Visual question answering has recently been settled as a fundamental multi-modal reasoning task of artificial intelligence that allows users to get information about visual content by asking questions in natural language. In the cultural heritage domain, this task can contribute to assisting visitors in museums and cultural sites, thus increasing engagement. However, the development of visual question answering models for cultural heritage is prevented by the lack of suitable large-scale datasets. To meet this demand, we built a large-scale heterogeneous and multilingual (Italian and English) dataset for cultural heritage that comprises approximately 500K Italian cultural assets and 6.5M question-answer pairs. We propose a novel formulation of the task that requires reasoning over both the visual content and an associated natural language description, and present baselines for this task. Results show that the current state of the art is reasonably effective but still far from satisfactory; therefore, further research in this area is recommended. Nonetheless, we also present a holistic baseline to address visual and contextual questions and foster future research on the topic.

The COVID-19 pandemic highlighted the importance of validated and updated scientific information to help policy makers, healthcare professionals, and the public. The speed in disseminating reliable information and the subsequent guidelines and policy implementation are also essential to save as many lives as possible. Trustworthy guidelines should be based on a systematic evidence review which uses reproducible analytical methods to collect secondary data and analyse them. However, the guidelines’ drafting process is time consuming and requires a great deal of resources. This paper aims to highlight the importance of accelerating and streamlining the extraction and synthesis of scientific evidence, specifically within the systematic review process. To do so, this paper describes the COKE (COVID-19 Knowledge Extraction framework for next generation discovery science) Project, which involves the use of machine reading and deep learning to design and implement a semi-automated system that supports and enhances the systematic literature review and guideline drafting processes. Specifically, we propose a framework for aiding in the literature selection and navigation process that employs natural language processing and clustering techniques for selecting and organizing the literature for human consultation, according to PICO (Population/Problem, Intervention, Comparison, and Outcome) elements. We show some preliminary results of the automatic classification of sentences on a dataset of abstracts related to COVID-19.

Distinguishing whether some code portions were implemented by humans or generated by a tool based on artificial intelligence has become hard. However, such a classification would be important as it could point developers towards some further validation for the produced code. Additionally, it holds significant importance in security, legal contexts, and educational settings, where upholding academic integrity is of utmost importance. We present EX-CODE, a novel and explainable model that leverages the probability of the occurrence of some tokens, within a code snippet, estimated according to a language model, to distinguish human-written from AI-generated code. EX-CODE has been evaluated on a heterogeneous real-world dataset and stands out for its ability to provide human-understandable explanations of its outcomes. It achieves this by uncovering the features that for a snippet of code make it classified as human-written code (or AI-generated code).