All Projects

[No Longer Available] Understanding Social Behaviour from the Seamless Interaction Dataset

Rachel McDonnell (email | all projects)

How do two people coordinate their movement, gaze, gestures and speech during a conversation, and what can these behaviours tell us about the social interaction taking place? In this open-ended project, the student will use Meta’s Seamless Interaction dataset, a large-scale multimodal dataset of face-to-face interactions containing processed body motion, facial behaviour, gaze, audio, speech … Read more

[TAKEN] Looking at Each Other: A Dataset of Social Gaze in Conversation

Rachel McDonnell (email | all projects)

Eye gaze plays a fundamental role in face-to-face communication, signalling attention, engagement, turn-taking and social intent. However, gaze is often studied from the perspective of only one participant. This project will create a new dataset of synchronised eye movements from pairs of people engaged in natural social interactions, using two wearable Pupil Labs eye-tracking headsets. … Read more

[TAKEN] Does This Voice Fit? Face–Voice Congruence in Virtual Humans

Rachel McDonnell (email | all projects)

The voice of a virtual human is a powerful cue to its identity, personality, age, emotion and social characteristics. Psychological research on cross-modal face–voice perception suggests that people have expectations about which voices belong with which faces. People can make associations between unfamiliar faces and voices, while characteristics such as vocal pitch, apparent age, femininity/masculinity, … Read more

Working with Parsing Corpora

Martin Emms (email | all projects)

There are various corpora of ‘gold standard’ syntactically analysed natural language. Below is a sample from the ‘Penn TreeBank’, analysing the sentence ‘W.R. Grace holds three of Grace Energy ‘s seven board seats’ ( (S (NP-SBJ (NNP W.R.) (NNP Grace) ) (VP (VBZ holds) (NP (NP (CD three) ) (PP (IN of) (NP (NP (NNP … Read more

Computational Treatement of Historical Sound Changes

Martin Emms (email | all projects)

BRIEFLY: for pairs of languages there seem to be ‘sound substitutions’ that systematically map words of one language to words of similar meaning in another, something explicable by supposing that the words in question languages are descended from a common ancestor, with systematic and divergent sound evolutions along the way. Working with pencil-and-paper in a … Read more

Reinforcement Learning and Game Playing

Martin Emms (email | all projects)

There is a theoretical and implementational field ‘Game Theory’ (GT), originating in the 1940s, concerned with ‘games’: situations where players interact with each other and possibly some other objects in a very formalised way, gaining individual ‘pay-offs’ when the ‘game’ ends. Tic-Tac-Toe is a so-called ‘2-person zero-sum game’ and GT is concerned with what are … Read more

Simulating Populations and their use of Language in Cooperation

Martin Emms (email | all projects)

Individuals in a species (esp. humans) may cooperate and they may communicate. Recent decades have seen quite a lot of efforts to use computational simulation — so-called individual/agent-based modelling — to investigate what cooperation is , and how communication factors into it, striving to fit this into a general picture of varying levels of survival … Read more

[TAKEN] The AI Museum Guide: Social Characters for Multiplayer VR

Rachel McDonnell (email | all projects)

Imagine visiting a virtual museum as a family after closing time, where the exhibits have mysteriously come to life and an AI-driven virtual museum guide needs your group’s help to discover what has happened. This project will develop a short multiplayer VR experience for 3–4 players, in which a MetaHuman guide dynamically responds to the … Read more

Continual Model-based Reinforcement Learning – No longer available

Vinny Cahill (email | all projects)

Model-based reinforcement learning (MBRL) is emerging as a powerful paradigm for building intelligent agents capable of operating in the physical world. Most existing MBRL algorithms assume that the environment is stationary, i.e., that its behaviour doesn’t change so that the learnt world model and the strategies than depend on it remain optimal over time. But … Read more