Tilburg University department Cognitive Science and Artificial Intelligence

Research

The research of the CS&AI research program focuses on the cognitive sciences and artificial Intelligence, combining the measuring of mental and behavioral processing, the modelling of these processes in computational models, and the testing of these computational models with experiments. A considerable part of the research is based on the development and use of methods from data science. The research topics addressed by the group are wide and varied and are all related to aspects of cognitive science and artificial intelligence, including computational linguistics (language and text analytics), image recognition, affective computing (social signal processing), deep learning (neural networks), speech and voice analytics, multimodal communication, automated face and gesture analytics, decision making, cognitive modeling, virtual, mixed and augmented reality, (serious) gaming, robotics, and avatars.

Within our education – which in our view can ultimately not be seen as independent from research – the CS&AI group is responsible for the Cognitive Science & Artificial Intelligence (CS&AI) Bachelor, the CS&AI Master, as well as the university-wide Master Data Science & Society (formerly known as Data Science: Business and Governance). The latter has now been accredited nationally and the former two are in the process of being accredited nationally (the Bachelor CS&AI passed a first phase in the accreditation process in August 2018). The CS&AI group also contributes to the Eindhoven University of Technology / Tilburg University combined Bachelor Data Science and the Master Data Science & Entrepreneurship (Jheronimus Academy of Data Science or JADS). It should be pointed out that in particular the CS&AI Bachelor and Master programs demonstrate the department’s vision on research-based education. Students perform research within the projects of CS&AI, either internal projects or projects in an environment of public-private collaboration.

Research Units:

Human-AI Interaction

The Human-AI Research Unit focuses on the theories, techniques and implementations from cognitive science and cognitive psychology in their relation to the interaction between humans and artificial intelligence. Questions include (but are not limited to):

  • How can brain-computer interfacing systems be developed for adaptive human-technology interaction?
  • How can human social behavior enable artificial agents such as robots to deal effectively with social interactions?
  • How can traditional psychological measures (questionnaires) be linked to behavioral and psychophysiological data, how AI solutions can be implemented in applied psychology areas (e.g., identification procedures for students with learning issues)?
  • How can XR technologies influence perception and performance, both at a fundamental interaction level and also applied in areas such as healthcare and training?
  • How can we predict future experiences by using cognitive tasks as well as brain data to look at underlying coding principles of memory and learning?
  • How can we understand high-level cognitive processes (e.g., collaborative problem solving) during human interaction with social and technological environments in complex work domains?

Methods and Techniques

Methods and techniques of this research unit primarily come from the field of cognitive science and cognitive psychology (but extend to the field of data science and artificial intelligence) and include behavioral and neurophysiological measures, including EEG, response times techniques, virtual, mixed and augmented reality, social signal processing, and computational linguistics.

Research domains

Research domains include those generally covered by cognitive science, including but not limited to aviation, healthcare, education, and creative industries.

Applications

The research applications of the Human-AI Research Unit include improving interactions between healthcare professionals, embodied conversational agents for healthcare, improved simulators for aviation, augmented reality techniques to optimize surgical procedures, and intelligent tutoring systems for educational purposes.

Deep Learning for Perception

The research of the Deep Learning and Perception group focuses on deep learning and on models of perception. The deep learning topics covered include the analysis and classification of perceptual (visual and auditory) data. Prominent application domains are health, biodiversity, and industry. The introduction of deep learning in society requires a careful calibration of human and machine decision making. We study decision making under uncertainty in hybrid human-deep learning systems. In our fundamental research we study transparency and explainability of deep learning models, reducing the complexity of deep learning architectures, and generative models of perception.

Fundamental questions addressed in our research unit:

• How can we improve the accuracy and robustness of deep learning models?

• How can deep learning models be applied in hybrid-AI settings?

• How to balance human decision making with AI-informed decision making?

• How can overparameterisation in deep learning be avoided?

Application-oriented questions addressed in our research unit:

  • How can we combine brain-imaging data with neuropsychological assessments to predict the outcome of brain surgery?

  • To what extent can can human body-shape characteristics be reliably inferred from images?

  • To what extent can unsupervised and supervised deep learning support the authentication of artworks? How can we automatically recognise birds from their songs?

  • How can generative deep learning facilitate predictive maintenance in the energy sector?

Methods and techniques

The methods and techniques used are deep learning and (traditional) machine learning. With respect to machine learning, the methods and techniques used are: generative deep learning, variational auto- encoders, geometric deep learning algorithms, convolutional neural networks and their recent variants, such as vision transformers, and fuzzy cognitive maps.

Research domains

The main research domain is the branch of AI covering deep learning, machine learning, perception and decision making. The domain of hybrid AI, the efficient combination of artificial and natural intelligence, plays an important role in our applied projects. We investigate how the black box of deep learning can be opened. Addressing the computational and energy costs of overparameterisation in deep learning, we study the feasibility of simplifying network architectures.

Applications

Our applied research focuses on medical image analysis, e.g., segmentation and analysis of MRI im- ages (collaboration with ETZ hospital Tilburg) and wrist fracture detection (collaboration with JADS, Radboud University Medical Center and Jeroen Bosch Hospital, ’s-Hertogenbosch), human body shape estimation from images (collaboration with Zero Hunger Lab of Tilburg University), supervised and unsupervised analysis of drawings and paintings (collaboration with the Swiss company Art Recognition), birdsong recognition (collaboration with Naturalis), out-of-sample detection for smart industry (collaboration with JADS; CERTIF-AI project), generative networks for predictive maintenance to promote sustainable-energy use (collaboration with JADS; ILUSTRE project).

Computational Linguistics and Psycholinguistics

The Computational Linguistics and Psycholinguistics unit focuses on applying Machine Learning techniques to textual, audio and visual data to model human language learning and processing in theoretical and applied domains. A number of research projects attempt to improve our understanding of human cognition by building formal and computational models of human language acquisition and use. Knowledge of how humans process language is in turn used to develop better applied systems and general-purpose tools and techniques for processing large collections of linguistic and extralinguistic data.

The main research directions within this group can be divided into two broad themes of Computational Linguistics and Computational Psycholinguistics.

Computational Linguistics

  • Developing robust and general-purpose frameworks for analyzing the linguistic representations that emerge in deep models of language.
  • Modeling language learning grounded in perceptual context, specifically learning language from multimodal signals such as speech and vision.
  • Integrating linguistic features into Neural Machine Translation systems, and investigating gender bias in Natural Language Processing and Machine Translation.
  • Modeling cognitive processes that allow humans to identify words and syntactic structure by applying NLP techniques on data collected from naturalistic human interactions or from experimental studies.
  • Studying computational stylometry for inferring information from text, as well as developing methods to better understand and defend against its potential for invading the users' privacy.

Computational Psycholinguistics

  • Studying the relation between word form and lexical semantics and their impact on language learning, processing and change using large corpora.
  • Adapting novel statistical techniques for time-to-event analysis to psycholinguistic data, providing insight into the time course of language processing that is not available through more additional analysis techniques.
  • Exploring the cognitive mechanisms underlying sentence processing through formalizing and developing computational models of language processes.
  • Investigating the shared mechanisms of language and music perception such as rhythmic, syntactic and pitch processing.

Methods and Techniques

Research in the Computational Linguistics and Psycholinguistics group mainly relies on statistical machine learning and (hierarchical) Bayesian modeling, as well as advanced deep neural models for processing linguistic and visual data. 

Autonomous agents, robots, and games

At the core of the research represented by Research Unit 4, “Autonomous agents, robots, and games” is the concept of autonomous decision making, from a computational perspective, in an environment where a machine interacts with humans. Autonomous decision making entails, in this case, that machines decide upon their actions independent from and without direct interference of a human controller.

To be able to take intelligent decisions autonomously, a machine must be able to interpret the context it resides in; it must (in a rudimentary way) understand its environment and the other entities in it (whether they are humans or other computational agents), it must be able to process the supplied information, and it must be able to make a choice between all its possible action sequences so that it meets its desired goals.

A main activity of the research unit is the programming (and sometimes building) of artifacts, such as robots, simulations, and games, and to study the behavior of such artifacts in interaction with their environment.

Methods and techniques. To interpret data and information, and to allow a system to reach decisions, the research unit is not limited to a particular methodology or technique. Both symbolic and natural artificial intelligence methods and techniques are applied in research. Considering this broad basis, the foundations of artificial intelligence are also of interest to the research unit.

Research environments. Typical environments which are of interest to the research unit are autonomous agents (artificially intelligent programs which act autonomously to interpret data and undertake actions), robots (autonomous agents which act in the real world), and serious games (which may contain autonomous agents which act in a virtual world). Data and information are sometimes gathered through observations of humans, via sensor technology.

Applications. Typical application domains which are used in the research are health, education, training, entertainment, and safety and security.

Facilities

MindLabs

In 2019,  Mindlabs, the interdisciplinary research center where Minds, Media and Technology meet, was founded as a multi-partner initiative in the field of the development of learning and training applications. MindLabs is an initiative that brings together knowledge and expertise in areas related to human-centered AI, including robotics and avatars, serious games and learning, virtual and augmented reality and language and data technologies. MindLabs operates on the interface of human and artificial minds. It hosts a series of large AI-related projects, including the Tilburg University ICAI Lab “MasterMinds”. MindLabs functions complementarily to other initiatives in the province, including JADS and EAISI, and is located in a dedicated building in the Tilburg city center to house all relevant AI partners. 

Read more about MindLabs.

DAF-LAB

Experiencing Virtual Reality

The DAF Technology Lab provides high-tech facilities for students, researchers, and the business community. The combination of technology and behavioral sciences expertise offers unique possibilities for innovative teaching and research. The DAF Technology Lab consists of two spaces: the Experience Room and the Research Room.

Qualtrics accounts for TSHD staff and TSHD students

The Tilburg School of Humanities and Digital Sciences has come to an agreement with Qualtrics for a licence permitting all of its staff and students to create a Qualtrics account and hold it for as long as they are employed /registered as student at TSHD.

Find out how to create your own TSHD Qualtrics account (pdf-file).