Ioannis Psorakis

Multiclass Relevance Vector Machines: Sparsity and Accuracy

In this paper, we investigate the sparsity and recognition capabilities of two approximate Bayesian classification algorithms, the multiclass multi-kernel relevance vector machines (mRVMs) that have been recently proposed. We provide an insight into the behavior of the mRVM models by performing a wide experimentation on a large range of real-world datasets. Furthermore, we monitor various model fitting characteristics that identify the predictive nature of the proposed methods and compare against existing classification techniques. By introducing novel convergence measures, sample selection strategies and model improvements, it is demonstrated that mRVMs can produce state-of-the-art results on multiclass discrimination problems. In addition, this is achieved by utilizing only a very small fraction of the available observation data.

Inferring social network structure in ecological systems from spatio- temporal data streams

We propose a methodology for extracting social network structure from spatio-temporal datasets that describe timestamped occurrences of individuals. Our approach identifies temporal regions of dense agent activity and links are drawn between individuals based on their co-occurrences across these ‘gathering events’. The statistical significance of these connections is then tested against an appropriate null model. Such a framework allows us to exploit the wealth of analytical and computational tools of network analysis in settings where the underlying connectivity pattern between interacting agents (commonly termed the adjacency matrix) is not given a priori. We perform experiments on two large-scale datasets (greater than 106 points) of great tit Parus major wild bird foraging records and illustrate the use of this approach by examining the temporal dynamics of pairing behaviour, a process that was previously very hard to observe. We show that established pair bonds are maintained continuously, whereas new pair bonds form at variable times before breeding, but are characterized by a rapid development of network proximity. The method proposed here is general, and can be applied to any system with information about the temporal co-occurrence of interacting agents.

Dynamic Bayesian Combination of Multiple Imperfect Classifiers

Classifier combination methods need to make best use of the outputs of multiple, imperfect classifiers to enable higher accuracy classifications. In many situations, such as when human decisions need to be combined, the base decisions can vary enormously in reliability. A Bayesian approach to such uncertain combination allows us to infer the differences in performance between individuals and to incorporate any available prior knowledge about their abilities when training data is sparse. In this chapter we explore Bayesian classifier combination, using the computationally efficient framework of variational Bayesian inference. We apply the approach to real data from a large citizen science project, Galaxy Zoo Supernovae, and show that our method far outperforms other established approaches to imperfect decision combination.We go on to analyse the putative community structure of the decision makers, based on their inferred decision making strategies, and show that natural groupings are formed. Finally we present a dynamic Bayesian classifier combination approach and investigate the changes in base classifier performance over time.

Consistent individual differences in the social phenotypes of wild great tits, Parus major

Despite growing interest in animal social networks, surprisingly little is known about whether individuals are consistent in their social network characteristics. Networks are rarely repeatedly sampled; yet an assumption of individual consistency in social behaviour is often made when drawing conclusions about the consequences of social processes and structure. A characterization of such social phenotypes is therefore vital to understanding the significance of social network structure for individual fitness outcomes, and for understanding the evolution and ecology of individual variation in social behaviour more broadly. Here, we measured foraging associations over three winters in a large PIT-tagged population of great tits, and used a range of social network metrics to quantify individual variation in social behaviour. We then examined repeatability in social behaviour over both short (week to week) and long (year to year) timescales, and investigated variation in repeatability across age and sex classes. Social behaviours were significantly repeatable across all timescales, with the highest repeatability observed in group size choice and unweighted degree, a measure of gregariousness. By conducting randomizations to control for the spatial and temporal distribution of individuals, we further show that differences in social phenotypes were not solely explained by within-population variation in local densities, but also reflected fine-scale variation in social decision making. Our results provide rare evidence of stable social phenotypes in a wild population of animals. Such stable social phenotypes can be targets of selection and may have important fitness consequences, both for individuals and for their social-foraging associates.

Inferring social structure from temporal data

Social network analysis has become a popular tool for characterising the social structure of populations. Animal social networks can be built either by observing individuals and defining links based on the occurrence of specific types of social interactions, or by linking individuals based on observations of physical proximity or group membership, given a certain behavioural activity. The latter approaches of discovering network structure require splitting the temporal observation stream into discrete events given an appropriate time resolution parameter. This process poses several non-trivial problems which have not received adequate attention so far. Here, using data from a study of passive integrated transponder (PIT)-tagged great tits Parus major, we discuss these problems, demonstrate how the choice of the extraction method and the temporal resolution parameter influence the appearance and properties of the retrieved network and suggest a modus operandi that minimises observer bias due to arbitrary parameter choice. Our results have important implications for all studies of social networks where associations are based on spatio-temporal proximity, and more generally for all studies where we seek to uncover the relationships amongst a population of individuals that are observed through a temporal data stream of appearance records.

The role of social and ecological processes in structuring animal populations: a case study from automated tracking of wild birds

Both social and ecological factors influence population process and structure, with resultant consequences for phenotypic selection on individuals. Understanding the scale and relative contribution of these two factors is thus a central aim in evolutionary ecology. In this study, we develop a framework using null models to identify the social and spatial patterns that contribute to phenotypic structure in a wild population of songbirds. We used automated technologies to track 1053 individuals that formed 73 737 groups from which we inferred a social network. Our framework identified that both social and spatial drivers contributed to assortment in the network. In particular, groups had a more even sex ratio than expected and exhibited a consistent age structure that suggested local association preferences, such as preferential attachment or avoidance. By contrast, recent immigrants were spatially partitioned from locally born individuals, suggesting differential dispersal strategies by phenotype. Our results highlight how different scales of social decision-making, ranging from post-natal dispersal settlement to fission–fusion dynamics, can interact to drive phenotypic structure in animal populations.

Maritime abnormality detection using Gaussian processes

Novelty, or abnormality, detection aims to identify patterns within data streams that do not conform to expected behaviour. This paper introduces novelty detection techniques using a combination of Gaussian processes, extreme value theory and divergence measurement to identify anomalous behaviour in both streaming and batch data. The approach is tested on both synthetic and real data, showing itself to be effective in our primary application of maritime vessel track analysis.