黑料福利网

How AI shapes the workplace

A research project by Pascale Le Blanc, Human Performance Management Group, department of IE&IS

# The Human Performance Management Group studies human-centred AI and robotics.

# Research focuses on design choices, human鈥搑obot collaboration, and wellbeing.

# Computational models simulate optimal human鈥揂I collaboration for organizations.

# Companies are invited to participate in the research.

New technologies such as artificial intelligence and robotics increasingly impact many aspects of our work. In the Human Performance Management Group, Pascale Le Blanc and her colleagues focus on how organizations and their employees implement, adapt to and collaborate with these types of new technologies on the individual, team, and organizational level.

Le Blanc鈥檚 group looks at a variety of aspects, starting from the design phase. Researchers. for example, explore how AI and robots can be designed and implemented to fit real workplace practices and match employees鈥� needs and skills. The focus is on sociotechnical integration, ensuring that these technologies enhance both organizational performance and employee well-being.

From individual employees
On the level of individual employees, the Human Performance Management Group investigates how an effective and human-centered human-AI collaboration can be realized, and how such a collaboration augments  employee functioning. They, for example, look into how employees adapt to working with robots in industrial, service and health-care environments, with a focus on the psychological, behavioral, and work design implications of human-robot collaboration.

The group also researches potential synergies between AI and humans in teams. To this end, they for example look into common key requirements for human-human and human-technology collaboration and the differences in dynamics between human-human teams and human-AI teams. In addition, the researchers are generating a computational model to virtually simulate what rules lead to the emergence of teamwork when one of the team members is an AI.

To the entire organization
On the organizational level, there is ongoing work on how employees interact with AI decision-support systems and generative AI, and how this affects their well-being and performance.

Overall, the work of Le Blanc鈥檚 group provides insights into how new technologies like AI can provide added value in the workplace. From their work it is clear that these technologies have to fit the social context in which they are used. This requires taking human factors into account in the design and implementation of technological innovations, as well as enabling people to properly and satisfactorily use these new technologies in their working lives.

More information about the research group

Toward faster construction with less concrete
A research project by Rob Wolfs, unit of Structural Engineering and Design, department of the Built Environment

# Rob Wolfs develops a multi-scale approach to control 3D-printed concrete behavior.

# Numerical models, machine learning, and real-time sensor data are integrated.

# Continuous monitoring ensures print quality during construction.

# Adaptive feedback loops and reinforcement learning reduce defects and improve sustainability.

In the 3D Concrete Printing group, Rob Wolfs is developing a novel multi-scale approach to predict the behavior of 3D-printed  concrete structures with the aim to enable faster construction with less material.

3D-Printing of buildings can provide an answer to the housing shortage and the sustainability challenge in the construction industry. Unfortunately, the behavior of printed materials is more complex than that of traditional materials. To control the quality during manufacturing in real-time, Rob Wolfs addresses the complexity of 3D concrete printing across various time- and length scales through the use of numerical modelling, machine learning and experiments.

Minimize defects
In the CTRL+P project, for example, Reinforcement Learning is used to minimize defects during 3D printing and optimize the printing strategy. Optimizing the print head鈥檚 sequence and movement helps control thermal, moisture or material age gradients more effectively. This way, the researchers enhance efficiency by reducing the need for manual adjustments and experimental fine-tuning, while being adaptable to different printing methods.

Adapting on the spot
Another project Wolfs is working on aims to enhance the 黑料福利网 3D printer system with in-line and on-line measurement systems as a starting point for process optimization. By implementing monitoring technology like laser scanners, near-infrared scanners, and thermal sensors, large amounts of data are collected in a non-destructive way during 3D-printing. Through machine learning approaches these data are used to predict performance of the final printed objects. The aim is to improve the quality of printed objects by establishing adaptive feedback loops that enable continuous tuning of printing parameters to counteract feature fluctuations during production.

More information about the research group.
 

Getting value from advanced data analytics
Oktay T眉retken

# Technical expertise alone does not create AI impact.

# Many organizations remain stuck in pilot projects without scaling.

# Real value requires aligned capabilities: strategy,
processes, data architecture, and skilled teams.

# The CMM-ADA model helps assess maturity and build a focused growth roadmap.

How can organizations leverage advanced data analytics, including AI? What organizational capabilities do they require to translate data into strategic, tactical, and operational value? That is one of the primary questions in the research of Oktay T眉retken from the Information Systems group.

Firms often possess multiple databases containing terabytes of structured and unstructured data. By applying data analytics techniques, including AI, to analyze these vast data assets, organizations could gain a competitive advantage in their respective markets. However, many firms continue to struggle to translate their analytics efforts into tangible business value.  

In practice, firms often rely on high technical talent, which alone is insufficient. Many analytics initiatives remain at the pilot stage, resulting in models that are built but not used, as well as fragmented roles, practices, skills, and strategic alignment. The value an organization can derive from advanced data analytics depends on its ability to develop and apply a coherent set of capabilities. These capabilities extend beyond technical expertise and include the ability to define a data analytics and AI strategy, manage data architecture, establish relevant processes, and assemble the right team. Managers, therefore, need guidance on how to systematically develop and improve these capabilities. Without a clearly defined roadmap, organizations are likely to struggle with prioritization and achieving consensus across stakeholders.

Assessing capabilities
T眉retken and his colleagues, Ginger Korsten, Banu Aysolmaz, and Baris Ozkan, have developed the Capability Maturity Model for Advanced Data Analytics (CMM-ADA) to provide a structured foundation for organizations to assess and enhance their advanced data analytics and AI capabilities in support of broader digitalization goals. The model has been developed through close collaboration with many industry experts and academic researchers. It has been validated via industry-wide surveys and multiple organizational assessments conducted in companies across the Netherlands.

Organizations can use the model to self-assess their advanced data analytics and AI capabilities, identify where they excel, see areas that need improvement, and create a roadmap for improving these capabilities.

Open to collaboration
The research group continues to seek partnerships with organizations interested in applying the CMM-ADA. These collaborations not only support organizations in advancing their analytics maturity, but also enable the researchers to refine the model and its components, and to validate its effectiveness in diverse organizational contexts.

Early cancer detection with AI

A research project by Fons van der Sommen, Architectures for Reliable Image Analysis, department of Electrical Engineering

# 黑料福利网 researchers developed SPECTRE, an AI model for early detection of abnormalities in CT scans.

# Trained on ~250,000 public CT scans, it serves as a strong medical foundation model.

# SPECTRE outperforms existing models without additional training and is open source.

# First deployment of 黑料福利网鈥檚 new supercomputer SPIKE-1, enabling large-scale scan processing.

黑料福利网 researchers developed SPECTRE: a medical AI model that helps doctors identify a wide range of abnormalities in CT scans at an early stage, enabling faster diagnosis of cancer and other diseases.

In computer vision, and especially in healthcare, AI models are mostly trained on small, in-house datasets for a specific task. Researchers from the Department of Electrical Engineering have taken a novel approach. They used some 250,000 CT-scans with accompanying radiology reports and electronic health records from public medical databases to train a foundation model.

The resulting model SPECTRE is a baseline system upon which other, more specialized AI applications can be built that can, for example, automatically recognize organs for guidance during operations, predict disease progression, or detect early onset tumors or other abnormal patterns in medical imaging data.

黑料福利网 supercomputer
The project marked the first practical deployment of the SPIKE-1 compute cluster that 黑料福利网 is leasing from NVIDIA. This cluster operates in a sustainable data center in Finland, and is able to process hundreds of CT-scans simultaneously.

SPECTRE outperforms existing segmentation models on several organ and tumor segmentation benchmarks, and shows state-of-the-art performance compared to other foundation models on tumor biomarker classification, without any further task-specific finetuning.

The model will be available open source, enabling hospitals and research institutions to develop their own tailored variants.

Autonomous intelligence for materials discovery

A research project by Sof铆a Calero, Materials Simulation & Modelling, department of Applied Physics

# The Materials Simulation and Modelling group develops data-driven, self-learning materials.

# By combining molecular simulations, statistical physics, and AI, they predict and optimize material behavior.

# Explainable AI and continuous feedback between simulations and experiments enhance discovery.

# The approach accelerates sustainable innovations, including energy conversion and CO鈧� capture.

Designing materials that can learn from data and guide their own discovery lies at the core of the research at Sof铆a Calero鈥檚 Materials Simulation and Modelling group. The group aims to create a form of materials science that not only predicts behavior but also learns from it, evolving through continuous interaction between models, data, and physical principles.

The Materials Simulation and Modelling group combines multiscale molecular simulation, statistical physics, and artificial intelligence to develop frameworks that enable autonomous reasoning across scales, linking atomistic detail to functional performance.

From simulation to autonomous discovery
Traditional materials design relies on trial and error, limited by the complexity of the chemical space and the cost of experiments. Calero鈥檚 group introduces self-improving systems capable of learning from both simulations and experiments to identify, evaluate, and optimise materials without human intervention. These systems use physics-informed learning to understand how molecular structure determines behaviour, accelerating the path from concept to functionality.

Building explainable intelligence
At the heart of this vision is the creation of digital frameworks that connect simulation, machine learning, and experimental validation in a continuous feedback loop. These tools allow materials to explain themselves, linking data, models, and theory into an interpretable and trustworthy scientific process.

A new paradigm for materials science
Through this integrated approach, Calero鈥檚 group aims to establish a new paradigm of autonomous scientific discovery. The research provides predictive tools for sustainable technologies in energy conversion, carbon capture, and environmental processes, laying the foundations for a future in which science and computation evolve together.

More information about the research group
 

Gaining clinical insights from biomedical signals

A research project by Aaqib Saeed, Computational Design Systems cluster, department of Industrial Design

# Aaqib Saeed develops AI that learns directly from biomedical signals to support clinical decision-making.

# By combining self-supervised learning with large language models, his team interprets heart sounds  and ECG data and answers open diagnostic questions.

# Tools such as CareAQA and the CareSound benchmark advance scalable, data-driven diagnostics for real-world clinical use.

Using artificial intelligence as a reliable doctor鈥檚 assistant. That is the ultimate goal for Aaqib Saeed from the Department of Industrial Design. With his group, he is developing intelligent systems that can interpret and answer nuanced questions about medical data.

Aaqib鈥檚 research is dedicated to creating AI that can interpret complex biomedical signals and engage in diagnostic reasoning about them with clinicians. He develops novel self-supervised learning frameworks that allow AI to learn directly from raw, unlabeled signal data, bypassing the need for manual expert annotation. This creates scalable, robust, and adaptable foundation models ready for real-world clinical challenges.

Reasoning stethoscope
In one of the projects, Aaqib and PhD student Tsai-Ning Wang digitized the classic stethoscope: they built an AI that listens to heart and lung sounds and enables symptom classification, disease diagnosis, and diagnostic reasoning with a large language model.

The model, called CareAQA, has proven to generate accurate and clinically relevant open-ended diagnostic answers. The researchers effectively combined audio understanding capabilities with the reasoning power of large language models to deal with differences in terms used to classify sounds like wheezes or heart murmurs, which can be early indicators of conditions like heart failure, asthma, or chronic obstructive pulmonary disease (COPD).

A major contribution of this project was the development and public release of CareSound, the first large-scale benchmark dataset designed to advance open-ended question-answering in health diagnostics.

Automated ECG interpretation
In a second line of research, Aaqib and PhD student Jialu Tang are  developing an AI that interprets a patient鈥檚 ECG and answers complex, free-form questions in natural language, such as 鈥業s there evidence of arrhythmia?鈥�. The Electrocardiogram-Language Model integrates a powerful ECG signal encoder with a large language model. The resulting combined model excels at learning, enabling it to accurately answer questions about cardiac conditions even with very few examples.

The researcher is actively seeking collaborations with clinical partners to translate these two research breakthroughs into tools that can empower clinicians and improve patient outcomes.

BayesBrain: Hybrid Brain-on-Chip AI Computing

A research project by Bert de Vries, Signal Processing Systems at the department of Electrical Engineering and Regina L眉ttge, Neuro-Nanoscale Engineering at the department of Mechanical Engineering

# BayesBrain develops a hybrid brain-on-chip AI computer that combines biological neural tissue with silicon systems.

# A Bayesian control agent and living neural circuits learn together under the Free Energy Principle.

# The project demonstrates ultra-low-power, next-generation AI hardware for real-world applications.

This project aims to develop the world鈥檚 first hybrid neuro-in-silico Artificial Intelligence (AI) computer, which combines biological brain tissue and silicon-based systems. The use of biological tissue in computing consumes much less power than silicon-based systems, making it an attractive option for AI computing. The project will create a hybrid learning system, incorporating an in-silico Bayesian control agent (BCA) with neural tissue in a microfluidic Brain-on-Chip (BoC). The system will be capable of solving real-world AI problems.

The Free Energy Principle, a leading neuroscientific theory, will govern all computation and communication inside and between the BCA and BoC. This principle supports a variational Bayesian machine learning interpretation. Initially, a pure silicon-based BCA will be developed to learn to balance an inverted pendulum by minimizing free energy on a factor graph.

Next, biological neural circuits of a microfluidic multi-compartment BoC device will replace progressively larger parts of the factor graph. The synthetic Bayesian agent will train the biological network by electrical stimulation. The two units will communicate through a novel protocol using existing software such as MEAViewer, CALIMA, NetCal, and SpikeHunter. 

The project aims to provide a proof-of-concept for ultra-low power hybrid brain-on-chip AI computing by upscaling the number of replaced sub-circuits. The project aligns with multiple EAISI program lines, particularly 鈥淐omputational AI Hardware and Software鈥� and 鈥淢erging Models and Data in AI.鈥�

Physics-Driven AI for Materials Discovery

A research project by Shuxia Tao, Intelligent Materials Theory, department of Applied Physics

# The Intelligent Materials Theory group combines first-principles physics with AI to accelerate materials discovery.

# By uniting quantum simulations and machine learning, they predict material behavior from atomic scale to device performance.

# This physics-driven AI enables the rational design of advanced materials for energy and quantum technologies.

Accelerating materials discovery by unifying materials theory and AI

Shuxia Tao, Intelligent Materials Theory, Department of Applied Physics

The discovery of new materials underpins many technological breakthroughs, from sustainable energy to quantum information technologies. To accelerate this process, the Intelligent Materials Theory (IMT) group, led by Shuxia Tao, integrates first-principles materials theory, physics, and artificial intelligence to predict the properties of materials before they are synthesized.

Because elements in the periodic table can be combined in virtually infinite ways, exploring all possible materials in experiments is infeasible. Predictive models that can reliably anticipate how new materials will behave under realistic conditions are therefore essential. Such models enable the rational design of semiconductors for solar cells, quantum materials for communication and computing, and functional materials for energy conversion and storage.

Material behavior emerges from a complex interplay of chemistry and physics across multiple length and time scales. Designing materials with targeted properties 鈥� such as efficient light-to-electricity conversion 鈥� requires understanding of how atomic-scale interactions propagate to mesoscale structures and ultimately determine device performance. Addressing this multiscale complexity is one of the central challenges in modern materials science.

Unlocking the predictive power of physics with AI
The IMT group combines quantum-mechanical electronic-structure theory with machine learning to model how photons, electrons, and ions interact to control energy and information conversion, storage, and transport. Using state-of-the-art first-principles methods, the group establishes rigorous links between atomic structure, chemical and physical response, and macroscopic functionality.

Machine learning plays a crucial enabling role: it extends the reach of quantum simulations to larger systems, longer timescales, and more complex materials, while maintaining physical fidelity. By learning from high-quality theoretical data, AI models accelerate predictions and strengthen the connection between theory and experiment.

Importantly, in the IMT vision, AI does not replace physics. Instead, it amplifies the predictive power of fundamental theory, allowing physically grounded models to operate at the scales relevant for real materials and devices.

Results and impact
The group has been able to address materials and phenomena that are difficult to access with conventional simulations, improving both predictive accuracy and scalability. Their integrated theory-AI approach has, for example, enabled the predictive modelling and rational design of and , with direct relevance to energy conversion and quantum technologies.