The Artificial Intelligence Research Institute (IIIA) offers up to 8 scholarship plans to introduce participants to a research career, as part of CSIC's JAE Intro ICU 2026 call.
These are the training plans offered and their mentors:
IIIA-01: MLOps for...
The Artificial Intelligence Research Institute (IIIA) offers up to 8 scholarship plans to introduce participants to a research career, as part of CSIC's JAE Intro ICU 2026 call.
These are the training plans offered and their mentors:
IIIA-01: MLOps for multi-modal machine learning in the SocialMinds laboratory.
Mentor: Dr. Séverin Lemaignan (severin.lemaignan@iiia.csic.es)
The newly established SocialMinds AI laboratory, part of the IIIA, is working on developing its machine learning workflows. We are seeking a motivated student to lead the design and implementation of our internal MLOps ecosystem. This role combines systems engineering with state of the art AI infrastructure to transform manual workflows into a unified, scalable platform. You will research, select and deploy a self-hosted solution to manage the lifecycle of our machine learning projects:
Data and Model Management: Implement storage solutions (for example, MinIO, DVC or MLflow) for tracking datasets and model versions.
GPU Orchestration: Configure a management layer to allocate and monitor our GPU cluster, ensuring efficient use across researchers.
Distributed Computing: Deploy frameworks to distribute large-scale computational workloads across multiple nodes.
Infrastructure as Code: Document and automate deployments using Docker, Kubernetes or Slurm to ensure a reproducible and stable environment.
You will gain hands-on experience in the “Ops” side of AI, which is rarely covered in academic curricula:
Systems Architecture: How to build an “AI Factory” from scratch.
Resource Optimisation: Management of high-performance hardware (NVIDIA GPUs) in multi-user environments.
Scalability: Identifying bottlenecks in large-scale models and addressing them through distributed training (Ray, Horovod or PyTorch Distributed).
IIIA-02: Robustness and Resilience of AI Systems.
Mentor: Dr. Daniel Gibert Llauradó (daniel.gibert@iiia.csic.es)
The primary objective of this project is to thoroughly examine the vulnerabilities and potential attack vectors present within AI systems deployed across diverse sectors such as cybersecurity and healthcare. The overarching goal is to gain a comprehensive understanding of the current limitations inherent in these systems and leverage these insights to fortify AI systems, ensuring robustness and resilience. The appointed candidate will be tasked with exploring the weaknesses within these AI systems, looking for vulnerabilities and developing attack methods designed to deliberately deceive and induce errors in the systems. Additionally, the candidate will explore an array of defense mechanisms aimed at mitigating the impact of adversarial attacks and failure models in AI systems.
IIIA-03: Using multiplayer game worlds to capture human social interaction data.
Mentor: Dr. Jordi Sabater Mir (jsabater@iiia.csic.es)
While modern humanoids have mastered physical tasks like walking and lifting, they still struggle to navigate the complex "unwritten rules" of human social interaction. We are looking for a motivated Master’s student to join us in solving this challenge through high-fidelity digital simulation. We believe that for robots and avatars to thrive in our world, they must learn from large-scale, ecologically valid social data. Our goal is to leverage the power of multi-player social simulation games to teach agents the nuances of human behaviour, from respecting personal space to mastering non-verbal cues.
As a JAE Scholarship recipient, you will play a pivotal role in evolving the RHYMAS platform. Built on Unreal Engine + Python, this innovative framework is designed for emergency training. Your task will be to adapt it into a sophisticated laboratory for social dynamics. You will: (1) Engineer Social Realism: Transform the platform to accurately simulate complex scenarios like group conversations or queuing. (2) Gamify Research: Re-engineer the framework into a "casual game" format, developing the interfaces and input systems needed for human players. (3) Capture Human Insight: Implement robust data-logging to record how humans navigate social friction, providing the "gold standard" data needed to train the next generation of AI.
We prioritise candidates with proficiency in Python, hands-on experience with Unreal Engine (including C++ programming, blueprints, and character animation), and the ability to design structured databases for data logging. We are seeking a creative problem-solver capable of bridging the gap between high-performance game development and the intricacies of psychological modelling
to drive our research forward.
IIIA-04: Theory and Applications of Multi-Objective Reinforcement Learning
Mentor: Dr. Manel Rodríguez Soto (manel.rodriguez@iiia.csic.es)
The purpose of this project is to investigate the formal frontiers of multi-objective reinforcement learning (MORL), an area with multiple applications such as Large Language Models (e.g., ChatGPT, DeepSeek), video games, or value alignment. The project will extend recent theoretical work by IIIA-CSIC on multi-objective reinforcement learning published at the world-renowned NeurIPS conference. We will put a strong focus on evaluating existing MORL algorithms with our novel theoretical tools. This evaluation will happen on two fronts. From an empirical front, we will design and deploy novel MORL algorithms exploiting our formal results. Then, we will compare these novel algorithms against the state-of-the-art in several test applications inspired by real-life problems (applying existing standard frameworks like MO-Gymnasium). From a formal front, we will focus on proving convergence theorems for the novel algorithms we develop. s. To guarantee that empirical results are formally significant, we will also perform experiments in controlled, mathematically provable test environments. The results of this research will put in the spotlight the capabilities and limitations of current multi-objective reinforcement learning algorithms.
IIIA-05: Application of Machine Learning in Epidemiology.
Mentor: Dr. Felip Manyà Serres (felip@iiia.csic.es)
Machine learning (ML) is a fundamental tool for early detection and prevention in epidemiology, as it enables the analysis of large volumes of data from health surveillance, genetic sequencing, mobility patterns, and environmental factors. By using ML models capable of identifying relevant patterns in disease dynamics—such as transmission routes, temporal trends, spatial clustering, seasonality, reproduction rates, and population susceptibility—these techniques facilitate the early detection of emerging health threats and support timely prevention and control strategies. Within this training plan, the student will work directly on the development and application of these ML methods, integrating into an interdisciplinary team that brings together experts in epidemiology and artificial intelligence. Throughout this process, the student will acquire skills in designing predictive ML models, processing epidemiological data, and interpreting results to support decision-making in public and animal health. As a result, the student will develop a solid understanding of the role of artificial intelligence in epidemiological surveillance and in improving early detection processes, ultimately contributing to more effective prevention, preparedness, and control of health threats.
IIIA-06: Generative Artificial Intelligence for Art-Therapy
Mentor: Dr. Juan Antonio Rodríguez Aguilar (jar@iiia.csic.es) / Dra. Maite López Sánchez
Nowadays, mental health and emotional well-being are of utmost importance, as they are essential for overall quality of life. In this context, art therapy has proven to be a valuable therapeutic approach, allowing individuals to explore and express emotions through creative processes.
With the recent advances in generative artificial intelligence, and particularly in conversational agents based on Large Language Models (LLMs) as well as other multi-modal models for image synthesis, new opportunities arise to support and augment therapeutic practices. These systems have shown strong capabilities in natural language interaction, emotional mirroring, and content generation. However, their use in art therapy remains largely unexplored and presents specific conceptual, technical, and ethical challenges.
The aim of this JAE intro project is to investigate to what extent Generative Models can be designed to assist art-therapy processes. Art therapy involves a triadic relationship between the therapist, the user, and the artwork. Interaction goes beyond dialogue alone, requiring the system to actively support the creation, transformation, and reflection around visual artefacts that are emotionally
meaningful for the user. Ensuring a therapeutically safe, non-judgmental, and reflective interaction is a key requirement, but yet of upmost importance is to ensure a (ethical) respectful interaction that supports emotional expression through visual media.
The project will involve the following main activities:
A review of the state of the art in generative AI, therapeutic chatbots, and value-alignment, together with an introduction to art therapy.
The design and development of a prototype combining conversational interaction based on LLMs with image generation capabilities.
The evaluation of the prototype through user studies
IIIA-07: Ethics and Artificial Intelligence in Global Health Governance: The One Health Approach
Mentor: Dra Núria Vallès Peris (nuria.valles@iiia.csic.es)
The research aims to analyse, from an ethical perspective that integrates the One Health approach, how the main international organisations conceptualise and guide the use of AI and robotics in the digitalisation of health.
Month 1– Theoretical framework
Document selection criteria will be defined and key international organisations (WHO, UNESCO, OECD, World Bank, FAO, among others) will be identified, as well as official document repositories.
Month 2 – Document collection
Strategic, regulatory and policy guidance documents on health digitisation, with a special focus on AI and robotics, produced by the international organisations will be collected in a database with basic metadata and a preliminary thematic classification.
Months 3, 4 – Thematic analysis
Qualitative thematic analysis will be applied to the selected documents, combining inductive and deductive approaches. Comparative matrices between organisations will be developed.
Months 4, 5 – Interviews
Semi-structured interviews will be conducted with officials of these international organisations for a comparison of the discourses in the documents with institutional practices, perceptions and challenges.
Month 6 – Integration and drafting
The project will culminate in a final report and recommendations for public policy and future research.
IIIA-08: Learning-Guided Heuristics for SAT and MaxSAT via Graph Neural Networks
Mentor: Dr. Jordi Levy (levy@iiia.csic.es)
The main objective of this project is to further investigate how Graph Neural Networks can be used to enhance key heuristics in SAT and MaxSAT solvers, with a particular emphasis on variable selection, clause weighting, and guidance of local-search or optimization procedures.
Rather than aiming at end-to-end learned solvers, the project focuses on hybrid approaches in which GNN-based predictions are integrated into existing symbolic algorithms, preserving their robustness while potentially improving performance on challenging instances.
The project will build upon existing SAT and MaxSAT solving frameworks and the experience acquired during the Master’s Thesis. The initial phase will focus on consolidating and extending implementations of GNN models tailored to graph representations of SAT formulas, together with baseline SAT and MaxSAT solvers.
As a first step, attention will be given to local-search and approximation algorithms, which typically require lower computational overhead and provide a suitable testbed for learning-guided decisions. In parallel, classical message-passing techniques such as belief propagation or survey propagation will be considered, both as standalone heuristics and as sources of informative signals for training
GNNs.
Training will be performed on small and medium-size instances, while evaluation will be carried out on larger and more diverse benchmarks. The performance of learning-guided heuristics will be systematically compared against state-of-the-art handcrafted heuristics, with respect to solution quality, runtime, and robustness.
IIIA-09: Solving Complex Scheduling Problems with the Help of Large Language Models
Mentor: Dr. Filippo Bistaffa (filippo.bistafa@iiia.csic.es)
Designing effective algorithms for complex scheduling problems (e.g., Job Shop Problem, Flow Shop Problem, or more generally, the Resource-Constrained Project Scheduling Problem) often relies on the availability of high-quality heuristic knowledge to handle realistic large-scale instances. Traditionally, such heuristics are crafted manually by experts, a process that is both time-consuming and highly dependent on domain knowledge.
Recent advances in large language models (LLMs) suggest new opportunities for supporting the algorithm design process by automatically extracting useful heuristic insights from textual problem descriptions. However, systematically leveraging these models for complex scheduling problems remains an open challenge, especially when the generated heuristics must be compatible with established optimisation and metaheuristic frameworks.
Along these lines, the following question arises: how can large language models be used as reliable assistants for deriving heuristic guidance in complex scheduling problems? Addressing this question requires identifying suitable ways of encoding problem constraints, guiding the interaction with the model, and validating the usefulness of the generated heuristics within algorithmic search.
Ultimately, this project aims to develop a methodology for extracting and validating heuristic information from large language models in the context of complex scheduling. As a primary use case, we will consider a scheduling problem in the domain of personalised education. For such a use case, the goal is to allocate a set of university subjects into a sequence of semesters so that students increase their skills for pursuing their desired career, while achieving the requirements for completing a Bachelor’s Degree. To assess the robustness and generality of the proposed approach, we will also evaluate it on established benchmark problems from the scheduling literature, including variants of the Resource-Constrained Project Scheduling Problem, comparing the LLM-suggested heuristics against state-of-the-art methods.
IIIA-10: Neurosymbolic Artificial Intelligence for Qualitative Spatial Reasoning
Mentor: Dr. Vicent Costa (vicent@iiia.csic.es)
Neurosymbolic artificial intelligence (AI) is a field in AI that aims to merge knowledge-based symbolic approaches with neural network-based methods. It is primarily driven by application-level considerations, such as explainability and interpretability, and seeks to combine the strengths of both approaches while overcoming their respective limitations. Qualitative Spatial Reasoning (QSR) is a research area within AI that focuses on automating reasoning about continuous aspects of the physical world, specifically space, to support problem-solving, planning, and scene understanding using qualitative rather than quantitative information. The main goal of this project is to use and integrate principles and techniques from neurosymbolic AI to design hybrid systems for QSR. Unlike purely numeric approaches, this project will focus on the symbolic representation and reasoning of spatial entities (such as topology, orientation, and distance). The application domains will be aligned with the tutor's previous work.
IIIA-11: Neurosymbolic Computation for an Embodied Approach to Diagrammatic Reasoning
Mentor: Dr. Marco Schorlemmer (marco@iiia.csic.es)
To address current barriers and limitations in the reasoning, abstraction, and analogy-making capabilities of neural-network-based AI systems, particularly in diagrammatic understanding and reasoning, we aim to conduct fundamental research to explore a new paradigm for neurosymbolic AI grounded in the principles of embodied cognition. This paradigm should combine the neural-network-based computation of a given input –the image of an abstract diagram like those used in mathematics and computer science– with the symbolic representation of image-schematic structure and dynamics, thus modelling an ‘embodied understanding' of the information processed by the neural network. (Image-schemas are a framework proposed in cognitive linguistics to capture the recurring dynamic patterns of our perceptual interaction and motor programs that give coherence and structure to our embodied experience.) Some neurosymbolic approaches typically aim to integrate the connectionist level directly with the level of predicates and logic. In contrast, our approach integrates the connectionist level with an intermediate level, where qualitative representations of perception can be combined with image schemas to support an ‘embodied understanding’ of an initial perception. This would enable the neurosymbolic system to perform reasoning, abstraction, and analogy in an integrated manner that is easier for humans to understand, owing to the embodied cognition framework.
