The Vrije Universiteit Brussels has several jobs available as part of a larger project to stimulate AI in Flanders. These jobs are on the PhD and on the Postdoc level, are fully funded and have a small budget for travel and equipment. More details of each job can be found by clicking on the links below. More information about working at the VUB and in Brussels can be found here.
How to apply
Fill out this form in which you upload a letter of motivation, a CV and other requested documents (specified in the descriptions of the individual jobs). Please express in your letter of motivation which job(s) you are interested in. You can express interest in multiple jobs. Positions will be filled as soon as a suitable candidate is found. The starting date is negotiable, but preferably no later than September 2019.
The AI Flanders project is an overarching project involving all Flemish universities and several research institutes, aiming to stimulate AI research in Flanders, both on the academic and industrial level. The project consists of four grand challenges, of which the VUB is involved in two, namely multi-agent and collaborative AI and human-like artificial intelligence.
Multi-agent and collaborative AI
The goal of this grand challenge is to research how multiple AI agents, each with their own goals can collaborate to reach a joint desired state. The overall long-term goal of this challenge is to develop multi-agent AI systems that are able to cope with open-ended problems, where new (kinds of) agents enter the system, with new capabilities and goals, and to integrate them seamlessly, keeping the system viable. Aspects that are investigated in this challenge include joint multi-agent online learning (how to learn efficiently or transfer knowledge, while respecting constraints and privacy), robustness in dynamic multi-agent systems (how to update the system when agents join or leave; how to avoid agents gaming the system), strategic decision making (how to recognize, reason and anticipate on the decision making of other agents), collaborative decentralized query execution (how to cooperatively solve queries in privacy-sensitive environments), and information gathering and sharing (how to learn and share new information in a MAS without compromising privacy). There are several proofs of concept foreseen for the research in this challenge: machine fleet control, assembly, and healthcare.
Human-like artificial intelligence
The goal of this Challenge is to design and build autonomous, intelligent, trustworthy entities that communicate and collaborate seamlessly with humans in natural and complex environments. This entails communication in ways that are natural for humans, such as natural language, but also the ability to provide multi-step human-like reasoning by perceiving and understanding the complex environment. This will allow to enhance society and workplace with artificial entities that can clearly understand the environment they are interacting with, feature the same level of adaptivity to unseen tasks as humans, while appropriately interpreting the social and physical environment, and involving, informing and supporting human colleagues.
This focus on complex reasoning in an interaction between man, machine and its environment, as human intelligence and physical capabilities work in harmony with machines via intuitive and social interaction, has strong potential in domains such as industry, mobility & society. Increasing the reasoning and understanding capabilities of AI systems and facilitating seamless interaction with intelligent technology has both short term application domains in more predictable environments and allows – in the long term – to achieve human-like intelligence.
List of available jobs
Multi-agent and distributed AI
- Distributed data intelligence: one position (preferably filled by a PhD student, but good PostDocs will also be considered) on representation of, and reasoning with distributed data, ranging from the development of languages for agents to declare their data capabilities to algorithms that combine data from different agents. (contact: Prof. Bart Bogaerts)
- Hybrid MAS: A PhD position and a short-term visiting postdoc position in modelling and experimentally investigating how collective behavior emerges can emerge in populations of agents that do not all share the same interests (and potentially with both human and artificial agents). (contact: Prof. Bart De Boer)
- Fleet Control: A PhD position and a short-term visiting postdoc position in learning control in a fleet of devices which are the same but not exactly the same. The key research question is who should learn or copy the behaviour of whom when? Application are industrial production machines, wind mils, hybrid engines, etc. (contact: Prof. Ann Nowé)
- Co-operating agents and robots: A PhD position and a postdoc position (short-term or long-term) for developing novel methods for incorporating (approximate) domain knowledge in co-operative multi-agent learning systems. In particular we want to extend methods for single agent reinforcement learning that incorporate domain knowledge into the learning process to multi-agent systems in a co-operative setting. (contact: Prof. Bram Vanderborght)
- Formal verification of Multi-agent systems: A PhD position and a short-term visiting postdoc position for developing novel approaches to formally verify multi-agent learning systems. More precisely, we are interested in how a policy obtained trough learning can be formally verified using formal verification techniques such as ATL, LTL, etc. Experience in either learning systems (single or multi-agent) or formal verification is highly recommended. (contact: Prof. Ann Nowé)
Human-like Artificial Intelligence
- Acquiring domain knowledge through natural language dialogue (PhD position). The selected candidate will work on a project that investigates how gaps in domain knowledge of either a human or an intelligent system can be identified, and filled through natural language dialogue. For doing so, he or she will need to combine symbolic techniques from computational construction grammar and dialogue modelling, with the goal of building a conversational agent that (i) can interact naturally on both the grammatical and discourse level, (ii) reason about the knowledge that needs to be acquired either by the human or by the agent and (iii) integrate the acquired knowledge into its knowledge base for later reuse. (contact: Prof. Katrien Beuls)
- Sequential Hierarchical Learning Systems (A PhD position and a postdoc position) Sequential data often obeys a hierarchical structure either hidden in the semantics such as in language or explicit such as in mathematical expressions. When defining an action, also granularity within the hierarchy comes into the picture. Here, we develop new hierarchical statistical learning methods to enable grounded learning systems, that are capable of explanations themselves, in several directions. The temporal associations learned in this process, diachronic and synchronic, will be used to form a rich world model, which is capable of formulating its own representations in a human-like way. The architecture affords abstraction, and we will extend the reasoning system to include analogy and/or metaphor and thus allow transfer learning. (contact: Prof. Geraint Wiggins)
- Explainable constraint solving (PhD position or post-doc). The goal is to investigate why a solution to a Constraint Satisfaction Problem is optimal or satisfiable, in terms of high level constraints or counter-factual solutions. Familiarity with constraint solving (CP, SAT, SMT or similar) required. (contact: Prof. Tias Guns)