ELLIS Unit Lisbon: Advancing Human-Interacting AI and Natural Language Processing

ELLIS Unit Lisbon, established in 2020 as Portugal’s only ELLIS Unit, focuses on Human-Interacting Responsible AI and advancing Natural Language Processing. Based at Instituto Superior Técnico and composed of three leading research institutes, it promotes ethical, impactful AI through strong academic, industry, and policy collaborations. The Unit includes 21 ELLIS Members and actively shapes national and European AI innovation.

Mission & Vision 

The ELLIS Unit Lisbon was created with a mission to become the leading AI and ML research unit in Portugal, establish strong links with Europe, and boost collaborative research and training in the field through participation in events, joint activities, and mobility programs. With a focus on Human-Interacting Responsible AI, research within the Unit aims to develop robust, open, explainable, trustworthy, and efficient AI.

The Unit has strong connections to various research institutions and industry partners, which are key to translating fundamental AI and ML research into ethical products and help maximise the social, environmental, and economic impact of the Unit’s research in Europe and the world. Within Portugal, the Unit has played an important role in shaping national AI policy and innovation, and several of its researchers have established advisory links with governmental bodies. 

The creation of the ELLIS Unit Lisbon has played an important role in the development of AI and ML synergies both within the Lisbon ecosystem and with our European colleagues, and we look forward to establishing further collaborations and helping push boundaries. - Mário Figueiredo, Director of the ELLIS Unit Lisbon

Key Numbers and Milestones

• The Unit Lisbon joined ELLIS in 2020 and is currently the only Unit in Portugal 

• It is composed of three institutions, Instituto de Telecomunicações (IT), Instituto de Engenharia de Sistemas e Computadores (INESC-ID), and Institute for Systems and Robotics (ISR-Lisboa)

• The Unit is hosted within the Instituto Superior Técnico, the school of engineering of the University of Lisbon, and has grown to include 21 ELLIS Members, including 8 Fellows and 3 Scholars.

• Since its creation, the Unit has hosted 7 PhD students from the ELLIS PhD Program

• Published 40 articles in collaboration with other ELLIS Members

• Participated in 11 collaborative events with other Units, and held two Unit PhD and Postdoc Symposia

The Unit Lisbon is particularly strong in Natural Language Processing (NLP), and ELLIS Fellow and Co-Director of the ELLIS NLP Program, André Martins, is co-leading the development of an open-source European large language model, EuroLLM, which supports all 24 official EU languages, plus 11 other important languages. These models are currently being adopted by the Portuguese government and are expected to impact education, public administration, and business sectors. We anticipate these models to be adopted by other EU governments.

Research Areas 

The Unit Lisbon aims to use efficient, reliable, theory-grounded ML to create human-interacting responsible and sustainable AI systems, and conducts cutting-edge research in the following themes: 

• Natural Language Processing - concerned with the analysis, interpretation, and generation of human language and its automation in technology. Research in NLP focuses on computational linguistics, speech recognition and processing, multilingual LLMs, machine translation, and retrieval-augmented language generation.
• Machine Learning Foundations - focuses on developing machine learning tools and algorithms with robust AI applications. Research in this field includes pattern recognition, representation, evaluation, and optimisation of models, with widespread applications to healthcare, communication, surveillance, and other areas.
• Reinforcement Learning & Robotics - applies reinforcement learning, active learning, and decision-making systems to the development of intelligent and social artificial (robotic) agents. The research includes the combination of ML and AI with social modeling to study hybrid societies of humans and machines and understand multi-robot collaboration.
• Computer Vision & Embodied Cognition - designs naturalistic artificial (robotic) systems that can understand and interact with the world around them by using vision and other cognitive systems to model the world.  
• Health - develops computational approaches and algorithms for analysing clinical datasets, and explainability mechanisms tailored for computer vision medical applications to help improve patient diagnosis and treatment.
• Infrastructure - develops relevant infrastructure and systems that will support advances in AI and improve its capability. 

Ongoing Collaborations

Since its inception, the Unit Lisbon has established strong collaborations with other ELLIS Units. In NLP, there are ongoing partnerships with the ELLIS Units Edinburgh, Amsterdam, Munich, and Darmstadt. One of the most notable of these collaborations is the EuroLLM project funded through the EuroHPC programme and co-funded by the EU. Launched in 2024 and led by André Martins, the team is currently developing the most robust and advanced open-source multilingual LLM in Europe, capable of understanding and generating text in all 24 EU languages and other strategically important ones, contributing to creating a competitive and innovative European AI landscape. The project has already released models EuroLLM-1.7B and EuroLLM-9B, which have over 200K downloads on Huggingface.

In Robotics, the Unit has connections with the ELLIS Units Lausanne and Genoa, and through ELLIS Member Pedro Lima, currently participates in the euROBIN network, a consortium of 31 partners spread across 14 countries that aims to develop cognition-enabled embodied AI in intelligent machines (robots) that can interact with humans.

In Health, the Unit has established close collaborations with the ELLIS Units Heidelberg and Zurich primarily through the Olissipo twinning project, led by ELLIS Member Susana Vinga, which aimed to strengthen computational biology and innovation in the Lisbon Unit. Through this project, various joint events were organised, including Summer and Winter Schools.

At a national level, many of the Unit Lisbon Members are involved in the Center for Responsible AI, a consortium involving 10 startups, 8 research centres, and 1 law firm, funded through Recovery and Resilience funds, and which aims to develop trustworthy, fair, transparent, and sustainable AI products. Some of the products being developed by this consortium involve using AI for telerehabilitation and physical therapy, developing culturally-aware customer service translation systems, and combining non-invasive neural interfaces with LLMs to allow for communication in patients with ALS and other conditions where communication has been impaired.

Supporting and Fostering Young Talent

Since 2011, the Unit Lisbon has hosted the annual Lisbon Machine Learning School (LxMLS). LxMLS is one of the top summer schools in ML and Natural Language Processing and attracts approximately 200 students from all over the world for a whole week of intense teaching and training. This Summer School provides opportunities for early career researchers (ECRs) in the Unit Lisbon to become involved in teaching and mentoring younger students by participating in tutorials and laboratory classes.

The Unit Lisbon holds an annual PhD and Postdoc Symposium intending to promote the work of ECRs within the ML and AI community at Técnico. The Symposium features flashtalks and poster presentations by the students and postdocs and provides a platform for ECRs and the rest of the community to get to know each other’s work.

ELLIS itself has been fantastic at providing opportunities for scientific exchanges for ECRs. Last year, several students from the Unit used funds from the ELISE and ELSA projects to travel to other ELLIS Units and gain knowledge of new techniques that are crucial for their research. These funds have been fundamental for networking and fostering collaboration.

Efforts in Public Engagement

Explaining AI to the general public: The Unit Lisbon has actively participated in promoting and explaining AI to the general public, primarily through the participation of several of our members in various programmes for national tv networks and radio. Notably, ELLIS Fellow Arlindo Oliveira has written several popular science books about AI for the public and writes a monthly column in a national newspaper about developments in AI. Arlindo and Mário Figueiredo, our Director, have  participated in several television and radio programs and conference panels that discuss AI and its impact on society. 

“Girls in AI”: In terms of outreach programs, the Unit Lisbon is committed to promoting gender diversity and equality in AI through the “Girls in AI” campaign, which aims to inspire young girls to follow a career in AI and STEM in general. 

“European Researchers Night”: Equally, the Unit’s research has also been exhibited in past editions of the “European Researchers Night”, an event open to the public to increase interest and participation in science.