In my years as a PhD student and post-doc researcher, my colleagues and I worked on numerous topics in natural language processing, including lexical distributional semantics (i.e. representing word meaning, word similarity and word relation), event knowledge representation (i.e. representing knowledge about events and their participants), sentiment analysis of non-literal tweets (i.e. deciding whether ironic or metaphorical tweets are positive or negative), grammar checkers (i.e. correcting orthographic, grammatical and stylistic writing mistakes), style transfer (i.e. turning texts from a style to another, such as from formal to informal), fake news detection, and information extraction (i.e. identifying relevant information in documents, reports etc).
In 2022, the Women’s Brain Project and Elsevier collaborated for the publication of a volume on Sex and Gender Bias in Technology and Artificial Intelligence, with a focus on Biomedicine and Healthcare Applications.
In 2019, I was invited to give a talk at the White House about the possible applications of Natural Language Processing in administration tasks. In the months that followed, I developed a proof of concept for the Office of Records Management (ORM). Since then, I have contributed to two editions (2019 and 2023) of the Artificial Intelligence and Machine Learning factsheet of the Belfer Center, Harvard, for the American Congress.
Between 2018 and 2019, in collaboration with the University of Pisa and the University of Udine, our MIT lab developed neural models that exploit graphical representations and reinforcement learning to study how to measure news divergence and automatically retrieve facts that may support or deny statements.
Our project dealt with the purpose of identifying fake news by observing (1) how texts about the same topic diverge form each other, and (2) whether it is possible to fact check the statements in those texts.
The project included the development of a dataset based on FEVER. Annotations were provided by TransPerfect and financed by Meta.
In the United States, the Center for Data Innovation has called for a national strategy to guide the integration of AI with healthcare.
Between 2018 and 2019, we focused on various types of cancer and heart failure data. We cooperated closely with hospitals and research institutes (e.g., Massachusetts General Hospital, Dana-Farber Cancer Insitute and more) to develop models for the automatic extraction of information from medical reports and images. Such information was then exploited for population-based studies, and for supporting risk assessment and clinical decisions. Relevant research was also carried out against COVID-19 and future pandemics.
The Prayer is an experimental setup that explores the possibilities of an approximation to celestial and numinous entities. It performs a potentially never-ending chain of religious routines and devotional attempts for communication through self-learning software.
A concept by Diemut Strebe, The Prayer is probably the first robot that speaks and sings to God—all Gods. A rough design (inspired by a machine produced by Japanese scientists that replicates the human vocal tract) is combined with a cutting-edge neural language model, fine-tuned on thousands of prayers and religious books from all over the world.
COVID-19 forced the world to reconsider its habits. Our homes have returned to be the place of salvation from an invisible enemy that advances silently, destroying years of work and reaping victims.
And it is precisely from our homes that we need to elaborate a new understanding of the limits of the society in which we have lived so far. A society whose resistance is being severely tested by this pandemic.
#DoItAtHome is a podcast series by Nicola Marino and Enrico Santus that aims to investigate today’s problems to design a better world for tomorrow.
Safe Paths is an MIT-led, free, open-source technology that enables jurisdictions and individuals to maximize privacy, while also maximizing the effectiveness of contact tracing in the case of a positive diagnosis.
The Safe Paths platform comprises both a smartphone application, PrivateKit, and a web application, Safe Places. The PrivateKit app will enable users to match the personal diary of location data on their smartphones with the anonymized, redacted, and blurred location history of infected patients.
Digital contact tracing uses overlapped GPS and Bluetooth trails that allow an individual to check if they have crossed paths with someone who was later diagnosed positive for the virus. Through Safe Places, public health officials are equipped to redact location trails of diagnosed carriers and thus broadcast location information with privacy protection for both diagnosed patients and local businesses.
In collaboration with the University of Udine, The Hong Kong Polytechnic University and Hong Kong University, we developed Artificial Intelligence and Natural Language Processing systems to understand the reasons behind vaccine hesitancy, contrasted misinformation and developed personalized educational strategies to mitigate this problem.
The first step in this process was the development of a social listening platform, COVID-19 Vaccine Opinion Analysis, which utilized state-of-the-art Natural Language Processing techniques (i.e., SpanBERT) to monitor daily vaccine opinions on Twitter. The system is able to extract information such as location, topics, sources of information, sentiment and side effects.
Featured in the first issue of the quarterly review of innovation for Italy: Center for studies on health and public administration, and published in collaboration with AiSDeT (Italian Association of Digital Health and Telemedicine). The journal on precision medicine intends to provide advanced, scientific and critical information on innovation processes in healthcare and their impact on the organizational and governance aspects of healthcare.
Find out more | Press