Jay Young - Background


  • PhD Artificial Intelligence (2015)
    Learning by Observation using Qualitative Spatial Relations. An approach to Learning by Observation -- a kind of learning task where an agent learns a particular task by observing, but not interacting, with an expert -- in noisy, spatially-situated domains, using Qualitative Spatial Relations. We show how our work with QSRs allows us to represent noisy, metric observations about the world as abstract symbols, and compactly represent spatio-relational features that the original, metric representation does not have. We present a technique for searching the parameter space that controls the degree of these abstractions, so that specific spatial features can be discovered that provide higher degrees of predictive power regarding the actions chosen by an observed expert in a particular state. This is then used to learn the behaviour of both humans and other AI systems. Additionally, we use these learned behaviour models to bootstrap reinforcement learning-based systems, reducing the amount of state-space exploration that must be done, and allowing them to begin exploring from (albeit biased) known-good positions. Supervised by Dr. Nick Hawes
  • MSc Advanced Computer Science, with Distinction. (2011)
    Using Genetic Algorithms as a basis for learning goal priorities towards autonomous, self-directed AI systems. Essentially we looked at applying some of the ideas of self-motivation used on our robot Dora the Explorer to an AI for playing the Real-Time Strategy game Starcraft. The system grew much more complex than expected, and necessitated the addition of an Evolutionary Learning approach in order for it to determine how to configure itself to perform certain tasks (which, in this case, consisted of the need to defeat opponents in the game). We then required mechanisms to allow the system to autonomously navigate it's own learned space of possible configurations. I was again supervised by Dr. Nick Hawes. This work was published in an an AIIDE paper .
  • BSc Computer Science (Hons), First Class. (2010)
    My Dissertation was a 3D simulation framework for real-time construction and experimentation with biologically-inspired AI systems. Something like Netlogo but a lot more work! I was supervised by Professor Jonathan Rowe.