Objective

The objective of the IEEE CIS Task Force on Multitask Learning and Multitask Optimization is to advance the development of novel algorithms and the theoretical understanding of methods that address multiple learning or optimization tasks jointly. It particularly encourages adaptive techniques that automatically identify and exploit synergies among tasks, allowing knowledge gained from one task to improve performance on others. The task force also provides a forum for academic and industrial researchers to exchange ideas and explore new research directions toward more efficient and intelligent decision-making systems.

In 2026, the Task Force on Multitask Learning and Multitask Optimization organized three special issues—“Advances in Metaheuristic Optimization Algorithms,” “Recent Advances in Mathematical Optimization and Related Topics,” and “Symmetry and Metaheuristic Algorithms”—and held the webinar “Metaheuristic Optimization” and the workshop “Metaheuristic Operators.” Planned activities for 2026–2027 include serving on the program committee of the IEEE Latin American Conference on Computational Intelligence, organizing the “Multitask Optimization” workshop, and submitting a proposal for the special issue “New Trends in Metaheuristic Methods.”

Activities

Alma Rodriguez

VICE CHAIR

Erik Cuevas

CHAIR

Alberto Luque

VICE CHAIR

Objective

To enable participants to understand, design, implement, and evaluate the mechanisms that drive metaheuristic optimization algorithms. Through practical exercises, participants will examine how solution generation, information sharing, adaptation, and selection influence search behavior, and learn to combine these mechanisms to address premature convergence, diversity loss, and inefficient use of computational resources.

Topics 

  1. Fundamentals of metaheuristic mechanisms
    Distinguish complete algorithms from their underlying mechanisms. Examine how representations, search operators, selection rules, and information sources interact during optimization.

  2. Solution representation and initialization
    Explore continuous, discrete, and combinatorial representations. Compare random sampling, Latin hypercube sampling, opposition-based initialization, and problem-informed strategies for generating initial solutions.

  3. Exploration mechanisms
    Study methods for discovering new search regions, including large perturbations, random walks, differential movements, and heavy-tailed sampling. Discuss how step size and sampling distributions affect search coverage.

  4. Exploitation and local refinement
    Examine mechanisms that improve promising solutions through neighborhood search, elite guidance, adaptive step reduction, and local optimization. Analyze their contribution to convergence and their risk of restricting the search too early.

  5. Selection, replacement, and elitism
    Compare tournament, rank-based, and fitness-based selection, together with survivor replacement strategies. Investigate how selection pressure and elite preservation influence solution quality and population diversity.

  6. Information sharing and population structures
    Analyze interactions based on leaders, centroids, neighborhoods, and multiple subpopulations. Explore how communication structures determine the spread of useful information and the emergence of collective search behavior.

  7. Diversity preservation and recovery
    Introduce distance measures, crowding, niching, archives, restarts, and population regeneration. Discuss when diversity supports progress and how to detect stagnation before applying corrective mechanisms.

  8. Constraint-handling mechanisms
    Examine penalty functions, feasibility-based selection, repair operators, and constraint-preserving representations. Discuss how these choices affect the search for feasible, high-quality solutions.

  9. Adaptive control and operator selection
    Explore parameter schedules, feedback-based adaptation, self-adaptation, and success-based operator selection. Use search indicators to adjust operator probabilities, perturbation strength, and computational effort.

  10. Hybridization and knowledge transfer
    Combine complementary mechanisms within a single framework. Examine global–local search integration, cooperation among populations, and transfer between related optimization tasks, including the possibility of harmful transfer.

  11. Experimental evaluation and ablation studies
    Design experiments that isolate each mechanism’s contribution. Compare variants under matched evaluation budgets, use repeated runs and statistical analysis, and assess solution quality, convergence, robustness, and computational cost.

  12. Practical mechanism-design exercise
    Implement a baseline optimizer, introduce selected mechanisms, and evaluate their effects on representative problems. Participants will justify their design choices and identify which mechanisms improve performance under specific conditions.

Upcoming events

Workshop: Metaheuristic Mechanisms (April 5/27)

Themes and contents

1. Foundations of multitask learning and multitask optimization

Definitions and motivations; differences among multitask learning, transfer learning, multiobjective optimization, and multitask optimization; task relationships; shared knowledge; positive and negative transfer; representative applications.

2. Fundamentals of large language models

Transformers and language modeling; pretraining, instruction tuning, and inference; prompting and contextual learning; representations and embeddings; capabilities and limitations of LLMs in learning, reasoning, and optimization.

3. Multitask learning with LLMs

Joint learning across language tasks; shared parameters and task-specific components; task selection and data balancing; instruction-based multitask training; parameter-efficient fine-tuning; task interference and evaluation of generalization.

4. Multitask optimization algorithms

Formulation of multiple optimization tasks; evolutionary multitasking; unified and task-specific representations; knowledge transfer mechanisms; adaptive transfer; task similarity estimation; prevention of negative transfer.

5. LLMs as assistants for multitask optimization

Task description and decomposition; identification of potential task relationships; generation of initial solutions, heuristics, and search operators; assistance with solution transfer; feedback from objective evaluations; feasibility checking and computational cost control.

6. Optimization of LLM-based multitask systems

Selection of prompts, examples, and task configurations; hyperparameter and adapter optimization; allocation of computational resources across tasks; use of metaheuristics; distinction between improving individual tasks and balancing competing system objectives.

7. Experimental design and performance evaluationIndependent and joint-task baselines; benchmark selection; task-level and overall performance; transfer gains and losses; ablation studies; repeated runs and statistical analysis; accounting for tokens, latency, and objective evaluations; reproducibility and data leakage prevention

8. Applications and integrative project

Applications in engineering design, scheduling, scientific text analysis, and code generation; implementation of an LLM-assisted multitask framework; comparison against independent task solutions; analysis of solution quality, transfer effectiveness, and computational cost.

Course: Multitask Learning and Multitask Optimization with LLMS (February 8/27)

Objective

To equip participants with the theoretical foundations and practical skills needed to design, implement, and evaluate methods that combine multitask learning, multitask optimization, and large language models. Participants will learn how to share knowledge across related tasks, identify beneficial and harmful transfer, and use LLMs to support the simultaneous solution of multiple learning and optimization problems, with particular attention to computational efficiency, reproducibility, and real-world applications.

Investigación

Dr. Erik Cuevas.

CONTACTO

erik.cuevas@cucei.udg.mx

Tel : +52 33 1378 5900 Ext 27714

© 2025. All rights reserved.