Mario Munoz-Organero

Machine learning (ML) models often benefit from learning across diverse datasets to improve generalization and performance. In scenarios where data is distributed across different institutions or sources, sharing raw data may not be feasible due to privacy or regulatory constraints. Federated Learning has emerged as a solution for these scenarios where data cannot be shared. However, challenges remain when training Federated Nodes on non-homogeneous datasets. Federated Transfer Learning has been proposed as a solution to federated learning scenarios in which data corresponds with different users and does not comprise the same set of features. However, privacy and efficiency challenges remain in existing solutions. This paper presents a novel architecture for Federated Transfer Learning scenarios that allows multiple models to train jointly across different datasets, sharing part of the weights for a common part in the models while maintaining a private component for each participant able to adapt to the particularities of each dataset. The proposed architecture allows all clients to work on a shared embedding space, so that the model can learn both information common to all clients and information specific to each institution during training. This means that each client can contribute its own knowledge without sharing private data, while the common part of the model is combined to improve global learning. This framework facilitates the development of robust machine learning models in environments where data cannot be centralized, enabling more effective learning from heterogeneous sources without compromising data ownership.

As we move from the Information Age to the Intelligence Age, universities must redefine themselves taking into account recent challenges. Two of these challenges are Micro-credentials and Artificial Intelligence (AI). Micro-credentials certify the learning outcomes of short-term learning experiences, which are typically more flexible and tailored to specific, job-relevant skills, meeting the increasing demand for continuous education. These learning experiences are normally referred to as micro-credential courses or micro-credential programs and disrupt traditional degree models. On the other hand, Artificial Intelligence is disrupting every single aspect of our life and work. Artificial Intelligence tools can work as assistants helping in all kinds of tasks that were previously reserved for humans. Artificial Intelligence can be used in education in many ways, such as to create personalized learning paths, create and optimize educational multimedia content such as text, voice, image, or video, tutor students, and monitor student progress in real-time, contributing to the acquisition of learning outcomes. Putting the two challenges together, Artificial Intelligence can also contribute to generate micro-credentials by ensuring learners demonstrate competence in practical, job-specific skills, enhancing their credibility for employers. Therefore, Artificial Intelligence can playa pivotal role in advancing micro-credential courses and programs, helping universities redefine their offerings to provide personalized, adaptable, and industry-relevant learning experiences. Artificial Intelligence can help in designing the content and the learning experience as with courses in general, but it can also help in designing the specificities of micro-credential courses. The specific “skills” provided by Artificial Intelligence for defining micro-credentials range from clerical work related to mastering formats such as ELM or OpenBadges that are needed to construct digital credentials to creative sug-gestions that aid in designing the details of the micro-credential course. Alternatively, the micro-credentials designed can include topics about Artificial Intelligence, helping the workforce on this important area. This paper analyzes the interplay of these two challenges and how they can help each other, making opportunities out of challenges.

Travel dynamics significantly impact commuter stress, influenced by traffic behavior, road conditions, travel modes, distance, and socio-demographic characteristics. Previous research on travel stress often exhibits limitations, including narrow scopes focusing on specific routes, vehicle types, or demographics. This study addresses these constraints by employing a comprehensive approach to analyze the influence of various travel attributes on commuter stress levels. An interview-based dataset was collected to capture the multifaceted experiences of road users. Five tree-based machine learning models–Decision Tree (DT), Random Forests (RF), Extra Trees (ET), Extreme Gradient Boosting (XGBoost), and k-Nearest Neighbor (k-NN)–were deployed for imbalanced multi-class classification. XGBoost demonstrated superior performance with the highest accuracy (73.33%) and precision (75.63%) with a standard deviation of ±5.9. A novel double hyperparameter optimization technique enhanced the prediction accuracy across all models, notably increasing the k-NN classifier’s accuracy to 19.99%. The SHAP (SHapley Additive exPlanations) method was utilized for model interpretability, revealing distance traveled per day as the most influential factor across stress levels, followed by mode of transport, gender, and age for low, medium, and high-stress categories, respectively. The study also examines the impact of features on individual commuter stress levels through random instance selection. This research provides valuable insights into the complex interplay between travel attributes and commuter stress, paving the way for the development of effective stress mitigation strategies and improved travel experiences for all road users.

Origin-Destination (OD) traffic flow estimations from traffic sensor data play an important role for transportation planning and management. This paper proposes a novel method to compare OD traffic estimated matrices (using data from traffic sensors). The proposed method uses the estimated OD traffic flow values together with COVID-19 incidence data in order to build a sequence of temporal graphs that are fed into a machine learning (ML) model. The ML model uses the input information to estimate/predict one-week ahead COVID-19 incidence data. A tailored Graph Neural Network (GNN) and Long Short-Term Memory (LSTM) model is designed adapted to the input information. The paper evaluates the proposed method with 3 different OD estimation alternatives and compares the accuracy achieved by different configurations of the ML model with a traffic agnostic baseline model. Data from 44 provinces in Spain during 2021 providing daily COVID-19 incidence data and 635 geo-located traffic sensors providing monthly traffic counts are used to evaluate the results. The 3 traffic-aware OD estimation methods were able to outperform the baseline model, achieving model gains up to 136%. The major application of the results of this paper is a novel mechanism to validate prior OD traffic matrices.

In Colombia, coffee futures contracts represent essential financial agreements that allow producers and buyers to establish prices, quality, and conditions for future transactions in the coffee market. Despite the evident benefits of stability and predictability, this practice faces significant sustainability challenges that threaten its long-term viability. One of the reasons is the significant lack of transparency in the supply chain. Farmers, affected by abrupt price fluctuations and adverse weather conditions such as the El Niño phenomenon, experience an increase in market prices, leading to the non-delivery of the final product, and contract breaches as they find better prices in the local market. In this context, smart contracts emerge as a promising technological solution to address these problems. These contracts enable the verification of each step in the process, from harvest to final sale, within a blockchain. Therefore, this research designs a smart contract managed through a platform called SmartBeanFutures, which records the clauses of futures contracts using the IERC721 framework, allowing the generation of a unique and non-repeatable asset. It aims to sell, promote, and manage coffee sale prices during the agreement’s signing, creating a transparent environment for chain actors. This proposal undergoes evaluation in a test environment, providing farmers access to the designed platform. Following the validation of the proposal, it was identified that over 74% would use this type of contract in their agricultural processes, highlighting that implementing this technology contributes to eliminating intermediaries in the chain and gives farmers more control over their participation in the market.

The possible applications of GenAI (Generative AI) in education alone are so manyfold and overwhelming that it is useful to have an overview of the many possibilities that are opening up. In this paper, we try to organize some of the low-hanging fruits that can help instructors, learners, and educational managers use GenAI applications to improve educational performance. For instructors, GenAI can help in gaining a deeper understanding of the topics to be taught, preparing educational materials, and facilitating the enactment phase in class. Learners can be assisted in getting personalized content and feedback, having GenAI as a participant in forums, or for self-reflection and emotions detection. Managers and other stakeholders can profit from Academic Analytics, bias detection, course repurposing, and many other uses. In this paper, we also present a use case detailing some initial actions we are implementing for a Programming with Java course. One action is to explicitly identify, in each problem set, the competencies being developed. Another one is the development of a chatbot, fine-tuned with the course material, which can be used by students as a tutor. The third action is to use a GenAI tool to generate questions aimed at assessing whether students truly grasp the programming project they have supposedly developed. AI is here to stay, in spite of the issues it opens up, and therefore it is never too early to start experimenting with it in practice.