Subject

Data scarcity poses a significant challenge in deep learning. Specifically for air pollution, data scarcity can be due to lost signals or unmeasured locations by sensors. The problem is usually solved with data reconstruction or generation models. Addressing the issue of data reconstruction and generation requires innovative deep approaches, for instance, with deep generative models. Deep generative models like Variational Graph Autoencoders (VGAEs) offer promising solutions by generating fake data with characteristics similar to the input data and leveraging variational inference mechanisms. Similarly, Markov Random Fields (MRF) provide approximate inference for graphical models, enhancing the reconstruction process. This master thesis proposal aims to explore the integration of VGAEs and MRFs to improve air pollution forecasting by posing the problem as a data reconstruction and generation task.

Kind of work

The student will study deep generative models, focusing on VGAEs, and comprehend graph properties and the functioning of graph neural networks. Additionally, understanding MRFs and their integration into neural network layers will be crucial. The primary task involves merging VGAEs with MRF layers to develop a model for data reconstruction and generation. The thesis will employ existing datasets on air pollution data. The model can also leverage other context features such as people density in urban areas, traffic, satellite images or weather data.

Framework of the Thesis

The thesis relates to various projects running in the ETRO department, including projects in collaboration with industrial partners such as imec and hospitals.
The thesis will leverage long standing results in the department, including datasets and methods. Some publications are available below:
• MRF model: Fake News Detection using Deep Markov Random Fields. Duc Minh Nguyen, Tien Huu Do, Robert Calderbank, Nikos Deligiannis. NAACL 2019. (https://aclanthology.org/N19-1141/)
• VGAEs: Variational Graph Auto-Encoders. Thomas N. Kipf, Max Welling NIPS 2016 (https://arxiv.org/abs/1611.07308 )
• VGAES for data reconstruction: Matrix Completion With Variational Graph Autoencoders: Application in Hyperlocal Air Quality Inference. Tien Huu Do et al. ICASSP 2019 https://arxiv.org/abs/1811.01662

Number of Students

1 or 2

Expected Student Profile

Good knowledge of Machine Learning, AI and data processing. Good programming skills in Python (particularly PyTorch).