ROLI is developing a roadmap towards a vision of promoting the joy in the making of music, by reducing the entry barriers to music creation and empowering everyone to express themselves through music creation. Current music related digital entertainment products are mostly outdated; the "Guitar Hero" and "Singstar" style of interactions still dominate the music games market. Existing music games have yet to take advantage of human-machine interface technologies or mediums such as virtual reality and augmented reality. More importantly the current entertainment experience contributes little to learning actual musical skills and music creation. This EngD project will investigate new ways to make the experience of learning and creating music more fun and engaging through technology-driven gamification; and also explore specific creative solutions to the above challenges.

This Engineering Doctorate (EngD) research project builds on a 2016 to 2020 research collaboration between Emteq Ltd and BU CDE. CDE Research Engineer Ifigenia Mavridou's research findings show that the Emteq VR technology has significant ability to recognise valence and arousal measurements from participants exposed to affective environments. However, processing signals with different temporal characteristics, and subsequently inferring the affective state of a person based on those signals, is a challenging task. This task requires the development of novel algorithms for event detection and signal synchronisation. Furthermore, the relationship between these objective measures of affective states and the feelings of presence and immersion in the VR environment are not fully understood and further work is required on this topic.

This study aims to explore how social prescribing (SP) can be better used to improve the health and wellbeing of older people who are awaiting social care. It shall focus on domiciliary (home) care (DC), due to the significantly high number of individuals currently on a waiting list, and with conditions which can often be improved through community-based, non-clinical services. SP is a relatively new, health-based intervention, which connects individuals with community support and has greatly improved the health and wellbeing of older people when delivered effectively. However, there is evidence displaying low levels of interactions amongst this group, particularly amongst those awaiting DC. Through conducting in-depth interviews and diary keeping with social prescribers and older people awaiting DC, this study shall explore contemporary experiences with SP, gaining a detailed understanding surrounding which barriers may be restricting its effectiveness, thereby able to explicitly guide policy progressions.

Each fellowship will last up to 18 months to cover: 1. a 3-month inception phase for set up activity 2. a 12-month placement with the host organisation 3. an impact phase lasting up to 3 months Fellows will co-design projects and activities with their host and produce analysis to inform government decision-making across a range of policy priorities. Fellows will also engage across the host organisation, building effective working relationships and supporting wider knowledge exchange with researchers. This will be supported through their embedded role within the host organisation, including line management support.

Our goal is to obtain fast objects' and features' recognition from a series of images or video stream to allow augmented reality applications. Our starting hypothesis is that machine learning approaches can be applied as powerful regressors for the prediction of scene features such as flat surfaces, simple geometric objects (e.g. spheres and cylinders) and other geometric primitives. The main aim is good computational efficiency of the proposed methods which enables this approach to be used on the wide range of mobile devices (e.g. smartphones, tablets, etc.). The method would be to first generate a vast dataset of the videos with scene description to obtain ground truth data. The learning process will include setting correspondences between geometric features and the images. Both supervised learning (using of ground truth) and unsupervised learning (by using state of the art scene understanding method developed in computer vision) will be tested to obtain the most efficient framework which is going to work on different hardware. Another engineering aspect will include the problem of data storage and access, as the target hardware configuration will be limited in resources and inevitably some information will need to be stored on a server, thus diminishing the efficiency of the method. Research into finding a good balance between efficiency and required hardware resources will be another sub-project. The implementation of the architecture will be done using state of the art libraries for machine learning such as TensorFlow by Google or Torch by Facebook; this reflects that the majority of existing methods in machine learning applications for computer graphics use either of these libraries.