The adoption of algorithms based on Artificial Intelligence (AI) has been rapidly increasing during the last few years. However, some aspects of AI techniques are under heavy scrutiny. For instance, in many use cases, it is not clear whether the decisions of an algorithm are well informed and conforming to human understanding. Having ways to address these concerns is crucial in many domains, especially whenever humans and intelligent (physical or virtual) agents must cooperate in a shared environment. In this paper, we apply an explainability method based on the creation of a Policy Graph (PG) based on discrete predicates that represent and explain a trained agent’s behaviour in a multi-agent cooperative environment. We show that from these policy graphs, policies for surrogate interpretable agents can be automatically generated. These policies can be used to measure the reliability of the explanations enabled by the PGs through a fair behavioural comparison between the original opaque agent and the surrogate one. The contributions of this paper represent the first use case of policy graphs in the context of explaining agent behaviour in cooperative multi-agent scenarios and present experimental results that sets this kind of scenario apart from previous implementations in single-agent scenarios: when requiring cooperative behaviour, predicates that allow representing observations about the other agents are crucial to replicate the opaque agent’s behaviour and increase the reliability of explanations.

The growing interest in the application of Large Language Models (LLMs) for healthcare comes with a demand for better open-source LLMs, and stronger reassurances regarding their performance. To advance in this direction, this work conducts a thorough and transparent study of LLM model training and benchmarking in healthcare, releasing as open assets all resources needed for reproducing the Aloe models and its results (weights, data and code). This includes details on optimized data preprocessing and training, combining curated public data with synthetic samples for a total of 1.8B training tokens; enhanced safety, induced through Direct Preference Optimization (DPO), aligning Aloe models for ethical robustness and against jailbreaking attacks; and finally model performance, evaluated thoroughly through close-ended, open-ended, safety, and human assessments. To boost inference efficacy and test the upper bounds of open LLM performance, Aloe models are integrated with a Retrieval-Augmented Generation (RAG) system. The resultant models deliver competitive performance across healthcare benchmarks and medical fields while significantly improving safety and bias resilience. Model weights are released for research-only purposes, together with training and evaluation datasets, and RAG inference code. To enable the responsible release of such technology, this work is supported by a detailed healthcare-specific risk assessment. Building on top of base models like Llama 3.1 and Qwen 2.5, the Aloe models and their development recipe set a high standard for open-source medical LLMs, balancing top-tier performance with high ethical requirements.

Agents are a special kind of AI-based software in that they interact in complex environments and have increased potential for emergent behaviour. Explaining such emergent behaviour is key to deploying trustworthy AI, but the increasing complexity and opaque nature of many agent implementations makes this hard. In this work, we propose a Probabilistic Graphical Model along with a pipeline for designing such model -- by which the behaviour of an agent can be deliberated about -- and for computing a robust numerical value for the intentions the agent has at any moment. We contribute measurements that evaluate the interpretability and reliability of explanations provided, and enables explainability questions such as `what do you want to do now?' (e.g. deliver soup) `how do you plan to do it?' (e.g. returning a plan that considers its skills and the world), and `why would you take this action at this state?' (e.g. explaining how that furthers or hinders its own goals). This model can be constructed by taking partial observations of the agent's actions and world states, and we provide an iterative workflow for increasing the proposed measurements through better design and/or pointing out irrational agent behaviour.

In deep learning, transfer learning (TL) has become the de facto approach when dealing with image related tasks. Visual features learnt for one task have been shown to be reusable for other tasks, improving performance significantly. By reusing deep representations, TL enables the use of deep models in domains with limited data availability, limited computational resources and/or limited access to human experts. Domains which include the vast majority of real-life applications. This paper conducts an experimental evaluation of TL, exploring its trade-offs with respect to performance, environmental footprint, human hours and computational requirements. Results highlight the cases were a cheap feature extraction approach is preferable, and the situations where an expensive fine-tuning effort may be worth the added cost. Finally, a set of guidelines on the use of TL are proposed.

The COVID-19 pandemic has already caused more than 150,000,000 cases worldwide. In Spain this has lead to a massive and simultaneous saturation of all sanitary regions. Coherently, the quick and consistent understanding of the COVID-19 disease requires of the combined analysis of thousands of medical records generated by dozens of different institutions. In the context of the publicly funded CIBERES-UCI-COVID project, we have gathered, cleaned and preprocessed data from heterogeneous sources – more than 30 hospitals, with different data entry systems – in order to produce a unified database, of more than 6.000 patients, that is used in several clinical studies being carried by different multidisciplinary groups. In this paper, we identify the complexities we encountered, the solutions we applied, and we summarise the statistical and machine learning techniques we have applied for the studies.