Bias detection in the computer vision field is a necessary task, to achieve fair models. These biases are usually due to undesirable correlations present in the data and learned by the model. Although explainability can be a way to gain insights into model behavior, reviewing explanations is not straightforward. This work proposes a methodology to analyze the model biases without using explainability. By doing so, we reduce the potential noise arising from explainability methods, and we minimize human noise during the analysis of explanations. The proposed methodology combines images of the original distribution with images of potential context biases and analyzes the effect produced in the model’s output. For this work, we first presented and released three new datasets generated by diffusion models. Next, we used the proposed methodology to analyze the context impact on the model’s prediction. Finally, we verified the reliability of the proposed methodology and the consistency of its results. We hope this tool will help practitioners to detect and mitigate potential biases, allowing them to obtain more reliable models.

Among the existing eXplainable AI (XAI) approaches, Feature Attribution methods are a popular option due to their interpretable nature. However, each method leads to a different solution, thus introducing uncertainty regarding their reliability and coherence with respect to the underlying model. This work introduces TextFocus , a metric for evaluating the faithfulness of Feature Attribution methods for Natural Language Processing (NLP) tasks involving classification. To address the absence of ground truth explanations for such methods, we introduce the concept of textual mosaics . A mosaic is composed of a combination of sentences belonging to different classes, which provides an implicit ground truth for attribution. The accuracy of explanations can be then evaluated by comparing feature attribution scores with the known class labels in the mosaic. The performance of six feature attribution methods is systematically compared on three sentence classification tasks by using TextFocus , with Integrated Gradients being the best overall method in terms of faithfulness and computational requirements. The proposed methodology fills a gap in NLP evaluation, by providing an objective way to assess Feature Attribution methods while finding their optimal parameters.

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.

AI explainability improves the transparency of models, making them more trustworthy. Such goals are motivated by the emergence of deep learning models, which are obscure by nature; even in the domain of images, where deep learning has succeeded the most, explainability is still poorly assessed. In the field of image recognition many feature attribution methods have been proposed with the purpose of explaining a model's behavior using visual cues. However, no metrics have been established so far to assess and select these methods objectively. In this paper we propose a consistent evaluation score for feature attribution methods -- the Focus -- designed to quantify their coherency to the task. While most previous work adds out-of-distribution noise to samples, we introduce a methodology to add noise from within the distribution. This is done through mosaics of instances from different classes, and the explanations these generate. On those, we compute a visual pseudo-precision metric, Focus. First, we show the robustness of the approach through a set of randomization experiments. Then we use Focus to compare six popular explainability techniques across several CNN architectures and classification datasets. Our results find some methods to be consistently reliable (LRP, GradCAM), while others produce class-agnostic explanations (SmoothGrad, IG). Finally we introduce another application of Focus, using it for the identification and characterization of biases found in models. This empowers bias-management tools, in another small step towards trustworthy AI.