263134 research outputs found
Sort by
Deep Learning voor mens-computercommunicatie
No full textWith the rapid advancement of artificial intelligence (AI) technologies, machines are increasingly becoming intelligent assistants that enhance productivity across various human activities. In this context, human computer communication has gained both theoretical importance and practical value. Early chatbot systems were typically based on a modular pipeline architecture, including components for information extraction, intent detection, dialogue policy, and response generation. More recently, end-to-end systems powered by large language models (LLMs) have become dominant.
Beyond functional performance, the underlying infrastructure, such as web load balancing to ensure reliable and scalable user interactions, also plays a critical role.
This thesis addresses both the model-level functionalities and the infrastructure-level challenges in human-computer communication. Specifically, it tackles:
(1) improving dialogue policy strategies,
(2) effective information extraction under data scarcity,
(3) fine-grained intent detection,
(4) enhancing LLM reasoning under non-ideal conditions, and
(5) enabling scalable load balancing under high user traffic.
We present systematic experiments exploring how advanced deep learning techniques can improve the aforementioned components, analyze limitations in LLM reasoning, and develop infrastructure-aware solutions for large-scale deployment.
First, we address the overestimation bias in reinforcement learning-based dialogue policy optimization. We propose a dynamic partial average estimator (DPAV) that computes a weighted average between predicted maximum and minimum action values, where the weights adapt dynamically based on the task. DPAV is integrated into a deep Q-network and theoretically shown to offer provable convergence and tighter bounds on estimation bias compared to existing methods.
Second, we propose a few-shot named entity recognition (NER) pipeline that improves information extraction in low-resource scenarios. The method includes a steppingstone span detector pretrained on open-domain Wikipedia data, which reduces redundant feature learning. In addition, an LLM is leveraged to generate reliable type referents, eliminating the sample dependency problem.
Third, we develop a clustering method for fine-grained intent detection.
The method uses semantic similarities in a logarithmic space to guide sample distributions in the Euclidean space and to form distinct clusters that represent fine-grained categories.
A centroid-based inference mechanism is introduced to support real-time applications.
Fourth, we introduce and empirically explore a new research direction, evaluating LLM reasoning under the summarizing inference scenario. We fine-tune three state-of-the-art LLMs and a large vision-language model using a policy gradient algorithm and evaluate them on five benchmark datasets. Despite improvements in the setting without summarizing information, model performance degrades significantly in summarizing inference, highlighting limitations in advanced reasoning. We further propose a training method that explicitly trains models to consider multiple possible outcomes during learning.
Finally, inspired by causal dynamics in ecological systems, we investigate causal dependencies among web services to improve load balancing. We propose CCMPlus, a neural module that extracts causal relationships between services and integrates seamlessly with time series models to enhance traffic prediction. We also provide theoretical explanation that the generated causal correlation matrix
captures causal relationships among services.
In summary, this thesis presents a series of contributions that advance the capabilities of deep learning models in human-computer communication. By combining reinforcement learning, few-shot learning, contrastive learning, LLM reasoning and time series analysis, our work provides both methodological innovations and practical solutions validated through extensive experimentation.status: Publishe
Leren laat sporen na: hoe onderwijs het brein en gedrag verandert in de vroege kindertijd.
No full textAt the ages 5 to 7, there are massive developmental changes in children's neurocognitive functioning and brain development. These changes also coincide with the transition to formal education in primary school, during which there is a major change in modes of instruction from learning via less-structured playful activities in kindergarten, to learning via more structured and formal direct instruction in first grade. A critical question is how much the experience of formal schooling itself, as compared with age-related maturational changes, explains these developmental changes at the levels of behaviour and brain. This question is particularly left unanswered in the field of mathematical cognition. The current proposal will significantly move the needle by using the school cut-off design, a (quasi-) experimental approach to disentangle schooling- related vs. age-related maturational influences on children's development. This will be done by comparing the degree of change in arithmetic, its cognitive predictors and its associated brain networks at two time points, separated by one year, in children who just missed or made the cut-off for school entry. Three work packages will investigate how the experience of schooling changes (1) arithmetic and its cognitive predictors, (2) brain activity during arithmetic and (3) brain structures supporting arithmetic. This will be done by contrasting three hypotheses on the effects of schooling.status: Publishe
Putting the Venice Principles into practice: strengthening the independence of ombudsmen
No full textstatus: Published onlin
Karakterisering en verbetering van beeldvormende modaliteiten bij agressieve lymfoproliferatieve aandoeningen
No full textstatus: Publishe
Hybride polyoxometalaten voor toepassingen in supramoleculaire chemie en moleculaire machines
No full textInnovations in the field of supramolecular chemistry have provided some of the most creative and efficient solutions to modern-day problems. Studying chemistry beyond the molecule allowed chemists to exploit weak intermolecular forces to tackle problems in materials science, medicine, and pharmaceuticals. Moreover, supramolecular approaches have allowed us to expand our understanding of chemistry and physics toward the controlled motion of molecules at a molecular scale. This culminated in the development of the field of artificial molecular machines, where chemical and physical inputs are used to control molecular motion, such as rotation, translation, or perform other mechanical tasks.
Despite remarkable achievements, most research in these fields is focused on organic molecules. Very rarely can we find works that incorporate inorganic moieties such as metal-oxo clusters in these frameworks, which is a missed opportunity. Metal oxo clusters could provide new and interesting properties that, in combination with supramolecular systems and molecular machines, could offer new properties not easily accessible using purely organic frameworks. This is especially true for polyoxometalates (POMs), which are anionic metal-oxo clusters with well-defined structures and interesting redox and catalytic properties. These characteristics make POMs potential functional building blocks in supramolecular architectures and molecular machines.
In this thesis, I have explored new ways of incorporating POMs into supramolecular systems by functionalizing them with organic molecules to create hybrid polyoxometalates (HPOMs). Organic molecules were chosen based on the ability to facilitate the incorporation of the HPOM into various supramolecular architectures, including responsive materials, biochemical systems, and artificial molecular machines.
Chapter 1 provides an overview of current progress in the field of responsive supramolecular systems with POMs and discusses the state-of-the-art in artificial molecular machines. In the work presented in Chapter 2, a host-guest method, based on pillar[5]arene, was developed for controlled supramolecular assembly of HPOMs. These assemblies were further applied in catalysis, highlighting their potential as functional nanoarchitectures. Derivatization with organic and bioorganic molecules was also used in Chapter 3, to achieve targeted interactions of the HPOM with an enzyme. Namely, azide, α-D-mannopyranoside, and hexyne moieties were covalently grafted onto the POM, which enabled unprecedented site-selective binding of azide and α-D-mannopyranoside functionalized POMs within the enzymatic pocket of Hen Egg White Lysozyme, a crucial antibacterial enzyme. This represents a significant finding in designing POM-based systems for bioengineering applications. Chapter 4 expands on the idea of establishing a modular approach toward creating POMs with different properties and potential applications in supramolecular chemistry. In this case, copper-catalyzed click chemistry was used to synthesize new molecules with multiple POMs covalently linked together exhibiting emergent properties not accessible through individual components. The same synthetic approach was extended to create the first example of a rotaxane-type mechanically interlocked molecule containing a POM. Such a rotaxane with incorporated POM unit could act as a scaffold for functional molecular machines that combine motion with catalysis or electron transfer. In Chapters 5 and 6, the incorporation of POMs into artificial molecular machine architectures was studied. Chapter 5 explores coupling rotary motion with POMs, which was achieved through functionalization of an HPOM with a second-generation Feringa-type motor. Chapter 6 focused on achieving controlled translation motion of a macrocycle toward POM, through photoswitching of an azobenzene attached to the POM.
This thesis establishes a synthetic and conceptual framework for integrating POMs into supramolecular systems. The versatility of POMs as modular platforms for building supramolecular systems was demonstrated by combining them with organic moieties of different character, such as host-guest and light-responsive molecules. By expanding their structural diversity with tailored organic molecules, it was possible to establish control over the supramolecular behavior of the POMs. In the five experimental chapters, the developed systems demonstrate catalytic, redox, and dynamic stimuli-responsive behavior, creating new opportunities for inorganic cluster chemistry in the broader fields of molecular machines, smart materials, and nanoscale devices.status: Publishe
Effects of a 12-week intrinsic foot muscle strengthening program (STIFF) on gait, balance and concerns about falling in physically active older adults: An assessor-blinded randomized-controlled trial
No full textBACKGROUND: Falling is a major concern in the ageing population. Strengthening the plantar intrinsic foot muscles (PIFM) may improve gait and balance in older adults and, therefore, may have potential for fall prevention. RESEARCH QUESTION: The aim of the present study is to examine the effect of a PIFM strengthening program on gait, balance and functional outcomes in older adults. METHODS: For this assessor-blinded RCT, older adults (> 65 years) with potentially increased fall risk were recruited at functional exercise classes and randomly assigned to an intervention (12-week supervised and progressive PIFM strengthening program) and a control group. The trial outcomes were between-group differences in mean change from baseline in maximum gait speed (primary outcome), balance during gait, foot and ankle biomechanics during gait and concerns about falling and within-group differences in capacity and strength of foot muscles. RESULTS: Thirty-three participants were included. No between group differences were found for change in maximum gait speed. However, the intervention group showed a larger reduction in concerns about falling. In addition, the intervention group showed increased capacity and strength of foot muscles, but this was not related to other findings. SIGNIFICANCE: This study did not show an effect of PIFM strengthening training on maximum gait speed in older adults who are involved in a functional exercise program. However, it seems to reduce concerns about falling. This advocates further research on the benefits of integrating PIFM strengthening exercises in functional exercise programs. In addition, future studies are needed to unravel the mechanism behind the reduction in concerns about falling.sponsorship: This work was supported by the Dutch Research Council through the doctoral grant for teachers (grant number 023.013.063) . (Dutch Research Council through the doctoral grant for teachers|023.013.063)status: Accepte
Synthesis of metallic uranium microspheres via electrolytic reduction in molten LiCl
No full textsponsorship: The authors acknowledge the SCK CEN Academy for providing the funding for a PhD fellowship. (SCK CEN Academy)status: Accepte
π-DON: Physics-Informed Deep Operator Network for Control-Oriented Modeling of Thermal Systems in Cluster of Buildings
No full textsponsorship: Fonds Wetenschappelijk Onderzoek|1S66625N, KU Leuven|C24M/21/021status: Accepte
Hoe jij en ik samen vorm krijgen. Observaties van moeder-kind interacties in de context van autisme
No full textChildren develop in close interaction with their environments. During the first years of life, important environmental aspects are a child's interactions with their parents. To better understand the development of young children at elevated likelihood (EL) of autism, this doctoral project aimed to shed more light on how interactions between these children at EL of autism and their parents take shape. This doctoral project is part of the TIARA study (Tracking Infants at Raised likelihood of Autism), which tracked children at EL of autism from the ages of 5 to 36 months. Specifically, we investigated children with an older autistic sibling and children who were born very preterm. To investigate PCIs, we applied the theoretical framework of the bioecological model, as described by Bronfenbrenner and Morris (1998). Although this theory was already formulated in the previous century, and has been widely accepted, the basic principles posed by this model have been little reflected in autism research.
The bioecological model states that parent-child interactions (PCIs) are shaped by the context wherein they take place. Yet, in autism research, PCIs are almost exclusively studied during a free play context. In manuscript 1, we compared PCIs during free play and goal-directed play when children were 24 months. We noted that mothers showed more sensitive-responsive and positive behaviors, and less negative behaviors during free play compared to goal-directed play contexts. Furthermore, our results indicated that the relationship between mothers' and children's negativity only became apparent during goal-directed play, but not free play.
The bioecological model also emphasizes bidirectionality, meaning that both interaction partners mutually shape each other. In the autism field, the potential impact of parents' behaviors on child development has been rigorously studied, but little is known about the potential impact of child characteristics parents' PCI-behaviors. In manuscript 2, we investigated the latter. We zoomed in on a specific type of parenting behaviors: the degree to which parents synchronize their verbalizations to match with the child's focus of attention. Previous studies indicated that this parenting behavior is beneficial for children's subsequent language development. Our results indicated that children's early language abilities (10 months) were predictive for parents' subsequent synchronized verbalizations (14 months). Other child characteristics, such as early cognitive abilities or autism diagnosis, did not add predictive value beyond language abilities.
Thirdly, this model emphasizes that PCIs change throughout time, and encourages investigating PCIs using longitudinal designs. In manuscript 3, we observed PCIs at the ages of 5, 10, 14, 24, and 36 months. Based on trajectories of PCI-behaviors, we identified three subgroups of mother-child dyads: (a) dyads wherein children showed the most social-communicative behaviors, and mothers showed the most supportive behaviors across the first three years of life, (b) dyads wherein children initially showed the least social-communicative behaviors, and parents the least supportive behaviors, but both increased throughout time, and (c) dyads who showed notably more negativity in interactions, and wherein mothers showed the least supportive behaviors. Our results confirmed that PCIs change throughout time, and indicate that the course of PCI-trajectories relate to children's social-communicative behaviors and negative affectivity. The course of PCI-trajectories did not relate to children's research diagnoses of autism or parent-rated signs of internalizing or externalizing distress.
This doctoral project suggests that the principles of the bioecological model may apply to parent-child interactions in light of emerging autism, and encourages further investigation. In this thesis, we share critical reflections on our work and discuss clinical implications that can be drawn from this project.status: Publishe