Projects

As TNO, we look at the entire lifecycle of AI systems and seek technical and organisational solutions that improve AI applications’ lifecycle, reliability, and deployability.

‘ELSA’ forms an ecosystem within TNO in which we connect projects with focus on ethical, legal, and societal aspects. On this page you can find more information about the projects

Research in this programme line focuses on responsible and explainable AI, with specific attention to the correct use of data. To arrive at responsible decisions, a safe environment in which humans and AI can learn from each other (co-learning and secure learning) is needed

The development of AI algorithms and software that allow autonomous systems to be deployed safely and effectively in unpredictable and dynamic environments.

The objective of the AI Oversight Lab is to enable public entities to develop and use AI tools, while adhering to ethical principles and respecting human rights, and to facilitate oversight bodies in their auditing role.

The FATE flagship develops AI capabilities for a digital assistant that acquires and extends its expertise through continuous learning from multiple potentially confidential and biased (subject) data sources and from human experts who add to and reflect on the AI-outcomes. The system provides decision support for multiple user roles, such as a researcher, consultant, and subject.

In the GRAIL project, we are working towards a toolbox to assist companies in developing and using GenAI responsibly. This toolbox contains methods, guidelines, and functionalities covering three topics; governance, application, and evaluation of GenAI. Expertise and tools are co-developed as much as possible within GRAIL use case projects, in close collaboration with stakeholders. These use cases address concrete challenges faced by stakeholders, ensuring that the resulting solutions are directly relevant and applicable. This collaborative, use-driven approach ensures that the toolbox not only reflects real-world needs but also delivers meaningful and lasting impact.

Individually, major banks are facing significant challenges with respect to detecting money laundering. Due to privacy regulations and data confidentiality banks cannot readily share data, which makes it challenging to collaborate. The goal of this joint research project is to develop a Proof of Concept Multi-Party Computation solution for collaborative monitoring and detection of money laundering.

The SEAMLESS project focuses on upgrading system engineering methodologies so that they can ultimately reduce the likelihood and cost of failures of AI-based systems in an unpredictable world.

Adversarial attacks can manipulate our AI models. In this project, methods to detect such attacks are being studied in order to enhance the model's resilience.

How can AI algorithms based on learning and based on reasoning effectively be combined to ensure real time Situational Awareness (SA) while executing a robust mission plan for combat unmanned ground vehicles (CUGVs) to enhance their capabilities, including the ability to disturb/degrade/destroy enemy fire support attacks?

In this work we have investigated how a deep learning model can be fooled into misdetecting large objects like an aircraft by placing a relatively small sticker on top of the object.

Using NLP to extract company values and include these values in the skill matching done through explainable AI

What is ethical and responsible artificial intelligence ("AI")? AI MAPS stands for “AI Multi‐Agency Public Safety Issues."

Introducing an autonomous AI system for service and resource management in Mobile Communication Networks.

Detecting financial crime using synthetic data.

In this project, we designed the ASCENT (Ai System use Case Explanation oNTology) framework for annotation XAI solutions.

Self-learning systems for robust optimal performance

Making autonomous public transport aware of what is required at any moment for safe travel.

This project focuses on the automatic generation of instructions for humans and robots based on 3D CAD data.

Identifying heart failure patients at high risk using Multi-Party Computation.

Integrating human knowledge and AI techniques for better diagnosis and effective predictive maintenance of high tech systems.

Working towards a multi-decision criterion platform for a variety of stakeholders to facilitate the dialogue on healthy living environment, including policy decisions for emissions.

Enabling survival analyses on privacy-sensitive distributed cancer data using Multi-Party Computation

Use explainable and responsible AI to fight cybercrime and cyberattacks.

Work towards methodology for the integrated co-design of AI-based measurement- or diagnostic systems.

DIALOG is a proof of concept of a Large Language Model-based conversational agent allowing delegate tasks to autonomous systems via natural language.

What defence AI-based systems are acceptable from an ethical, legal, and societal point of view, and which are not? ELSA Lab defense focuses on responsible design, implementation and maintenance of military AI.

Developing insights on the impact and potential of Large Language Models.

Integration of real-time situational awareness in real-life automated vehicles

To effectively fuse the plans of multiple subsystems in a navy ship, we use AI methods to intelligently explore the problem space. We created a module that facilitates the merging of sub-plans with temporal dependencies. This is done by combining all events in the sub-plans, recalculating the temporal dependencies, finding possible incoherence in the final plan and pointing out where conflicts arise. We created a multi-dimensional utility function that is optimised towards achieving the mission goals.

Tailor-made support for operators in greenhouse horticulture

Using syntethic data generation for secure learning that preserves data privacy.

The GPT-NL initiative, a collaboration between TNO, SURF, and NFI, aims to support the Large Language Model community through the development of a state-of-the-art research facility and the pursuit of a competitive yet responsibly designed language model. We believe that technology should be reciprocal, trustworthy, transparent, and ensure sovereignty of our citizens and institutions and want to reflect these values with GPT-NL.

In this project, we use text mining techniques to mine forum data and create a search system that allows users to get insight in forum posts (either Dutch or English)

The Horizon Scanner can gather information from online sources, and provide and visualize potential trends.

Integrating Human Knowledge in the form of Attention in a Deep Learning Model

Within this project AI-based building models are developed to solve the problem of peak loads in energy demands caused by renewable energy sources.

Assess the readiness of Hybrid Intelligence (HI) technologies to arrive at a human-AI system for a societally relevant use case.

Designing a decision support system for the public prosecutor.

The Learning with Less Labels (LwLL) program aims to make the process of training machine learning models more efficient by reducing the amount of labeled data required to build a model by six orders of magnitude, and by reducing the amount of data needed to adapt models to new environments from tens to hundreds of labeled examples. This program is funded by the Defense Advanced Research Projects Agency (DARPA).

In this project, we develop a methodology for the identification, formalisation and implementation of the moral values and goals relevant in various high-risk contexts. Based on this methodology, it is our hope AI systems can be equipped with adaptive moral models that represent human moral standards.

In this project the main objective is to develop new AI methodology for explaining (uncertainty in) class predictions, with applications in the context of new mobility concepts.

Understanding the privacy-sensitive aspects of speech data

Developing a methodology for verification and validation of AI-components.

We investigate how a self-assessment capability can be obtained and exploited considering a typical use case in which limited training data is available, namely target recognition for buried objects in a marine environment in low-frequency sonar imagery.

The goal of the SALVA project is to use situation awareness for planning of an autonomous shuttle, so that the shuttle itself can decide to deviate from its original path (so called virtual rails) in order to be able to handle situations such as objects blocking the road, oncoming traffic, et cetera.

Enabling Privacy-Preserving Analyses on Healthcare Data

Enabling semantic communication for autonomous greenhouse systems

The SimulateVerify.AI project explores how scarcity of large, real-world datasets can be addressed by effectively using simulated data during the testing phase of deep learning models.

Developing an adaptive prediction system for a proactive maintenance approach of bridges.

The SNOW flagship aims to develop, integrate, demonstrate and evaluate hybrid Al capabilities for an autonomous system that can operate safely and effectively in an open world.

Applying privacy-enhancing technologies to speech data so more powerful models for automatic speech recognition can be developed

Generating courses of action in military operations.

Predicting the anthropogenic causes of subsidence in the Netherlands and unravel the contributions of deep versus shallow processes.

Using syntethic data generation for secure learning that preserves data privacy.

In this project we focus on the key factors of success and failure of implementing risk assessment tools in the justice domain, and identify technical, socio-technical, and social solutions to common problems.

The purpose of DASH is to allow humans to efficiently tell an AI-based system what to do without having to spell out all details, and without losing meaningful human control over that system.

This project studies and demonstrates the stakeholder needs for explainability and the required technologies to achieve explainable AI systems that support a sustainable energy system. We currently focus on demand-side flexibility or increased flexibility in how we consume energy to better match the characteristics of the energy grid.

Improve the matching between vacancies from companies and job seekers by developing an AI demonstrator that provides vacancy scores and recommendations of skills changes.