Hala Mostafa

Hala Mostafa

I am a Research Scientist at United Technologies Research Center (UTRC) in East Hartford, Connecticut. UTRC is the research arm of United Technologies Corporation, a conglomerate that includes businesses that make aircrafts, engines, escalators and HVAC systems. I apply decision theory and machine learning to solve a variety of real world  problems and interact with customers from very different backgrounds. In addition, I lead internally- and government-funded research projects in the areas of sequential decision-making, transfer learning with expert knowledge and heterogeneous domain adaptation. If you are a PhD student/graduate interested in these topics and looking for internship/permanent position, please contact me!

Before joining UTRC, I spent a year as a visiting Research Scientist at the Living Analytics Research Center, a new joint research initiative between Singapore Management University and Carnegie Mellon University. At LARC, we addressed a number of real-world problems together with our industrial partners. I used machine learning and mathematical optimization techniques to address issues like handling the congestion of visitors at venues like theme parks and world expos and studying offline and online customer transactions for more informed marketing.

Before joining LARC, I was a Research Scientist in the Distributed Systems Group at BBN Technologies, Cambridge, MA. I was the Technical Lead on a project where we used and enhanced distributed constraint optimization algorithms to address the issue of how a distributed system should utilize a set of (limited) resources to fulfill Quality of Service and security requirements. I was also involved in a number of proposal writing and business development activities.

I received my PhD from the University of Massachusetts, Amherst in 2011 with a thesis entitled "Exploiting Structure In Coordinating Multiple Decision Makers". I was in the Multi-Agent Systems Lab under the supervision of Prof Victor Lesser.

My thesis work was in the area of decentralized sequential decision-making under uncertainty. Within the framework of Dec-POMDPs, I studied the issue of how multiple decision makers share information to better carry out a task and how they maintain beliefs about each other in the absence of full communication.


My research interests include:

  • Decision making under uncertainty: Centralized and distributed sequential decision-making and planning. Coordination problems arising from shared tasks/resources in distributed systems.
  • Transfer learning: Capturing and leveraging external knolwedge on differences between source and target distributions. Approaches for dealing with heterogeneous data (different feature spaces).
  • Learning and decision-making over graphs: Studying the diffusion of entities and phenomena over networks. Making decisions to induce certain desirable diffusion properties.
  • Applied machine learning: Using graphical models for fault detection and prognostics and health management (PHM).

Research Projects at UTRC

  • Transfer Learning and domain adaptation
    In this project, we improved purely data-driven transfer learning by incorporating additional information provided by a subject matter expert (SME). In many settings, SMEs can have useful, though not necessarily precise, knowledge of how the source and target data distributions differ. We studied two transfer learning settings (covariate shift and functional change), identified types of SME knowledge that can be available in each, formulated optimization problems to incorporate this knowledge and demonstrated improved learning performance. .
  • Sequential decision making
    In this project, we investigate the use of (Paritally Observable) Markov Decision Processes in a variety of industrial settings like fleet management, prognostics and health management (PHM), and conducting maintenance activities. We are working on more user-friendly ways of capturing decision-making problems, as well as approaches to incorporate expert knowledge in our solutions.

Research Projects at BBN

  • Continuous runtime assessment and adaptaion of QoS and IA (2011 - 2014)
    This AFRL-funded project is concerned with the assessment and management of the levels of Quality of Service and Information Assurance delivered by a distributed system. QoS and IA compete for the same set of resources, often making it impossible to employ the full range of security mechanisms while maintaining the highest QoS levels. My role in this project concerns the tradeoffs that need to be made between QoS and IA to maximize user satisfaction given resource constraints. I formulated the problem as a Distributed Constraint Optimization Problem, proposed a value propagation algorithm to complement existing utility propagation algorithms, and deployed our assessment and adaptation techniques on a simulated distributed systems running on multiple virtual machines.
  • Communications in Extreme RF Spectrum Conditions (CommEx) (2012)
    The electromagnetic spectrum is used by friendly and adversary users, resulting in an increasingly congested, space, time and frequency environment. To communicate successfully amid these challenges, a communication system must exhibit significant adaptivity and flexibility. DARPA's CommEx program is interested in technologies and techniques that address communications in severe jamming and in the presence of adaptive jamming and interference sources. My role in this project concerns the sequential decision process of the communication system that needs to take into account uncertainty about both the adversary and the environment

Research Projects at UMass

  • Exploiting structured interactions among agents (2009 - 2011)
    In  real life situations where multiple decision-makers collaborate on a task, there is often a fair degree of independence among them; the agents/decision-makers interact in well-defined, structured manner. I exploited this structured interaction to realize representational and computational savings. For the former, I proposed a model that is much more compact than general decentralized MDPs and game trees. I realize computational savings through heuristics and solution algorithms that leverage the explicit representation of interactions in my model. I study both cooperative and self-interested agents.
  • Using mathematical optimization for sequential decision-making (Spring, Summer 2010)
    In this effort, I investigated what techniques and solvers from the field of mathematical optimization can be useful in solving decision- and game-theoretic models. I explored a number of formulations of general DEC-MDPs as mathematical programs . I also leveraged structured agent interactions to obtain more compact formulations. Other approaches I tried include homotopy methods for solving a formulation of my problem as a system of nonlinear equations, and decomposition methods for solving mathematical programs that become simple when fixing a set of complicating variables.
  • Distributed reasoning in a communication-limited environment (2007 - 2009)
    In this project, I studied communication in two settings. The first involved selfish agents that need to disclose some information in order to obtain a reward that depends on the quantity and quality of information disclosed by all agents. Each agent decides what it should disclose over the course of the game to maximize its reward and minimize its privacy loss. I formulated this problem as a sequential game of incomplete information and developed an approximate anytime hill-climbing algorithm for it. In the second setting, I studied communication among cooperative agents. I developed heuristics for introducing communication into a DEC-MDP in a controlled manner that limits the computational cost of reasoning about communication.
  • GILA: Generalized Integrated Learning Architecture (2008)
    GILA was one of the teams in the DARPA Integrated Learning program. We studied coordinating multiple learning agents with different learning algorithms whose performances depend on the context, and integrating their hypotheses to cooperatively and incrementally solve complex problems (e.g. air traffic flight planning). My role was to study the use of a feature-based context representation to choose which learners to use.
  • Coordination Decision Support Assistants (COORDINATORs) (2005-2006)
    The goal of this DARPA project was creating distributed intelligent software systems that help fielded units adapt their mission plans as the situation around them changes and impacts their plans. I was one of the leaders of the UMass effort which resulted in a new approach to coordinating the scheduling and execution of complex hierarchical task structures that are distributed among agents.



  • Program Committee Member

    • International Joint Conference On Artificial Intelligence (IJCAI) 2017

    • International Conference on Autonomous Agents and Multiagent Systems (AAMAS) 2013, 2014, 2015, 2017

    • Twenty-Sixth Conference on Artificial Intelligence (AAAI) 2012

    • Workshop on Multiagent Sequential Decision Making Workshop (MSDM) 2011-2015

    • Workshop on Optimization In Multiagent Systems (OptMAS) 2013, 2014, 2017

    • Workshop on Agents and CyberSecurity (ACySe) 2014

  • Member of the Doctoral Consortium Career Panel, AAMAS 2013.

  • Member of the CRA-W/CDC Discipline-specific Mentoring Panel, AAMAS 2013.

  • Member of the CS Women's Professionalism Panel at University of Massachusetts, 2012.

  • Reviewer for The Computer Journal

Curriculum Vitae