Matt Rounds

Intro

I am a student at the CDT in Data Science, at the University of Edinburgh, and in the first year of my PhD. My reasearch is motivated by an interest in how organisms (including humans) learn to interact effectively with their environment when they are constrained both by their limited access to that environment, and by their limited representational resources.

My current project models visual attention in conjunction search as a POMDP (Partially Observable Markov Decision Process) with a nonparametric prior over potential object categories in a scene. This allows the model to capture the principle features of overt human visual attention (scanpaths), in addition to covert indicators of attention (change in RT slope) when the model is 'surprised' with unexpected objects.

Projects

Non-Parametric Bayesian Modelling of Human Visual Attention (2016-present)

Modelling the interplay between covert and overt attention in the context of conjunction search - where performance decreases if distractors share features with the target object, and inattentional blindness - where an unexpected object is more likely to be spotted if its features overlap those which are task relevant. Supervisors: Frank Keller, Chris Lucas.

The Cognitive Plausibility of Deep Neural Networks with Attention (2016)

MSc(R) dissertation. An investigation into the extent to which a deep neural network using hard attention and REINFORCE to caption natural images learned policies of attention that were human-plausible. Supervisor: Frank Keller

Free Energy, Bistable Transitions, and Binary Disambiguation (2015)

MSc dissertation. Examining how a system that samples an image to minimise Free Energy acts to disambiguate between competing prior expectations of possible world states, and whether it could replicate a well-known result in bistable perception. Supervisor: Anil Seth

Phenomenological Concatenationism (2012)

MA dissertation. An argument in favour of a concatenationist account of phenomenology, structured primarily by attention. Defense of the account was grounded in both aesthetics and phenomenology. Supervisor: Peter Lamarque

Education

PhD Informatics , University of Edinburgh - in progress

MSc(R) Data Science , University of Edinburgh - Distinction

MSc Intelligent Systems , University of Sussex - Distinction

MA Culture and Thought after 1945 , University of York - Distinction

BA English and Related Literature , University of York - First Class Honours

Awards

CDT Data Science Scholarship (EPSRC grant EP/L016427/1)(2015-present)

Jan Metzger Scholarship for MSc in Intelligent Systems (2014)

Sussex Chancellor's Masters Scholarship (2014)

Teaching

TA/Marker: Computational Cognitive Science, University of Edinburgh (2016)

Tutor: Probabilistic Modelling and Reasoning, Machine Learning and Pattern Recognition, University of Edinburgh (2017,2016)

Contact

Elements

i = 0;

while (!deck.isInOrder()) {
    print 'Iteration ' + i;
    deck.shuffle();
    i++;
}

print 'It took ' + i + ' iterations to sort the deck.';