Social insects host a diversity of viruses. We examined New Zealand populations of the globally widely distributed invasive Argentine ant ( Linepithema humile ) for RNA viruses. We used metatranscriptomic analysis, which identified six potential novel viruses in the Dicistroviridae family. Of these, three contigs were confirmed by Sanger sequencing as Linepithema humile virus-1 (LHUV-1), a novel strain of Kashmir bee virus (KBV) and Black queen cell virus (BQCV), while the others were chimeric or misassembled sequences. We extended the known sequence of LHUV-1 to confirm its placement in the Dicistroviridae and categorised its relationship to closest relatives, which were all viruses infecting Hymenoptera. We examined further for known viruses by mapping our metatranscriptomic sequences to all viral genomes, and confirmed KBV, BQCV, LHUV-1 and Deformed wing virus (DWV) presence using qRT-PCR. Viral replication was confirmed for DWV, KBV and LHUV-1. Viral titers in ants were higher in the presence of honey bee hives. Argentine ants appear to host a range of’ honey bee’ pathogens in addition to a virus currently described only from this invasive ant. The role of these viruses in the population dynamics of the ant remain to be determined, but offer potential targets for biocontrol approaches.

The sheer volume of data to be produced by the next generation of radio telescopes—exabytes of data on hundreds of millions of objects—makes automated methods for the detection of astronomical objects ("sources") essential. Of particular importance are low surface brightness objects, which are not well found by current automated methods. This thesis explores Bayesian methods for source detection that use Dirichlet or multinomial models for pixel intensity distributions in discretised radio astronomy images. A novel image discretisation method that incorporates uncertainty about how the image should be discretised is developed. Latent Dirichlet allocation — a method originally developed for inferring latent topics in document collections — is used to estimate source and background distributions in radio astronomy images. A new Dirichlet-multinomial ratio, indicating how well a region conforms to a well-specified model of background versus a loosely-specified model of foreground, is derived. Finally, latent Dirichlet allocation and the Dirichlet-multinomial ratio are combined for source detection in astronomical images. The methods developed in this thesis perform source detection well in comparison to two widely-used source detection packages and, importantly, find dim sources not well found by other algorithms.

The next generation of radio telescopes will generate exabytes of data on hundreds of millions of objects, making automated methods for the detection of astronomical objects ("sources") essential. Of particular importance are faint, diffuse objects embedded in noise. There is a pressing need for source finding software that identifies these sources, involves little manual tuning, yet is tractable to calculate. We first give a novel image discretisation method that incorporates uncertainty about how an image should be discretised. We then propose a hierarchical prior for astronomical images, which leads to a Bayes factor indicating how well a given region conforms to a model of source that is exceptionally unconstrained, compared to a model of background. This enables the efficient localisation of regions that are "suspiciously different" from the background distribution, so our method looks not for brightness but for anomalous distributions of intensity, which is much more general. The model of background can be iteratively improved by removing the influence on it of sources as they are discovered. The approach is evaluated by identifying sources in real and simulated data, and performs well on these measures: the Bayes factor is maximized at most real objects, while returning only a moderate number of false positives. In comparison to a catalogue constructed by widely-used source detection software with manual post-processing by an astronomer, our method found a number of dim sources that were missing from the "ground truth" catalogue.

We present exploratory work into the application of the topic modelling algorithm latent Dirichlet allocation (LDA) to image segmentation in greyscale images, and in particular, source detection in radio astronomy images. LDA performed similarly to the standard source-detection software on a representative sample of radio astronomy images. Our use of LDA underperforms on fainter and diffuse sources, but yields superior results on a representative image polluted with artefacts --- the type of image in which the standard source-detection software requires manual intervention by an astronomer for adequate results.

We propose an online feature weighting method for classification by genetic programming (GP). GP's implicit feature selection was used to construct a feature weighting vector, based on the fitness of solutions in which the features were found and the frequency at which they were found. The vector was used to perform feature ranking and to perform meta-learning by biasing terminal selection in mutation. The proposed meta-learning mechanism significantly improved the quality of solutions in terms of classification accuracy on an unseen test set. The probability of success — the probability of finding the desired solution within a given number of generations (fitness evaluations) — was also higher than canonical GP. The ranking obtained by using the GP-provided feature weighting was very highly correlated with the ranking obtained by commonly-used feature ranking algorithms. Population information during evolution can help shape search behaviour (meta-learning) and obtain useful information about the problem domain such as the importance of input features with respect to each other.