Marek Petrik

Constraint relaxation in approximate linear programs

Approximate Linear Programming (ALP) is a reinforcement learning technique with nice theoretical properties, but it often performs poorly in practice. We identify some reasons for the poor quality of ALP solutions in problems where the approximation induces virtual loops. We then introduce two methods for improving solution quality. One method rolls out selected constraints of the ALP, guided by the dual information. The second method is a relaxation of the ALP, based on external penalty methods. The latter method is applicable in domains in which rolling out constraints is impractical. Both approaches show promising empirical results for simple benchmark problems as well as for a realistic blood inventory management problem.

A bilinear programming approach for multiagent planning

Multiagent planning and coordination problems are common and known to be computationally hard. We show that a wide range of two-agent problems can be formulated as bilinear programs. We present a successive approximation algorithm that significantly outperforms the coverage set algorithm, which is the state-of-the-art method for this class of multiagent problems. Because the algorithm is formulated for bilinear programs, it is more general and simpler to implement. The new algorithm can be terminated at any time and-unlike the coverage set algorithm-it facilitates the derivation of a useful online performance bound. It is also much more efficient, on average reducing the computation time of the optimal solution by about four orders of magnitude. Finally, we introduce an automatic dimensionality reduction method that improves the effectiveness of the algorithm, extending its applicability to new domains and providing a new way to analyze a subclass of bilinear programs.

Agile logistics simulation and optimization for managing disaster responses

Catastrophic events such as hurricanes, earthquakes or floods require emergency responders to rapidly distribute emergency relief supplies to protect the health and lives of victims. In this paper we develop a simulation and optimization framework for managing the logistics of distributing relief supplies in a multi-tier supply network. The simulation model captures optimized stocking of relief supplies, distribution operations at federal or state-operated staging facilities, demand uncertainty, and the dynamic progression of disaster response operations. We apply robust optimization techniques to develop optimized stocking policies and dispatch of relief supplies between staging facilities and points of distribution. The simulation framework accommodates a wide range of disaster scenarios and stressors, and helps assess the efficacy of response plans and policies for better disaster response.

Optimizing the end-to-end value chain through demand shaping and advanced customer analytics

Abstract As supply chains become increasingly outsourced, the end-to-end supply network is often spread across multiple enterprises. In addition, increasing focus on lean inventory can often create significant supply/demand imbalances over a multi-enterprise supply chain. This paper discusses a set of integrated analytics for supply/demand synchronization with a new emphasis on customer facing actions called demand shaping. Demand shaping is the ability to sense changing demand patterns, evaluate and optimize an enterprise supply plan to best support market demand and opportunity, and execute a number of demand shaping actions to “steer” demand to align with an optimized plan. First, we describe a multi-enterprise cloud-based data model called the Demand Signal Repository (DSR) that includes a tightly linked end-to-end product dependency structure as well as a trusted source of demand and supply levels across the extended supply chain. Secondly, we present a suite of mathematical optimization models that enable on demand up selling, alternative-selling and down-selling to better integrate the supply chain horizontally, connecting the interaction of customers, business partners and sales teams to procurement and manufacturing capabilities of a firm. And finally, we describe findings and managerial insights from real-life experiences with demand shaping in a server computer manufacturing environment.

Social media and customer behavior analytics for personalized customer engagements

Companies in various industries, including travel, hospitality, and retail, increasingly focus on improving customer relationships and customer loyalty. In this paper, we propose a new systems architecture that combines the textual content in social media messages with product information, such as the descriptions summarized in catalogs, in order to provide marketing campaign recommendations. Companies commonly build user profiles based on purchase histories and other customer-specific information; however, when dealing with social media, we often cannot match the social media users with the customers. In this regard, we address the problem of targeting individual social media messages for which no personalized profile information can be retrieved. Our solution combines two disparate computational toolboxes for text analytics—natural language processing and machine learning—in order to select social media users for whom to target with topic-specific advertisements. Natural language processing is used to analyze the context of social media messages, and machine learning is used to analyze product information, with the goal being to match social media messages to products and ranking potential advertisements. To demonstrate the framework, we detail a real-world application in the travel and tourism industry using Twitter® as the social media platform.