Last application date Jun 16, 2021 00:00
Department TW06 - Department of Electronics and Information Systems
Contract Limited duration
Degree Master of science in one of the following fields: Computational neuroscience, Artificial Intelligence (or similar), Computer Science (Engineering) - related degrees can be considered if the necessary background knowledge (described below) is present
Occupancy rate 100%
Vacancy type Research staff
PhD position – Biologically inspired feature learning in neural networks
The European Marie Curie training network SmartNets () brings together researchers from various fields, all studying biological or biologically inspired networks or graphs. The common goal of these partners is to study the interaction between the network topology, the way information is transformed in the network, and the dynamic changes in the network. One sub-topic in this project revolves around information processing in biological neural networks.
As is commonly known, biological brains are hugely more energy and data-efficient than artificial neural networks, meaning they need orders of magnitude less power to perform their computations and far fewer supervised examples to learn. As biological neuroscience collects more information about what exactly happens in living brains, computational neuroscience abstracts this information into models that can be used to evaluate the impact of local neuronal and synaptic effects on the scale of large networks. However, the connections between insights from computational neuroscience and artificial neural networks have long been lacking. Recently, there has been a surge in research that tries to bridge both worlds. However, this is far from evident, because both biological neural networks are optimized for biological substrates. Simply transferring everything we know about biological neural networks to the silicon-based substrates used to build computers (e.g., GPUs) does not make sense, from an efficiency perspective. In addition, biological ‘sensors’ are also highly optimized and perform a lot of preprocessing, such that the extracted information is already offered to the brain in a way that makes it easier to process.
The AIRO-team of IDLab at Ghent University is an engineering lab with a focus on AI and Robotics with a tendency for blue sky approaches and a very solid basis in state-of-the-art deep learning techniques.
Through our involvement in SmartNets, we want to identify elements of biological neural learning that can improve artificial neural networks, with respect to data efficiency and/or hardware efficiency. This includes new and biologically inspired approaches to feature learning but also a better match between sensors and neural processing.
One of the most prominent application domains of AI is computer vision. In computer systems, the input comes from camera’s, which generate frames of increasingly high resolution. A huge amount of processing power goes into extracting relevant features from extremely high numbers of high resolution pixel values, for each frame. When processing video data, the focus is still very strongly on frame-based information extraction, while temporal information is only considered after that. This is in very large contrast to biological systems, where a lot of information, both spatial and temporal is extracted already from lower resolutions, e.g., from peripheral vision. As an example, when seeing a movement in the corner of your eye, you very often already have a very good idea of what it is, only based on the size and movement pattern (e.g., a butterfly, a falling leaf, an insect, a person, …).
In this research position, we want to focus on more efficient ways to extract spatio-temporal visual information in artificial systems by studying how this is achieved in biological systems. One part will address improvements to artificial neural networks that extract and process such information and a second part will focus on proposing sensors systems that perform smart preprocessing in order to facilitate the task of such networks.
To perform this research, we are ideally looking for a student with a background in computational neuroscience that is strong enough to dive into the state of the art in the understanding and modelling of animal visual perception. You should, however, also have a strong interest in transferring this knowledge towards improvement in artificial intelligence and innovative sensor systems. Entering this research from a pure CS engineering background is also possible, but more difficult.
You will be enrolled at Ghent University for a PhD in Computer Science Engineering. However, your research will be highly interdisciplinary. You will need to combine in-depth understanding of biological learning, artificial learning and its efficiency as a hardware implementation.
We offer a fully funded PhD scholarship, funded by the SmartNets project for a maximum of 3 years (upon positive progress evaluation), extendable with at most 1 year from other project funding. The PhD research has both fundamental and innovative aspects. You will join a young and enthusiastic team of researchers, post-docs and professors. This PhD position is available immediately. More information on ESR-positions can be found here: https://ec.europa.eu/research/mariecurieactions/sites/mariecurie2/files/msca-itn-fellows-note_en_v2.pdf
As PhD student at Ghent university, you will collaborate with enthusiastic colleagues at IDLab-AIRO and our international partners in the SmartNets project. As an ESR in the SmartNets network, you will form an active training network with the other ESRs in the project and you are required to spend part of your PhD time (~ 2 times 3 months) with some of our partners.
- To be eligible for recruitment within an ITN project, you therefore must – at the starting date of your contract – be within the first four years after receiving you (first) master of Science diploma.
- You may not have resided in or carried out your main activity (e.g. work, studies) in Belgium for more than 12 months in the 3 years immediately before the start of your contract
- You have the degree of Master of Science in any of the following fields: Master of science in any of the following fields: Computational neuroscience, Artificial Intelligence (or similar), Computer Science (Engineering)
- you have a background in (computational) neuroscience and are proficient in programming
- you have an excellent academic track record (graduation cum laude or grades in the top 15% percentile)
- You are fluent in written and spoken English
- You are creative and prefer to find “out-of-the-box” solutions, You are particularly interested in blue-sky fundamental research, while keeping practical applicability in mind
- You are interested in and motivated by the research topic, as well as in obtaining a PhD degree
- You are systematic and organised, have excellent analytical skills, and can work independently as well as in team
- You have good communication skills, you have an open mind and a multi-disciplinary attitude.