimization problem [2]. The emphasis is on developing algorithms that avoid local minima and are insensitive to the initial placement of resources. The algorithm ensures that the coverage metric (between the mobile sites and resources) is continuously improved.

Publications

Sharma, P., Salapaka, S., and Beck, C., A Scalable Approach to Combinatorial Library Design for Drug Discovery
J. Chem. Inf. Model., 48, 1, 27 - 41, 2008.
Sharma, P., Salapaka, S., and Beck, C., Entropy Based Algorithm for Combinatorial Optimization Problems with Mobile Sites and Resources
Proc. of American Control Conference, 2008.
Sharma, P., Salapaka, S., and Beck, C., A Maximum Entropy Based Scalable Algorithm for Resource Allocation Problems
Proc. of American Control Conference, 516-521, 2007.
Sharma, P., Salapaka, S., and Beck, C., A Scalable Deterministic Annealing Algorithm for Resource Allocation Problems
Proc. of American Control Conference, 3092-3097, 2006.
Sharma, P., Salapaka, S., and Beck, C., A Deterministic Annealing Approach to Combinatorial Library Design for Drug Discovery
Proc. of American Control Conference, 979- 984 vol. 2, 2005.
Sharma, P., and Beck, C., Modelling and distributed control of mobile offshore bases
Proc. of American Control Conference, 5238- 5243 vol.6, 2004.

M.S. Thesis

Modelling and Distributed Control of Spatial Array Systems - Advisor: Prof. Carolyn Beck.


	\| Home \| Academics \| Publications \| Misc. Links \| Pics
	Current Research My current research involves developing scalable combinatorial optimization algorithms for solving static and dynamic resource allocation problems. Most of these problems are intimately related to a class of combinatorial resource allocation problems that have been studied extensively in various formulations such as the minimum distortion problem in data compression, facility location assignments, optimal quadrature rules and discretization of partial differential equations, pattern recognition, drug discovery, neural networks, and clustering analysis. In contrast, these problems are relatively recent in the control literature, having arisen in coarse quantization, coverage control, mobile sensing networks, and motion coordination algorithms. Computationally, these problems are typically complex and time intensive, if not intractable. Complexity is further complicated by their inherent non-convex nature, and the curse of dimensionality. Additionally, the huge size of the underlying datasets, and distributed implementation calls for algorithms that are scalable. We have used tools from information theory, graph theory, statistical physics and control theory to design such algorithms. We have implemented a Maximum Entropy based framework, and augmented it with additional constraints for addressing the issues of scalability and distributed implementation. We have successfully implemented these scalable algorithms for solving the library selection problems in combinatorial drug-discovery [1], by simultaneously addressing the key issues of diversity, representativeness, inclusion/exclusion and scalabilty. Similar framework is also used for solving the state-space discretization problem in quantized control. Presently, we are working on the coverage and tracking problem with mobile sites and resources. We have developed a Maximum Entropy based algorithm for solving this dynamic locational optimization problem [2]. The emphasis is on developing algorithms that avoid local minima and are insensitive to the initial placement of resources. The algorithm ensures that the coverage metric (between the mobile sites and resources) is continuously improved. Publications Sharma, P., Salapaka, S., and Beck, C., A Scalable Approach to Combinatorial Library Design for Drug Discovery J. Chem. Inf. Model., 48, 1, 27 - 41, 2008. Sharma, P., Salapaka, S., and Beck, C., Entropy Based Algorithm for Combinatorial Optimization Problems with Mobile Sites and Resources Proc. of American Control Conference, 2008. Sharma, P., Salapaka, S., and Beck, C., A Maximum Entropy Based Scalable Algorithm for Resource Allocation Problems Proc. of American Control Conference, 516-521, 2007. Sharma, P., Salapaka, S., and Beck, C., A Scalable Deterministic Annealing Algorithm for Resource Allocation Problems Proc. of American Control Conference, 3092-3097, 2006. Sharma, P., Salapaka, S., and Beck, C., A Deterministic Annealing Approach to Combinatorial Library Design for Drug Discovery Proc. of American Control Conference, 979- 984 vol. 2, 2005. Sharma, P., and Beck, C., Modelling and distributed control of mobile offshore bases Proc. of American Control Conference, 5238- 5243 vol.6, 2004. M.S. Thesis Modelling and Distributed Control of Spatial Array Systems - Advisor: Prof. Carolyn Beck.