DeepGeo · Deep Gaussian Processes for Geostatistical Data Analysis
Horizon 2020 — Marie Skłodowska-Curie Actions
- Duration
- 2018-07-23 → 2020-07-22
- EU contribution
- €183,455
- Participants
- 1
- Scheme
- MSCA-IF-EF-ST
Lines connect the coordinator with its partners.
Results in brief
Deep Gaussian Processes for Geostatistical Data Analysis
As cities expand, old industrial sites are repurposed for residential and recreational areas. Such sites, however, can be severely polluted, and discovering the exact location and composition of the pollution is very important for public health and wellbeing. Currently, taking soil samples from polluted sites is expensive and time-consuming, and one typically only looks for certain chemicals that are defined by the authorities (such as the EU). Over time, the list of chemicals to look for is updated, but once soil from a site has been analysed, it is discarded, making it impossible to reassess it according to the new guidelines. The overall objective of this project was to take advantage of artificial intelligence to make pollution assessment easier, cheaper, and more accurate, and to make it possible to discover new types of pollution at sites sampled years ago. During the project, we showed how we can push current state-of-the-art methods to their fullest. We also developed new methods that have a better understanding of when they are uncertain about a prediction, as well as methods that would allow us to discover new types of pollution by taking advantage of correlations with known types of pollution. Unfortunately, we also discovered that it is not possible to model soil pollution accurately with the quality of the soil samples that are being collected today due to regulatory requirements. The samples are taken too far apart, leading to blind spots between them, meaning that we potentially miss pollution hotspots.
Data: CORDIS, © European Union
Project objective
Urban soil contamination resulting from former land-use is important but challenging to measure. Direct measurements are expensive and time-consuming to acquire, making a city-wide assessment impossible. Current statistical methods for modelling the distribution of pollution in urban environments, such as kriging, often fail to do so properly, since the contamination is highly local and uncorrelated with the surroundings. The problems can be mitigated by using multi-output models, such as co-kriging, where several datasets are modelled concurrently. The methods are, however, slow to train and have limited flexibility. DeepGeo will develop state-of-the-art methods for assessing urban soil contamination and provide an open-source software library for geostatistical data analysis, directly making the novel discoveries available to a wide audience. DeepGeo aims to solve the mentioned problems by the use of deep Gaussian processes for estimating urban soil pollution. This recently developed class of models promises enormous flexibility and can model highly nonlinear correlations between outputs, making them far superior to standard co-kriging. They do, however, suffer from scalability issues and empirical studies show flexibility issues with increasing depth.DeepGeo will address the scalability issue by developing new algorithms for approximate inference and for inducing sparsity. Inspired by recent advances in training of deep neural networks, specialised covariance functions that allow for deeper Gaussian process architectures will be constructed. Finally, new and improved methods for learning complicated correlations between outputs will be investigated, thus increasing the amount of information that can be gained from already available data.By making the developed methods available as open-source software, DeepGeo seeks to reach a broad range of research fields as well as benefitting the geochemical industry.
Original text from CORDIS.
Participants
- THE CHANCELLOR MASTERS AND SCHOLARS OF THE UNIVERSITY OF CAMBRIDGE · CAMBRIDGECoordinatorUnited Kingdom
Links
Data: CORDIS, © European Union