Deep Learning and Data Analytics

Most modern games are starting to provide a way of generating their own new content and art, once these algorithms have been developed there is little need to spend time working on previously crucial tasks including artwork, AI behaviour and level design. On top of this, most top games are understanding the importance of streaming vast amounts of variable data from their games for further analysis, most of which is publicly accessible.

I propose the research using Deep Learning, specifically the unsupervised and supervised approaches involved with Convolutional Neural Networks to determine if in certain situations we can generate content which appears to a standard or beyond of that which is manually created such as visually stunning and interactive procedural terrain, AI which reacts in a much more robust and lifelike mannerism and art that can be conjured from the simplest inputs. This would fit in with several of the proposed interests that you have shown, "believable agents" and "game design".

The use of Deep Learning has been used frequently to generate new and intriguing content, in particular combining input images with particular styles known as Deep Art/Style. This particular method would be perfect for applying to the generation of new content textures for games.

Deep Learning and reinforcement learning have been used frequently to complete complex games and compete with even the best human players there are such as Chess and Go. I wish to further study the application of Deep Learning to more modern games to develop levels of AI that are quick to react, computationally cheap and can learn to adapt to their new environments based on real-time data feeds.

With previous work in my portfolio aimed at creating an entire engine for the rendering of large-scale planets I have a strong work environment to apply different ideas and research applications to showcase. During my degree, I have undertaken modules that are relevant to the direction of the research such as; Virtual Environments, Simulation and 3D Graphics and Artificial Intelligence. I feel a combination of these three modules produces a core backbone to the foundations of the research for AI in games.

Although I would prefer my focus to remain around Deep Learning of images the idea of analysing E-Sports data excites me, with previous time spent thoroughly involved in this scene and a background of data mining, analytical simulations of models and direct involvement with a specific API of a popular MOBA. The vast amounts of data that online games store provides a backbone for analytical projects, I would also like to investigate the correlation and prediction of gameplay via variable analysis applied with machine learning algorithms.