Christian Perez
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Christian Perez has a total of 18 co-authors. Here are the top co-authors by number of shared publications:
- Miguel A. Salido
Universitat Politècnica de València
8 shared publications - Carlos March
Delft University of Technology
5 shared publications - Laura Climent
Instituto de Automática e Informática Industrial, Universitat Politècnica de València, Valencia, Spain
3 shared publications - Alejandro Arbelaez
Instituto de Automática e Informática Industrial, Universitat Politècnica de València, Valencia, Spain
2 shared publications Miguel A. Salido
2 shared publications- Adriana Giret
Universitat Politècnica de València
1 shared publication
Latest Publications
- Open Access
- 3 October 2026
- Published by Elsevier in SSRN Electronic Journal
Scheduling the Solver: Learning-Guided Structure-Aware Solver Control for Job Shop Scheduling
Modern constraint-programming solvers expose internal search mechanisms whose value varies across instances, yet these mechanisms are typically fixed globally. We propose $\ASA$, a learning-guided structure-aware control method for the Job Shop Scheduling Problem (JSSP). Before search, eleven static descriptors summarize instance size, workload imbalance, lower-bound structure, and route geometry. Two frozen Random-Forest regressors predict the relative optimization advantage of alternative CP-SAT search regimes, and a conservative rule selects the regime used by a common exact backend. No solver outcome or online state is required at deployment. Policy development uses matched deterministic search effort to obtain stable learning targets; external evaluation remains strictly wall-clock based. The retained action portfolio exhibits 10.26% virtual-best versus single-best headroom, of which the learned policy captures 67.1% under strict out-of-fold evaluation. Held-out permutation analysis identifies processing-time variability as the dominant signal for the effective C_0--C_1 switch. After freezing the representation, models, and decision rule, the policy retains a 5.61\% improvement over the strongest fixed action on 54 previously unused families. On 103 protected JSPLIB instances, the same policy improves standard single-worker CP-SAT at 10, 60, and 600 s. At 600 s, mean relative percentage deviation falls from 9.87% to 6.19%, a 37.3% relative reduction. A secondary analysis reveals a long-horizon crossover for some decisions, showing that structural solver control is useful but not budget invariant. These results demonstrate that pre-search JSSP structure provides actionable information for learned control of an exact optimization backend.- 3 October 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessScheduling the Solver: Learning-Guided Structure-Aware Solver Control for Job Shop Scheduling
Modern constraint-programming solvers expose internal search mechanisms whose value varies across instances, yet these mechanisms are typically fixed globally. We propose $\ASA$, a learning-guided structure-aware control method for the Job Shop Scheduling Problem (JSSP). Before search, eleven static descriptors summarize instance size, workload imbalance, lower-bound structure, and route geometry. Two frozen Random-Forest regressors predict the relative optimization advantage of alternative CP-SAT search regimes, and a conservative rule selects the regime used by a common exact backend. No solver outcome or online state is required at deployment. Policy development uses matched deterministic search effort to obtain stable learning targets; external evaluation remains strictly wall-clock based. The retained action portfolio exhibits 10.26% virtual-best versus single-best headroom, of which the learned policy captures 67.1% under strict out-of-fold evaluation. Held-out permutation analysis identifies processing-time variability as the dominant signal for the effective C_0--C_1 switch. After freezing the representation, models, and decision rule, the policy retains a 5.61\% improvement over the strongest fixed action on 54 previously unused families. On 103 protected JSPLIB instances, the same policy improves standard single-worker CP-SAT at 10, 60, and 600 s. At 600 s, mean relative percentage deviation falls from 9.87% to 6.19%, a 37.3% relative reduction. A secondary analysis reveals a long-horizon crossover for some decisions, showing that structural solver control is useful but not budget invariant. These results demonstrate that pre-search JSSP structure provides actionable information for learned control of an exact optimization backend. - Open Access
- 30 August 2026
- Published by Elsevier in Computers & Operations Research
- Vol. 196
Dynamic idle-time bounding mechanism for job shop scheduling problems
- 30 August 2026
- Published by Elsevier in Computers & Operations Research
- Vol. 196
Open AccessDynamic idle-time bounding mechanism for job shop scheduling problems
- Open Access
- 8 July 2026
- Published by Elsevier in SSRN Electronic Journal
Predict for Pruning: Graph Neural Network-Based Makespan Prediction for Solver Pruning in Job Shop Scheduling
The Job Shop Scheduling Problem (JSP) remains challenging for exact optimization because feasible operation orderings grow rapidly with the number of jobs and machines. Constraint programming and mixed-integer programming solvers provide feasibility checks, incumbent management, and optimality certificates, but may spend substantial time exploring regions unlikely to contain competitive schedules. This paper introduces P4P (predict for pruning), a graph-based framework for safe solver guidance in JSP. Each instance is encoded as an edge-typed scheduling graph whose nodes represent operations and whose edges distinguish conjunctive precedence from disjunctive machine-conflict relations. A dual-backbone graph neural predictor combines SAGE-style aggregation with attention-based propagation to estimate makespan before search starts. The prediction is translated into a temporary upper-bound target with an explicit relaxation and fallback protocol, keeping learned guidance separate from solver certification. Experiments on IGJSP and JSPLIB show accurate pre-search estimates and solver-guidance effects whose speed-quality trade-off depends on calibration and instance regime. The proposed formulation reframes makespan prediction as an operational interface between graph learning and exact JSP solving.
- 8 July 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessPredict for Pruning: Graph Neural Network-Based Makespan Prediction for Solver Pruning in Job Shop Scheduling
The Job Shop Scheduling Problem (JSP) remains challenging for exact optimization because feasible operation orderings grow rapidly with the number of jobs and machines. Constraint programming and mixed-integer programming solvers provide feasibility checks, incumbent management, and optimality certificates, but may spend substantial time exploring regions unlikely to contain competitive schedules. This paper introduces P4P (predict for pruning), a graph-based framework for safe solver guidance in JSP. Each instance is encoded as an edge-typed scheduling graph whose nodes represent operations and whose edges distinguish conjunctive precedence from disjunctive machine-conflict relations. A dual-backbone graph neural predictor combines SAGE-style aggregation with attention-based propagation to estimate makespan before search starts. The prediction is translated into a temporary upper-bound target with an explicit relaxation and fallback protocol, keeping learned guidance separate from solver certification. Experiments on IGJSP and JSPLIB show accurate pre-search estimates and solver-guidance effects whose speed-quality trade-off depends on calibration and instance regime. The proposed formulation reframes makespan prediction as an operational interface between graph learning and exact JSP solving.
- 15 June 2026
- Published by Universitat Politecnica de Valencia
- p. 1-9
Badges as Visible Formative Feedback in Arduino PBL
Badges are common in gamified courses, yet their role as formative assessment in hands-on laboratories remains underexplored. We report an observational study from an Arduino project-based learning (PBL) course in which eight team badges were announced at the beginning of the next class together with the reason why they had been earned, turning recognition into recurring formative feedback. Using anonymized records from one cohort (N = 30 students; 15 fixed teams), we examined whether badge score, obtained by linearly rescaling earned badges (0–8) to a weighted 0–2 component, was associated with two graded sprints and the final practical grade. Spearman correlations showed positive associations with Sprint 1 (ρ = 0.69, p < .001), Sprint 2 (ρ = 0.50, p = .005), and final grade (ρ = 0.90, p < .001). High-badge students started near the ceiling, whereas the low-badge group improved more between sprints but remained behind. We discuss how a badge-announcement routine can support formative feedback, motivational gamification, and assessment transparency in Arduino PBL while remaining cautious about causal interpretation.
- 15 June 2026
- Published by Universitat Politecnica de Valencia
- p. 1-9
Badges as Visible Formative Feedback in Arduino PBL
Badges are common in gamified courses, yet their role as formative assessment in hands-on laboratories remains underexplored. We report an observational study from an Arduino project-based learning (PBL) course in which eight team badges were announced at the beginning of the next class together with the reason why they had been earned, turning recognition into recurring formative feedback. Using anonymized records from one cohort (N = 30 students; 15 fixed teams), we examined whether badge score, obtained by linearly rescaling earned badges (0–8) to a weighted 0–2 component, was associated with two graded sprints and the final practical grade. Spearman correlations showed positive associations with Sprint 1 (ρ = 0.69, p < .001), Sprint 2 (ρ = 0.50, p = .005), and final grade (ρ = 0.90, p < .001). High-badge students started near the ceiling, whereas the low-badge group improved more between sprints but remained behind. We discuss how a badge-announcement routine can support formative feedback, motivational gamification, and assessment transparency in Arduino PBL while remaining cautious about causal interpretation.
- 1 June 2026
- Published by IATED Academy in EDULEARN Proceedings
- Vol. 1, 1514
DESIGN AND EARLY EVALUATION OF A GENERATIVE AI TUTOR FOR SELF-REGULATED LEARNING IN AN INTRODUCTORY DATA STRUCTURES COURSE
- 1 June 2026
- Published by IATED Academy in EDULEARN Proceedings
- Vol. 1, 1514
DESIGN AND EARLY EVALUATION OF A GENERATIVE AI TUTOR FOR SELF-REGULATED LEARNING IN AN INTRODUCTORY DATA STRUCTURES COURSE
- 1 June 2026
- Published by IATED Academy in EDULEARN Proceedings
- Vol. 1, 1512
SPRINT-BASED PROJECT LEARNING AND GAMIFICATION IN AN INTRODUCTORY ARDUINO COURSE
- 1 June 2026
- Published by IATED Academy in EDULEARN Proceedings
- Vol. 1, 1512
SPRINT-BASED PROJECT LEARNING AND GAMIFICATION IN AN INTRODUCTORY ARDUINO COURSE
- 29 April 2026
- Published by Springer Nature
- p. 22-38
Cost-Efficient Charging Station Deployment for Electric Bus Fleets
- 29 April 2026
- Published by Springer Nature
- p. 22-38
Cost-Efficient Charging Station Deployment for Electric Bus Fleets
- 17 July 2025
- Published by Universitat Politecnica de Valencia
Análisis de Factores Influyentes en el Aprendizaje Colaborativo Mediante Perusall
Este estudio explora los factores influyentes en el aprendizaje colaborativo mediante un análisis realizado en la plataforma Perusall. Se examinó una muestra de 81 estudiantes, analizando métricas de interacción (anotaciones, comentarios, tiempo y porcentaje de lectura) y su relación con los resultados en evaluaciones teóricas y prácticas, considerando además variables demográficas como asignatura, género y edad. El uso de la estimación de densidad mediante kernel Beta permitió obtener una representación detallada de la distribución de las métricas, facilitando la identificación de patrones de compromiso. Los resultados revelan que niveles más altos de participación están fuertemente vinculados a mejores resultados académicos, destacándose especialmente en asignaturas de nivel avanzado y entre estudiantes de mayor edad. Estos hallazgos aportan valiosas evidencias para el desarrollo de herramientas digitales que potencien el aprendizaje colaborativo en la educación superior.
- 17 July 2025
- Published by Universitat Politecnica de Valencia
Análisis de Factores Influyentes en el Aprendizaje Colaborativo Mediante Perusall
Este estudio explora los factores influyentes en el aprendizaje colaborativo mediante un análisis realizado en la plataforma Perusall. Se examinó una muestra de 81 estudiantes, analizando métricas de interacción (anotaciones, comentarios, tiempo y porcentaje de lectura) y su relación con los resultados en evaluaciones teóricas y prácticas, considerando además variables demográficas como asignatura, género y edad. El uso de la estimación de densidad mediante kernel Beta permitió obtener una representación detallada de la distribución de las métricas, facilitando la identificación de patrones de compromiso. Los resultados revelan que niveles más altos de participación están fuertemente vinculados a mejores resultados académicos, destacándose especialmente en asignaturas de nivel avanzado y entre estudiantes de mayor edad. Estos hallazgos aportan valiosas evidencias para el desarrollo de herramientas digitales que potencien el aprendizaje colaborativo en la educación superior.
- 17 July 2025
- Published by Universitat Politecnica de Valencia
Tutor Virtual con Inteligencia Artificial: Evaluación Preliminar y Perspectivas para el Aprendizaje Autónomo
Este trabajo presenta el desarrollo e implementación de un tutor virtual basado en modelos GPT de OpenAI, aplicado en la asignatura de Estructura de Datos y Algoritmos del Grado en Ingeniería Informática. El objetivo del estudio es evaluar, mediante una intervención educativa a mitad de curso, si el uso de esta herramienta mejora la comprensión de conceptos clave y fomenta la autonomía en la resolución de problemas. La metodología combina el análisis de métricas de uso con un cuestionario estructurado en formato Likert, el cual evalúa diversas dimensiones de la experiencia de los estudiantes. Los resultados preliminares indican que los alumnos con menor conocimiento en inteligencia artificial valoran de forma significativamente positiva la herramienta y que un mayor tiempo de interacción se asocia con una percepción favorable de sus funcionalidades. Con base en estos hallazgos, se proponen mejoras en la personalización del contenido y una orientación inicial más detallada, con el fin de maximizar el impacto educativo del tutor. En conjunto, los resultados respaldan una futura implementación más amplia del tutor virtual en todos los grupos de la asignatura, adaptándose a las diversas necesidades del alumnado.
- 17 July 2025
- Published by Universitat Politecnica de Valencia
Tutor Virtual con Inteligencia Artificial: Evaluación Preliminar y Perspectivas para el Aprendizaje Autónomo
Este trabajo presenta el desarrollo e implementación de un tutor virtual basado en modelos GPT de OpenAI, aplicado en la asignatura de Estructura de Datos y Algoritmos del Grado en Ingeniería Informática. El objetivo del estudio es evaluar, mediante una intervención educativa a mitad de curso, si el uso de esta herramienta mejora la comprensión de conceptos clave y fomenta la autonomía en la resolución de problemas. La metodología combina el análisis de métricas de uso con un cuestionario estructurado en formato Likert, el cual evalúa diversas dimensiones de la experiencia de los estudiantes. Los resultados preliminares indican que los alumnos con menor conocimiento en inteligencia artificial valoran de forma significativamente positiva la herramienta y que un mayor tiempo de interacción se asocia con una percepción favorable de sus funcionalidades. Con base en estos hallazgos, se proponen mejoras en la personalización del contenido y una orientación inicial más detallada, con el fin de maximizar el impacto educativo del tutor. En conjunto, los resultados respaldan una futura implementación más amplia del tutor virtual en todos los grupos de la asignatura, adaptándose a las diversas necesidades del alumnado.
- Open Access
- 14 July 2025
- Published by Association for Computing Machinery (ACM)
- p. 159-162
IGJSP: A Reproducible Instance Generator for Job Shop Scheduling
- 14 July 2025
- Published by Association for Computing Machinery (ACM)
- p. 159-162
Open AccessIGJSP: A Reproducible Instance Generator for Job Shop Scheduling