I have been doing research in the following application areas of Operations Management and Management Science: (i) Scheduling; (ii) Logistics and Transportation; (iii) Supply Chain Operations; (iv) Pricing. These areas cover a wide variety of operational, tactical and strategic managerial decision making problems faced by most manufacturing and service enterprises. Optimal or at least near-optimal solutions to such problems are of paramount importance because often times 1% of improvement over an existing solution to such problems means millions of dollars of cost savings to companies. My research mainly develops mathematical models and solution algorithms for finding optimal or near optimal solutions to these problems using optimization techniques such as dynamic programming, integer programming, and stochastic programming. I also design fast heuristics for NP-hard problems and analyze performance of heuristics in terms of worst-case and/or asymptotic behavior. Applications of my work to date in each of these areas are summarized in the following.
Scheduling problems are commonly encountered in manufacturing and service industries. They are concerned about efficient allocation of resources (machines, manpower, utility) to tasks (jobs, customer orders) to achieve a desired performance of the tasks subject to resource availability. Optimal or near optimal solutions to these problems often play a critical role in achieving low cost and high resource utilization. One class of problems that I have worked on are production scheduling problems that arise in just-in-time manufacturing. In a JIT environment, both earliness and tardiness are not encouraged, and hence it is desired that a job be completed processing at a time as close as possible to its due date. A typical scheduling problem in such an environment is to schedule a set of jobs with a minimum total (weighted) earliness and (weighted) tardiness.
(ii) Logistics and Transportation
I have worked on a number of real-world logistics and transportation problems. One of the problems involves optimal dispatching of trucks commonly encountered by logistics service providers in practice. Issues such as DOT rules, multiple time windows, crossdocking, and dynamic order arrivals are addressed in conjunction with classical vehicle routing decisions. Large-scale optimization techniques (column generation coupled with fast heuristics) are developed. Other practical problems that I have worked on involve food distribution, container vessel operations, and truck loading operations.
(iii) Supply Chain Operation
I have worked on a class of so-called supply chain scheduling problems which address scheduling issues that cross multiple stages or multiple functional areas of a supply chain. One set of the problems that I studied involve integrating production and distribution operations at the tactical and scheduling levels. Such problems often occur in supply chains for time-sensitive, perishable, or make-to-order products. Examples include PC manufacturing and distribution, food preparation and delivery, newspaper printing and delivery, and concrete paste mixing and delivery, etc. In such applications, production operations are closely linked with outbound distribution operations without much finished product inventory in the middle. Hence it is critical to consider production scheduling and delivery routing and scheduling jointly. I have developed a number of models and solution algorithms.
I have been working with companies for several years on some practical markdown pricing problems involving multiple retail stores. Large retail chains frequently use markdown schemes to sell their products and it is important to develop a good markdown scheme by fully utilizing available data and analytical tools. In the markdown pricing literature, almost all existing papers consider problems with a single store. In the problem we studied, there are many stores (50 to 100) and the pricing decisions for different stores are coupled by a number of business rules. We developed an optimization based approach to generate a near-optimal inventory allocation across the stores and a markdown scheme for each store. Our solution outperforms all commonly used techniques in practice.
Computer and communications technologies have transformed businesses ranging from recorded music to securities markets. How did managers in the firms that have been dramatically impacted by technology transformations miss what was happening? From the stories of Merrill Lynch's first response to new entrants in the brokerage business offering online trade execution, it is clear that there was a wide-spread lack of awareness about transformational IT at the very highest levels of the organization. One of the many CEOS of Kodak in the 1990s, George Fisher, described in an interview how he was unable to change middle management to a digital mindset, instead of a focus on chemistry and film. Executives in firms that produce and distribute video content are struggling with the threats and opportunities of Internet delivery, just as music industry managers have been searching for more than five years for a business model that can co-exist with peer-to-peer file sharing technology.
My research is focused on IT-enabled transformations and on how we can meet their challenges and opportunities. We have looked at organizations as diverse as the New York Stock Exchange and Kodak, and are currently studying the impact of the Web on the newspaper industry and on where we learn about news. In all of these cases, managers have been unprepared for dramatic changes in their organizations.
Figure 1 describes the process by through which technology potentially transforms an organization.
The first task for managers is to recognize that an enabling technology will have an impact on her business. The manager can choose to adopt the technology voluntarily, or it can be forced on her by competitors. In either case, the end result requires massive changes in the organization and in the cognitions of managers in that organization. Our studies of the NYSE and Kodak found that these changes were very difficult to make, and it ended up taking traumatic shocks to bring about change. Figure 1 suggests that there is a high probability of failure in adopting transformational technologies because of these organizational and managerial challenges.
The bottom line is that managers have to recognize a transformational technology and then they have to manage the adoption process, a process that is likely to require major organizational and managerial change. My book Inside the Future: Surviving the Technology Revolution, which accompanies a PBS documentary co-produced by the Smith School and Maryland Public Television (www.transfromationage.org) offers advice for managers and individuals on how to cope with the significant changes technology enables in government, industry, organizations and our daily lives.
Mining the Annotated Biomedical Web
Funded by the NSF
The biomedical research enterprise has created a rich, publicly accessible Web of hyperlinked and curated data. In parallel, the healthcare enterprise (hospital systems, physician offices, insurers), the NIH, and individuals, are creating personal health information (PHI), and specialized portals (dbGaP, eMERGE) are emerging, to provide restricted access to de-identified data. In order to improve interoperability, these communities have created a number of ontologies such as GO, MeSH, SNOMED-CT and UMLS. Data entries (records) in these resources are typically annotated with concepts or controlled vocabulary (CV) terms from one or more of these ontologies. The data entries are often hyperlinked to entries in other repositories, creating a richly curated Web of semantic knowledge comprising this annotated and hyperlinked data. We are developing a set of tools exploiting techniques from data management, information retrieval, optimization approaches and approximation algorithms, data mining and visualization to help the scientist better understand and explore this wealth of knowledge. Results of this research may lead scientists to formulate interesting hypotheses relating genes, diseases, individuals and their response to treatments. This can lead to personalized treatments and can empower an individual to contribute personal knowledge.
GeoNets: A Semantic Dataspace for Humanitarian Assistance
Funded by the NSF
Access to up-to-date and quality information can have a significant impact on the humanitarian relief community as they coordinate relief efforts. In addition to data that is created and curated by experts, there is a vast volunteer community who are empowered by the social Web to blog and generate community curated content. Our research will explore the following challenges in setting up the GeoNets semantic dataspace for humanitarian assistance: GeoNets Semantic Dataspace: GoNets will leverage methodologies for event detection, document clustering, query answering, ranking and personalization to create a GeoNets semantic dataspace. A front end intuitive user language will be defined for users to specify their profiles and express their queries. A combination of techniques from query answering, ranking and optimization will be developed to provide relevant and important answers efficiently. GeoNets Quality Assessment:
We will develop a methodology to involve users in evaluating the quality of results to their queries. Quality criteria may include timeliness, accuracy, popularity and relevance. The feedback may be explicitly obtained from users or it may be implicit, e.g., popular content. User profiles and ranking will also be used to assess and improve the quality of the retrieved answers.
Monitoring, Sensing and Effective Retrieval from the Social Web
The social Web as captured by blogs and tweets represents a digital slice of thoughts and actions of Netizens. While the preponderance of this data is only of interest to the creator and a small social network, the social Web has the potential to track the emergence of information about disasters and diseases, to follow social trends or commodity price fluctuations, to serve as a vast database for validation of queries, etc. Research challenges include methods to answer the following questions: When did topic A emerge? Who is most likely to blog about topic A or who is most likely to follow topic A? Has the conversation about topic A reached a critical mass and then did that occur? The computational challenges include document similarity and clustering, maintaining social networks and blogger profiles, personalized ranking, and optimal monitoring strategies.