Collaborations

Companies

Fabbrica d'Armi Pietro Beretta

We reviewed how digital representations of people can help researchers study movement and physical activity. The review brought together research on human digital twins in sport and rehabilitation.

Airbus

We took part in a joint demonstration with Airbus in the DAIS programme in 2017. The linked technical report records the work presented together.

We studied how AI can account for uncertainty when assigning a case to a category, including identifying stages of a cyber attack. Other work examined how to reason about probabilities when the evidence itself is uncertain.

BAE Systems AI Labs

We reviewed methods for understanding why deep learning systems produce particular predictions. The survey brought together research on explaining model behaviour, rather than treating an AI answer as a result that cannot be examined.

Blue Prism AI Labs

We studied how AI can distinguish gaps in its knowledge from variation in the data. The work examined ways to learn from incomplete information and represent uncertainty in predictions.

Honeywell

We studied how people and software can work together to analyse intelligence information. The work examined how to reason with different kinds of evidence and included an evaluation by experts.

IBM Research Hursley (UK)

We studied how people can understand AI recommendations and judge when to trust them. The work connected explanations and uncertainty with making sense of complex situations, and also examined privacy rules and social media moderation.

IBM T. J. Watson Research Center

We studied how people and AI can reason with complex information and judge the reliability of an answer. Shared work also examined the risks of sharing information, ways to limit the spread of sensitive data and the application of privacy rules.

NEC Laboratories Europe

This research connection links academic interests in AI with NEC’s European industrial research community.

Institutions and research centres

US Army Research Laboratory (DEVCOM)

We studied how AI can make sense of changing, uncertain information. The work combined learning and reasoning to recognise events in sensor data, examine cyber threats and help people judge how much confidence to place in an answer.

US Air Force Research Laboratory

We studied how to interpret sensor data when computing resources are limited. The work combined AI methods to recognise events and explored how Wi-Fi signals can be used for passive radar.

CNRS

We helped compare software that reasons about arguments supporting or challenging one another. The shared report describes the first international competition assessing these systems on common problems.

INRIA

We studied how a system of arguments changes its conclusions when it receives different inputs. The work developed formal models for describing what goes into the reasoning process and what comes out.

NATO CMRE

We studied how to make AI interpretation of passive radar data easier to understand. The work examined a learning method for interpreting these signals, with a focus on making the model’s behaviour more explainable.

The Alan Turing Institute

We studied how AI agents can develop ways to communicate about space and time that people can interpret. Other work examined how to guide an agent’s learning by combining rewards that reflect different values.

UK Defence Science and Technology Laboratory

We studied how people can judge when to trust AI by seeing explanations and uncertainty alongside its answers. Shared work also examined how to support scientific enquiry when sources are uncertain.

Universities

Bar-Ilan University

We explored how software can help people follow an online debate by representing the arguments and how they support or challenge one another. The work took the form of a pilot study.

BI Norwegian Business School

We studied how to draw conclusions from a network of related information when the probabilities are themselves uncertain. The work developed a method for carrying that uncertainty through the calculation.

Carnegie Mellon University

We studied how to assess the risks of sharing information and how trust can inform that assessment. Other work examined how to reason with structured knowledge when some of it is uncertain.

Delft University of Technology

We studied how to protect privacy against systems that use Wi-Fi signals to sense people or activity. The work analysed, implemented and evaluated a method that changes how wireless signals are transmitted.

Heidelberg University

We explored how to make online debates easier to follow by representing which arguments support or challenge others. A pilot study examined the use of this approach in online discussion.

Holon Institute of Technology

We examined how reasoning about causes relates to reasoning through arguments. The shared work explored the connection between explaining why something happens and assessing reasons for a conclusion.

Hong Kong University of Science and Technology

We studied how AI can recognise complex events as data arrives from connected sensors and devices. The work explored models that combine learning with reasoning, including ways to train them using generated examples.

King's College London

We worked on a special journal issue about how software agents can exchange and assess arguments. This editorial collaboration brought together research on reasoning through disagreement in systems with multiple agents.

KU Leuven

We studied how AI can recognise events when data may be manipulated, and how it can distinguish missing knowledge from variation in the data. Shared work also examined how software agents negotiate with limited resources at the edge of a network.

Leipzig University

We helped develop common tests for software that evaluates competing arguments. The work included a framework for comparing systems and a report on the first international competition in this area.

Maastricht University

We examined how explanations based on causes connect with arguments for or against a conclusion. The shared work explored how these two forms of reasoning relate to one another.

Massachusetts Institute of Technology

We worked on a shared way to represent arguments so software can describe and exchange them consistently. The collaboration also included a special journal issue on argument-based reasoning among software agents.

Mimar Sinan Fine Arts University

We studied how knowledge held by local communities finds its way into plans for managing World Heritage sites. The work developed a method drawing on several disciplines to examine that transfer of knowledge.

Ohio State University

We studied a way to make it harder to infer private information from Wi-Fi signals. The work analysed and tested changes to wireless transmission intended to counter unwanted sensing.

Ozyegin University

We studied how AI can represent uncertainty and help people interpret complex situations. The work covered learning from uncertain data, explaining predictions and reasoning with people, as well as applying privacy rules.

Politecnico di Milano

We studied how AI can interpret sensor data, recognise events with limited computing resources and represent uncertainty when data is missing. Shared work also addressed Wi-Fi privacy, tropical cyclone formation and how to model urban planning around people’s needs.

Politecnico di Torino

We studied how attackers can exploit similar-sounding names across languages to deceive people online, and how AI can generate or help detect deceptive names. Shared work also examined resilience against persistent cyber attacks.

Sapienza University of Rome

We studied how AI can learn models for planning actions from visual data and connect what it sees with the concepts used in a plan. Shared work also examined recognising events and their causes with limited computing resources.

Swansea University

We examined how to make legal reasoning explicit by representing evidence and arguments about a conclusion. The shared studies analysed reasoning in the Karadžić trial judgment, including the question of intent.

TU Darmstadt

We studied how to protect people from unwanted sensing through Wi-Fi signals. The work analysed, implemented and evaluated a method that varies wireless transmission to make such sensing more difficult.

University College London

We studied how to account for risk and uncertainty when AI assigns a case to a category. The collaboration also reviewed methods for understanding and explaining the predictions of deep learning models.

University of Aberdeen

We studied how people and software can assess evidence, trust and the risks of sharing information. Shared work also examined software agents negotiating contracts and methods for comparing systems that reason about competing arguments.

University of California, Los Angeles (UCLA)

We studied how AI can turn streams of sensor data into an understanding of events. The collaboration combined learning and reasoning, examined how to explain results and uncertainty, and included expert evaluation of people working with software in intelligence analysis.

University of Dundee

We worked on a common format for describing arguments and the relationships between them. The aim was to let software represent and exchange argument structures using shared concepts.

University of Hagen

We studied algorithms for deciding which arguments can be accepted when arguments challenge one another. The work addressed how software can calculate defensible conclusions from these relationships.

University of Huddersfield

We developed and compared algorithms that determine which conclusions can be defended when arguments conflict. The work included software tools, common test problems and ways to choose or combine algorithms for difficult cases.

University of Koblenz

We helped assess software that reasons about competing arguments through a common international competition. The shared report describes the systems and results of that comparison.

University of Liverpool

We studied how software can reason about arguments that support or challenge conclusions, including structures that can continue indefinitely. The collaboration also included editorial work on argument-based reasoning among software agents.

University of Luxembourg

We studied how new information affects the conclusions produced by a system of arguments. The work developed formal descriptions of the relationship between the system’s inputs and outputs.

University of Michigan

We studied how AI can learn from incomplete data and reason when probabilities are uncertain. The work developed methods to keep track of that uncertainty as information is combined.

University of Oslo

We studied how people can remain involved when AI combines uncertain information to understand a situation. The work used networks of related information that represent both beliefs and uncertainty.

University of St Andrews

We studied tools that help analysts connect facts, examine arguments and prepare reports. The work included CISpaces, a system for highlighting weaknesses in a case, and an expert evaluation of collaboration between people and software.

University of Tokyo

We studied how a dialogue can help participants work towards reconciliation. The work developed formal models for changing the direction of deliberation when searching for a way to resolve disagreement.

University of Turin

We studied how to extract structured knowledge from cybersecurity logs using language models, and how attackers exploit similar-sounding names across languages. Other shared work examined cyber resilience and how systems of arguments respond to new information.

University of Utrecht

We worked on shared concepts for representing arguments in software. The resulting format describes arguments and their relationships in a way that different systems can use consistently.

University of Zagreb

We explored how software can represent the structure of an online debate so people can see which arguments support or challenge others. A pilot study examined this approach in online discussion.