Successful social interactions require humans to accurately perceive the behaviour and actions of other individuals. Observing actions and behaviour provides a way for us to understand the intentions and minds of others, which has important implications for the way we interact. In recent years, a network of brain areas dedicated to the processing of action information has been uncovered. Despite this, however, it is still unknown what organising principles underlie our representation of the actions of other individuals. This project will address this problem by determining the psychological and neural 'space' in which actions are represented, and evaluating the dimensionality and shape of 'action-space'. The project will compile a comprehensive database of actions, generate a new large data-set of action evaluations, interrogate this new data set using novel experimental methodology and advanced quantitative statistics, and employ these findings to interrogate already existing large neural data-sets. Prior research has indicated that we have multiple different maps and models of external stimuli, where their characteristics are represented in different dimensional spaces. For example, face evaluation occurs along either 2 (Oosterhof & Todorov, 2008) or 3 (Sutherland et al. 2013) important dimensions. However, there is currently much debate as to whether faces are represented with respect to a theoretical average face (norm-based coding), or whether faces are represented as example faces across face space (exemplar based coding). Action information can be more important for social decision making than face information (Aviezer et al. 2014), however, we don't know the organising principals of how actions are represented with respect to each other. Understanding this has important implications for understanding how we see, represent, remember, and make social judgements based on the actions of other individuals. Initially, I will collate action videos from a number of large action databases to generate a broad stimulus set of approximately 500 naturalistic actions that vary as widely as possible. I will then record multiple participants' unconstrained descriptions of these actions, for each action (Expt. 1; Year 1 Term 1). The data from this experiment will provide common action descriptors on which a separate group of participants will provide ratings of the 500 actions (Expt. 2; Y1T2). An additional experiment will ask participants to provide similarity ratings between the 500 actions (Expt. 3; Y1T2). These experiments will generate a very large data-set that can be interrogated using advanced quantitative statistical techniques (multidimensional scaling; principal components analysis) to determine the dimensions on which humans evaluate naturally occurring actions. The multidimensional action space will be validated by testing with a novel set of 500 different social action videos (Expt. 4; Y1T3). Using adaptation procedures, the project will then evaluate the 'shape' of action-space in order to determine whether we represent actions on a norm or exemplar based code (Expt. 5+; Year 2). Later in the project (Year 3), based upon the multidimensional framework established earlier, it will be possible to probe existing databases to determine the neural basis underlying how we make sense of social actions. Databases interrogated will include: 1. A large data-set of single neuron data of cells that respond selectively to complex social actions. 2. Structural brain data at the York Centre for Neuroimaging. This will for the first time tell us how the different dimensions on which human action are evaluated is coded at the single cell level, and in different regions of the cortex. In addition, I will use fMR adaptation techniques during neuroimaging to determine the neural basis of how we evaluate social actions.

One main issue facing fisheries management is uncertainty regarding how fish populations will respond to changes in fishers’ behaviour or the environment. With fish resources under increasing pressure, accurate, cumulative histories of anthropogenic and environmental change are a key tool in developing effective management policies. Archaeology can help overcome this issue by providing detailed, long-range histories of local inshore fisheries and their exploitation by humans, but only if techniques for the identification and analysis of fishbone are refined. Fishbones are underrepresented in the archaeological literature because they are less stable than other taxa. Identification to species is often difficult or impossible. During my MSc I developed an identification system for fishbone: ZooMS (Zooarchaeology by Mass Spectrometry), based upon protein barcoding. As proteins can be cleaved enzymatically and analyzed by mass spectrometry in a repeatable way, protein barcoding is used widely for quick and inexpensive protein identification. Mass spectra reflect the differences in protein sequence and can therefore be reproducibly linked to a particular protein or fragment. Since I left the lab, this method for fish identification has stalled, despite earnest requests from the community for a robust method. ZooMS uses peptide fingerprinting of collagen as a method for rapid identification of archaeological bone. Identifying masses to peptides of known sequence is required. For mammals, sufficient sequence information is available, but, for freshwater fish, species are highly diverse and few sequences are available. This project will develop biomolecular techniques to overcome these hurdles, creating a database of fish collagen sequences and testing the method at several archaeological sites. With these new techniques archaeologists can provide more accurate histories of fisheries. The results will enhance our knowledge of part of our diet inform fisheries management.

The Internet of Things and connected devices is creating opportunities for the increased automation of many complex tasks. From self-driving cars and unmaned aerial vehicles to high performing AIs in increasingly complex multi-player game environments. In these scenarios, it is easy to specify a goal for the AIs to achieve, but it is much more difficult to define the strategy that the AI should follow in order to achieve the environment's goals. In these multi-agent environments, training becomes difficult because it is necessary to discover many varied strategies for all agents present in the environment. Classical approaches of training an AI to perform well on a environment with multiple agents suffers from the problem of overfitting. Where learning agents become good operating with or against themselves, but considerably drop performance when matched against other agents that act differently to those they have previously encountered. This issue is specially disastrous when one knows not of any existing good opponent strategies to test against. In the absence of a dataset of already existing agent strategies, Multi Agent Reinforcement learning offers the posibility of safely learning a good behaviour in these kind of environments by trial an error through simulation. Making it possible to learn how to solve these problems by training from the experiences encountered in the simulated environments. The focus of the PhD research is to address an open question in Multi Agent Reinforcement Learning that helps to mitigate the overfitting problem. How can we qualitatively analyze the way that different agent strategies in an environment influence the eventual strategy of a learning agent. Using Reinforcement learning techniques, it is possible to create a dataset of varied agent strategies starting without any knowledge of what a good strategy is in a given environment. Once we have a dataset of strategies, it is possible to create an algorithm that will switch??? agent strategies to present to a learning agent to maximize its performance and robustness in the environment. Making it possible to create an AI that not only performs well in a given environment, but is also robust against unseen agent strategies. The intended aim of Daniel's project is to develop Multi Agent Reinforcement Learning algorithms for learning strategies to control swarms of agents, which are capable of outperforming multiple possible opponent strategies for a specific task. This will also enable the system to control vast numbers of connected AIs simultaneously. The project will first require a system to be built that allows for the creation and testing of Multi Agent Reinforcement Learning algorithms. This system will be built on top of popular open source artificial intelligence frameworks, such as Keras and Tensorflow. Due to the computational complexity of modern Reinforcement Learning techniques, these tasks require programming proficiency in various programming languages, in addition to having a practical understanding on how to test, train and deploy AIs in highly parallelizable computing clusters. The research outcomes of the project will be tested on the platform provided by Accelerated Dynamics, a London based robotic start-up. Once the algorithms have been developed and thoroughly tested through simulation, they will have the opportunity to be tested in multiple real world scenarios using swarms of aerial drones. Thereby promoting and encouraging the sharing of efforts, insights, resources and research outcomes between industry and academia.

Econometric modelling of healthcare costs serves many purposes: to obtain key parameters in cost-effectiveness analyses; to implement risk adjustment in insurance systems; and to examine the impact of risk factors such as smoking and obesity. Modelling healthcare costs is challenging because the cost data are typically non-negative, heavy tailed and highly skewed. Filling the gap in the literature: SPEM will develop an ambitious research programme that simultaneously meet these conditions: (1) no need for retransformation of costs; (2) be able to estimate both the conditional mean and the whole distribution; (3) no need to differentiate zero costs from positive costs; (4) be less parametric and more flexible; and (5) be able to accommodate panel data. Such a method does not exist in the literature. Another highlight of SPEM is that the new method will be used for out-of-sample prediction and full distributional analysis which are typically not considered in the semiparametric framework. Promoting more informed decision making: SPEM will produce accurate and robust estimates of the relationship between childhood obesity and healthcare costs, which are crucial in the design and evaluation of government programmes aimed at treating and preventing childhood obesity. This will be achieved through an empirical application. The Longitudinal Study of Australian Children (6 waves: 2004-2014) and linked records from Medicare will be used to investigate the relationship between childhood obesity and healthcare costs.