Carlos Linares López

Carlos Linares López

Universidad Carlos III de Madrid
5 previous affiliations
    • Associate Professor · Since 2016
    • Director del Grado en Ingeniería Informática · 2021–2025
    • Profesor Titular Interino · 2004–2016
  • Deimos (Spain)2002–2004
    • Project Manager · 2002–2004
    • External Consultant · 2001–2002
    • PhD in Computer Science and Artificial Intelligence · 1997–2001
    • MsC in Knowledge Engineering · 1995–1997
  • INtelligent ADvisors (INAD)1998–2000
    • consultant · 1998–2000
3Machine learning papers tracked by Lacuna
Is this you?

Research Overview

Carlos Linares López focuses on automated planning, heuristic search, and the optimization of sequential portfolios. His work explores methods for deriving admissible heuristics in cost-optimal planning, including symbolic merge-and-shrink techniques and abstraction heuristics for bidirectional search. More recently, his research addresses the configuration and ordering of component solvers to maximize the performance of automated planning portfolios over time.

Research Trajectory

The available evidence captures a research trajectory from 2013 to 2016 centered on automated planning and heuristic search algorithms. A primary focus of this work is the development and refinement of admissible heuristics for cost-optimal planning, alongside the optimization of solver architectures.

In 2013, Linares López contributed to Symbolic Merge-and-Shrink for Cost-Optimal Planning as a second author. This research investigated the combination of two established approaches for deriving admissible heuristics: Symbolic Pattern Databases (PDBs) and Merge-and-Shrink (M&S) abstractions. The resulting method, Symbolic Merge-and-Shrink, utilized M&S abstractions as a relaxation criterion for symbolic backward search. The empirical evaluation demonstrated that this combined approach effectively leveraged the strengths of both underlying techniques, yielding heuristics that matched or exceeded the performance of the individual methods across various planning domains.

Expanding from heuristic design to the broader architecture of automated planning systems, Linares López served as the senior author on Sorting Sequential Portfolios in Automated Planning in 2015. While portfolios of problem solvers had previously shown advantages over single-algorithm approaches in tasks like Boolean satisfiability (SAT) and automated planning, the specific impact of solver ordering on temporal performance remained underexplored. This work proposed a greedy approach to sorting component solvers within a sequential portfolio, aiming to maximize the probability of achieving peak performance at any given point in time. The study showed that this ordering strategy efficiently achieved near-optimal performance over time and generalized effectively across problem sets.

Returning to core search algorithms in 2016, Linares López co-authored Abstraction Heuristics for Symbolic Bidirectional Search. Because the integration of heuristic functions into bidirectional search does not universally improve computational performance, this research proposed a conditional application of abstraction heuristics, deploying them primarily when standard symbolic bidirectional uniform-cost search becomes unfeasible. The study adapted partial and perimeter abstractions for bidirectional contexts, utilizing A* search to traverse abstract state spaces and generate perimeters. The results demonstrated that this targeted use of abstraction heuristics could further improve the efficiency of cost-optimal planning systems.

Current Directions

The most recent papers in the evidence are from 2015 and 2016, indicating a dual focus on optimizing search algorithms and improving solver portfolio configurations. Recent work investigates the conditional use of abstraction heuristics to salvage unfeasible symbolic bidirectional searches, as well as the statistical ordering of sequential portfolios to maximize solver performance over time. These directions highlight an ongoing interest in both the low-level mechanics of cost-optimal planning—specifically through advanced heuristic design—and the high-level orchestration of automated planning systems.

Profile updated

See something wrong? Report a mistake