Research School IrregularPublished: Thursday, 12 July 2018
A newsletter for PGR
Research School Irregular - Contents
[ top ]Research School Irregular
Prof Gavin Brown and team are investigating ensemble methods in deep learning, recently examining a new class of “modular” loss function and its theoretical/empirical relations to more mainstream approaches like dropout. A postdoc is required to augment the existing team, initially for 1 year, or 2-3 depending on experience. The appointee should be strong in Machine Learning, mathematics (esp. linear algebra and multivariate calculus) and coding, ideally in PyTorch.
Contact Gavin.Brown@manchester.ac.uk for further info
[ top ]Industrial Placements and Interships
New Internship Opportunity at BBC Sport for PhD Students
BBC R&D are seeking a Web Development intern for a project aiming to use AI techniques to dynamically reframe images, and understand which parts of an image are safe to overlay with text. The internship will last for three months (commending in September/October), be based at MediaCityUK in Salford, and have a stipend of £1500/month.
Applicants must have permission to work in the UK, and be able to take an interruption to their studies (if applicable) in order to complete the internship.
To apply, please email email@example.com with a 2 page CV and a short description (one page or less) of how your skills and experience are relevant to the project goals.
Recent work on the AI Production in R&D has demonstrated that we can analyse images to produce interesting metadata about them, and that we can produce algorithms which can dynamically frame images according to aesthetic criteria. As well as being useful for TV production, these approaches may apply in non-video situations (such as responsive web design) by allowing designers to take the content of images into consideration when producing designs. Also, this could enable the deferral of some design choices to the moment a page is rendered in the browser, taking into account more client context.
The project will investigate whether automated understanding of the content of images could free designers to produce more engaging designs. It would investigate one or both of the following approaches:
Develop a prototype responsive website that can change the framing of an image to be appropriate for the size on the screen.
Develop an algorithm that can overlay text upon images without visual disruption by automatically marking areas of the image as safe to overlay, by modelling image salience.
BBC Sport are directly interested in using the outcomes from this research to improve their products.
Human Computer Interaction
Image analysis (desirable)
In order to ensure that your application is processed in time, please send your CV to firstname.lastname@example.org by Friday 3rd August 2018.
If you have any questions regarding applications please contact email@example.com