PhD student - Active learning for solid waste characterization Ghent University, Department of Telecommunications and information processing Belgium

PhD student - Active learning for solid waste characterization


Last application date
Jan 31, 2019 23:59


Department
TW07 - Department of Telecommunications and information processing


Contract
Limited duration


Degree
Master's degree in Data Science, computer science, informatics, mathematics, physics


Occupancy rate
100%


Vacancy Type
Research staff

 

Job description

Vito and UGent are looking for candidates for a PhD research position on active learning for solid waste characterization. The selected candidate will be on Ghent University's payroll with the goal of obtaining a PhD Degree. The actual research will take place mostly at the Vito site in Mol.

 

Research question:

  • Improved training strategies: How to train a classification model for solid waste characterization by multimodal images (x-ray, 3D and color) with minimal or no training data labeling by humans.

  • Active and semi-supervised learning: How to reduce the time spent by humans on training the deep learning networks, by clever "human-in-the-loop" strategies. This also involves research into efficient interactive tools, for visualizing and interacting with the learning process.

  • Transfer learning: how to optimally make use of older training data and/or older trained networks in retraining for a new use case. How to “teach” or to steer how the algorithm calculates particle “similarity” for unseen particle classes?

  • Low level image preprocessing to reduce the input parameter space of deep learning and hence reduce the number of required labeled training samples.

 

Background:

The characterization of waste also has many important applications in existing recycling processes:

  • Process engineering: determine the technical and economical feasibility of waste sorting processes and design of new waste sorting processes

  • Online process optimization: measure, control and optimize sorting processes

  • Quality control: warrant the quality of recycled products to establish market trust

 

Currently, waste characterization is still often done manually. This approach is slow, subjective, expensive, unpleasant and eventually it delivers only little information. There is a need for a fast, objective and automated method that delivers data on a much more detailed level. Therefore VITO started the development of a multi-sensor characterization device and the required machine learning algorithms. Installed sensors include dual energy x-ray transmission, 3d laser triangulation and a high resolution color camera.

 

Currently, the algorithms have been trained to successfully recognize about a dozen of material types within mixed waste streams. Algorithm training was performed by feeding the device “pure“ material streams, of which many were manually prepared to ensure correct labeling of the data. Not surprisingly, this preparation of “pure” (mono-material) streams turned out the be a major challenge in the training process of a waste characterization device.

 

Therefore VITO wants to investigate how to develop a semi-supervised learning framework that can learn to classify waste particles with a minimum of labeled particles, while still learning from the similarities between unlabeled particles. This approach is known as active learning. During the training process, the system should classify particles in existing classes and suggest new classes to the user. For this, interactive visualization tools should be developed that allow the user to explore and interact with the particle similarity that the algorithm calculates (e.g. t-SNE).

 

Profile of the candidate

Researcher aiming for a PhD degree

 

How to apply

 

Candidates should send a CV, a detailed transcript of courses, and where possible a pdf of scientific documents (master thesis, scientific papers) they have written and which may help us to assess their scientific attitude. They should also add a motivation letter.

 

Applications should be sent to wilfried.philips@UGent.be and to roeland.geurts@vito.be.

 

Alternatively candidates can apply online: https://vito.be/en/jobs/active-learning-solid-waste-characterization


If you apply for this position please say you saw it on Computeroxy

Apply

All Jobs

FACEBOOK
TWITTER
LINKEDIN
GOOGLE
https://computeroxy.com/phd-student-active-learning-for-solid-waste-characterization,i6683.html">

ubc reklama

Anu

cambridge

geneva

kuwait

Melbourne

nottingham

Nus

sfu

southampton

texas tech

Toronto

uni copenhagen

unsw

Uwo