Research
Applying academic research to everyday problems
Our partners work together on research focused on practical solutions for universal challenges.
From R&D teams at leading tech startups, to our esteemed academic researchers, we’re focused on tackling tangible issues using a responsible AI framework. Learn more about our research projects, and how they’ll lead to future product innovation.
Energy-Efficient and Sustainable AI
This project will investigate and develop new scientific methods and technologies for making AI systems more energy and data efficient — an important goal for the European Green Deal.
This project will follow a three-pronged approach to enhance the efficiency and scalability of AI models: system-level optimization, cloud-level optimization and self-adaptation.
INESC-ID
Researchers will investigate ways to compress and distill pretrained models (like BERT, GPT-3 and ViLBERT) without compromising their accuracy.
IT
INESC-ID
Unbabel
This project aims to understand how nature evolves highly efficient systems like our brains, and how we can apply the same principles to ML.
CISUC
IT
Unbabel
Privacy-Preserving AI Systems
This project will develop methods for policy-compliant, privacy-preserving, and confidential-compute AI systems.
This project aims to find a balance between ensuring data privacy and security, and leveraging customer insights in a competitive landscape.
INESC-ID
YData
Since many ML/AI applications require data sharing, we will build software development frameworks that help preserve privacy.
Fraunhofer
FEUP
YData
YooniK
We will develop new methodologies and guidelines to help AI/ML developers ensure compliance with data protection (GDPR, HIPAA, CCPA, PIPEDA) and AI (EU AI Act) regulations.
Fraunhofer
YData
This project will look into meta-learning and data generation/manipulation in order to build robust algorithm cards that evaluate models based on responsible AI principles.
FEUP
Fraunhofer
INESC-ID
Transparent, Fair and Explainable AI
Understanding the decision-making processes of complex models is key to develop interpretable and explainable human-centered systems that interact positively and fairly with humans, including non-experts. This project will address these needs to allow a fairer, unbiased, reliable and more accountable use of AI in making decisions that affect human beings.
Combining the power of Explainable AI and causal inference to create transparent and interpretable models, unraveling causal relationships and providing insights for responsible decision-making.
CISUC
Champalimaud Foundation
Fraunhofer
INESC-ID
ISR-Lisboa
IT
Automaise
Unbabel
YData
This project will investigate and implement model-agnostic approaches, to estimate measures for the reliability/confidence/uncertainty of AI tools.
CISUC
Fraunhofer
INESC-ID
ISR-Lisboa
IT
Unbabel
CISUC
AI tends to reflect our own social biases. With this project, we’ll develop bias detection techniques, data “unbiasing” operations, and a methodology to automatically assess a model's fairness.
CISUC
Fraunhofer
ISR-Lisboa
Unbabel
YData
We will develop trustworthiness mechanisms to support human-computer collaboration, and explore those through co-creation tasks.
CISUC
INESC-ID
ISR-Lisboa
Unbabel
Language Technologies and Embodied Human-AI Interaction
This project will focus on investigating new techniques for robust and trustworthy NLP and vision technologies, envisioning a scenario where humans and AI systems work together collaboratively.
This project will look into NLP and MT and its integration with complex knowledge bases, such as medical terminology, and wider media-related applications.
IT
INESC-ID
Automaise
NeuralShift
Priberam
Unbabel
We will develop explainability mechanisms tailored for computer vision medical applications, in order to improve patient diagnosis and treatment.
ISR-Lisboa
INESC-ID
Emotai
NeuralShift
Social AI aims at creating AI that responds and interacts with humans in a natural, responsive and responsible way, creating a symbiotic relationship.
INESC-ID
ISR-Lisboa
We propose a new approach to brain-computer interface (BCI) for language input based on a seamless interface combined with a large pre-trained neural language model.
Champalimaud Foundation
ISR-Lisboa
IT
Emotai
Unbabel
Recent advances in robotics have created a new generation of teleoperation robots. We’ll make them more natural by designing systems that comply with human social norms.
ISR-Lisboa
INESC-ID
Multilingual and Contextualized Conversational AI
This project will advance the state of the art in conversational AI by making progress in two fronts: removing language barriers, by designing systems that are multilingual (via machine translation technologies), and by enhancing dialogue systems with contextualization, with particular focus on customer service communication.
This project will investigate conversational AI and retrieval-augmented language generation, improving the generation of trustworthy dialogue responses based on previous context and external information.
INESC-ID
Unbabel
Visor.ai
We’ll be continuing work on context-aware MT based on conversational data such as emails and chat, to improve quality and speed of translations.
IT
INESC-ID
Unbabel