2 PhD-studenten in Neuro-AI of de Ontwikkeling van Visie (m/v/x)

The Curriculum of Sight - How Visual Processing Develops in Brains and Machines.

For animals, it is imperative to learn as quickly as possible. To do this, we know that the mammalian visual system utilizes "inductive biases"—inherent assumptions that streamline learning for specific environments. In this project, we aim to investigate how rapid learning in mammals is facilitated by the learning curriculum shaped in the developmental stages of human infants. By comprehensively measuring and documenting these developmental stages and the associated inductive biases, we aim to identify key mechanisms that drive rapid learning in the visual system. The goal is to create a robust mechanistic neural network model of the visual system that not only mimics its processing capabilities but also its adaptability, leveraging early developmental data and ecological validity. Infant data will be acquired in collaboration with Dr. Tessa Dekker (University College London) and Dr. Ingmar Visser at UvA. The project thus aims to quantify and model the remarkable learning efficiency of the human visual system.

The project is an interdisciplinary collaboration between the the Machine Learning group at CWI in Amsterdam (Prof.dr Sander Bohte) and the Brain & Cognition group (Dr. Steven Scholte) at the UvA, where the PhD at UvA will focus more on quantifiying and modeling developmental stages, and the PhD at CWI on developing detailed neural models.

PhD student (m/f/x) on the subject of Fast and low-dose medical imaging

The project is part of the IMAGINE open innovation lab (https://research.umcutrecht.nl/news/imagine-takes-off/), and aims to develop novel computational methods for faster medical imaging with reduced radiation dose. The focus will be on (cone beam) CT and its use in image-guided interventions in oncology. One of the main challenges is to deal with high noise levels (due to the low radiation dose). One possible way to address this is to incorporate information gained from previous scans (e.g., MRI) of the same patient.

As a PhD student on this project, you will:

• Design and conduct research on computational methods for CT reconstruction in the context of image-guided interventions, focusing on improving quality of cancer treatment for patients as well as reducing workload for clinicians. Collaborate with experts in AI, data science, and medical imaging to develop integrated solutions
• Engage in translational research, bridging the gap between technological innovation and clinical application
• Contribute to publications and presentations to disseminate findings within the scientific community
• Participate in interdisciplinary meetings and workshops to foster knowledge exchange and innovation

Your work is vital to advancing less invasive treatment options, reducing patient recovery times, and optimizing healthcare resources. Your day-to-day research will take place in the computational imaging group at CWI, where you will have a chance to interact and collaborate with fellow PhD-students and research staff with a background in mathematics, computer science, and imaging science. It is expected that you will frequently interact with the other members of the IMAGINE open innovation lab (situated in Utrecht), as well as contribute to teaching at the Mathematical Institute at Utrecht University.

PhD position in Fundamental Techniques in Table Representation Learning

Goal of the Table Representation Learning (TRL) Lab

Approximately 120 zettabytes of data has been collected worldwide but less than 1% is actually used. Structured data as found, for example, in tables, spreadsheets, and relational databases, is prevailing in organizations and typically informs important decisions in governments and humanitarian organizations, healthcare and finance. Yet, while AI has demonstrated a high impact on applications on text and images, proportional progress on tabular data is lacking. With the TRL Lab (Table Representation Learning Lab), we aim to close this gap, by developing AI models and tools for tabular data, to help organizations, of any size, domain, and level of data literacy, get insights from structured data, efficiently, accurately and securely.

Goal of this PhD project
High-capacity neural models, such as transformers, have been pivotal for establishing general-purpose models for a wide variety of natural language tasks. Despite successful adaptations for structured data, our research has identified shortcomings for fundamental properties of tabular data. This research position will focus on exploring fundamental techniques for tabular-native models. This can involve, for example, studying new TRL model architectures, serialization and tokenization techniques, among others. A strong interest and background in AI and/or NLP are desired.

What you will be doing

• Inform a research agenda on the PhD topic for a timespan of four years.
• Develop fundamental AI techniques, new TRL models, and systems specific for tabular data.
• Publish reusable software and data artifacts where relevant.
• Communicate research outcomes through papers and talks at conferences, workshops, and beyond.
• Actively collaborate with other researchers in the TRL Lab (students, 4-5 PhDs, postdocs, PI) and external collaborators (e.g. University of Amsterdam, the UN, and Amsterdam UMC).
• Assist in relevant teaching activities at universities, such as thesis supervision and assisting in courses.

Postdoc (m/f/x) to develop computational models for electric breakdown of gases

Electric gas discharges occur in nature, most prominently in air in the form of lightning and its less visible precursors. Similar gas discharges occur in high-voltage equipment, such as the switchgear used in electric grids. Our transition to sustainable electric energy requires major extensions of this grid. However, switchgear and other high-voltage equipment currently often operate with SF6 gas, which is the worst greenhouse gas known. Within the project, we investigate electric discharges in “green” alternatives to SF6. For further information, see https://homepages.cwi.nl/~ebert/2023-ERCIM-News.pdf

We are seeking a postdoc to develop computational models of electrical breakdown of gases. Breakdown occurs in several stages: streamer discharges form, some of them can turn into leaders due to gas heating, eventually resulting in a spark. Simulating all these stages in detail is computationally unfeasible, so the goal is to develop a simplified model focusing on the leader stage. You will:

• Analyze experimental data and microscopic simulations
• Identify relevant physical features and parameters
• Apply machine learning techniques to build predictive models
• Collaborate closely with experimentalists and modelling experts

Project Environment
This position is part of a collaborative research project involving:

• Two PhD students at TU Eindhoven, conducting experiments
• A PhD student at CWI, focused on modelling streamer discharges
• A committee of industrial users, ranging from start-ups to multinationals, providing feedback several times per year

Our modelling group at CWI Amsterdam has extensive expertise in modelling the dynamics of gas discharges. We have developed several microscopic models for the early stages of electric discharges as well as analytical approximations to grasp their multiple spatial and temporal scales. Experimental partners at TU Eindhoven include the group of Sander Nijdam, an expert in measuring streamer discharges with highest resolution, and the group of Tom Huiskamp, who focusses on new solid state high voltage sources and their applications.

PhD student in AI methods for integrated hydrogen-electricity markets (m/f/x)

Our energy systems are undergoing rapid changes, with increasing adoption of distributed renewable generation (from PVs and wind turbines), new forms of demand (from EV charging, heating) and storage. This poses significant challenges for our power networks, as seen in the significant network congestion issues emerging in many regions of the Netherlands. In recent years, hydrogen has emerged as a significant energy vector, with potential to address many of the challenges posed by the energy transition. The Netherlands national hydrogen research programme on hydrogen, Groenvermogen NL (GVNL, https://groenvermogennl.org/) was established to perform fundamental and practical research required for realizing and accelerating the hydrogen economy.

WP7 of Groenvermogen (called Hy-SuCCESS: Social, User aCCeptable, Economically Sustainable Systems for hydrogen) studies the development of hydrogen systems from a socio-technical perspective. It considers both the economic and business cases of hydrogen systems, but also the system integration aspects, legal aspects and the user acceptability of hydrogen solutions in different application areas. It is based on a highly interdisciplinary research consortium, that brings together economists, business school scientists, system modeling and optimization researchers, computer scientists, legal experts and social scientists working on energy topics.

Description of the PhD project
The project of the PhD student based at CWI in Amsterdam will study integrated hydrogen-electricity markets. In particular techniques from Artificial Intelligence and multi-agent systems for modelling new types of markets are particularly relevant for this position. From an application perspective, in most application areas where hydrogen is deployed, it has to inter-operate with other energy carriers and consumers, in particular electricity, transport, gas, heating etc. For example, a hydrogen based-fuel cell or electrolyser will have to supply a number of services, ranging from supplying local transport or industry needs, to acting as a source of storage and flexibility for the power networks. Here, flexibility is the capability to change some individual energy supply or demand in time, size, or location: this can relieve power network congestion, which is an increasingly important problem in the Netherlands. From a market-based perspective, this means hydrogen assets will have to participate simultaneously in a number of markets, including local hydrogen and electricity markets, but also flexibility markets related for managing network congestion.

Some specific topics that are relevant for this PhD position include (non-exclusive list):

• Automated bidding strategies for participating in multiple markets (hydrogen, electricity) over different time scales, such as short-term forward markets and spot markets (operating in real-time). This can be achieved through a variety of optimization, machine learning or AI-based heuristics.
• Optimization of revenue stacking models for hydrogen assets that have to supply a number of market-based services simultaneously (e.g. deliver hydrogen for transport, or act as a form of energy storage for electricity system)
• Design of new types of short-term electricity markets, that incentivize the participation of hydrogen assets, in combination with other types of power system assets (batteries, etc.). From a research perspective, this can involve techniques from algorithmic game theory and AI-based mechanism design.
• Validation of new types of markets (both those designed above and others) through principled multi-agent simulations, complex systems analysis or other data-driven simulation methods.

Fundamental AI techniques relevant to this project include a wide range of computational intelligence and ML techniques, distributed, multi-agent optimization, market-based coordination between agents, automated negotiation and algorithmic game theory approaches. In terms of the optimization aspects, relevant topics include: single and multi-criteria optimization, dynamic constraints, uncertainty and risks, costs of actions, and economic/technical trade-offs.

The PhD student will start with a survey identifying promising alternatives and then delve deeper into specific solutions and approaches and real case studies and data. The projects will involve regular discussion and presentation of the results to other consortium members from other universities, HBOs and companies. Given this is a large, interdisciplinary project, partners may have potentially, different expertise backgrounds. Hence, effective communication skills are required.