Export
Anastasia Paparrizou
· 3 previous affiliations
- Researcher · Since 2023
- Researcher · 2016–2023
- Postdoc · 2014–2016
- Current
- University of Western Macedonia2009–2013
- PhD student · 2009–2013
Research Overview
Anastasia Paparrizou focuses on the intersection of constraint programming, data mining, and machine learning. Her collaborative work explores how data-driven insights can be systematically integrated into optimization problems to dynamically revise constraints and criteria. Her most recent contributions center on inductive constraint programming frameworks that bridge predictive modeling and complex scheduling or resource allocation tasks.
Research Trajectory
Between 2015 and 2017, Paparrizou participated in a large collaborative effort aimed at bridging the historical gap between machine learning and constraint programming. Traditionally, constraint programming has been utilized for complex real-world optimization problems, such as planning, scheduling, and resource allocation. This domain connects to the research area of The Limits of Local Consistency in Constraint Satisfaction and the broader theme of Balancing Consistency and Search in Constraint Satisfaction. However, standard constraint programming software historically struggled to dynamically exploit the vast amounts of data generated by these problems to update schedules and plans in real time.
To address this limitation, Paparrizou was a contributing middle author (among ten researchers) on a project proposing a novel framework known as the Inductive Constraint Programming (ICON) loop. The initial preprint of this work, The Inductive Constraint Programming Loop, was released in 2015. The framework systematically gathers and analyzes data to dynamically revise and adapt constraints and optimization criteria. This approach aligns closely with the research direction of Aligning Predictive Models with Downstream Optimization and the related area of Decision-Focused Learning, where statistical predictions are treated not as final outputs but as direct inputs to downstream optimization solvers.
This collaborative framework was subsequently formalized and disseminated across multiple venues to reach different sub-communities within artificial intelligence. In 2016, the team published a peer-reviewed iteration, The Inductive Constraint Programming Loop, focusing on the intersection of data mining and constraint programming. The following year, the framework was presented to a broader audience in IEEE Intelligent Systems as The Inductive Constraint Programming Loop. Throughout this period, Paparrizou's role as a middle author reflects her participation in a foundational effort to unify predictive data mining with rigorous constraint solving, moving away from isolated machine learning paradigms toward integrated optimization systems.
Current Directions
The most recent papers in the evidence are from 2017 and focus on the continued integration of machine learning techniques into traditional optimization pipelines. This work highlights ongoing challenges in the research area of Optimization Under Performative Distribution Shifts, where the deployment of a model or a schedule inherently changes the underlying data distribution. By participating in the development of dynamic, data-responsive constraint solving frameworks, Paparrizou's documented contributions emphasize the shift away from static optimization toward systems that continuously adapt to new information and evolving constraints.
Profile updated
| Title | Venue | Year |
|---|---|---|
| The Inductive Constraint Programming Loop | IEEE Intell. Syst | 2017 |
Loading new papers…
Loading research neighbors…
See something wrong? Report a mistake