No description available.

Recommender systems automate the process of recommending products and services to customers based on various types of data including customer demographics, product features, and, most importantly, previous interactions between customers and products (e.g., purchasing, rating, and catalog browsing). Despite significant research progress and growing acceptance in real-world applications, two major challenges remain to be addressed to implement effective e-commerce recommendation applications. The first challenge is concerned with making recommendations based on sparse transaction data. The second challenge is the lack of a unified framework to integrate multiple types of input data and recommendation approaches.This dissertation investigates graph-based algorithms to address these two problems. The proposed approach is centered on consumer-product graphs that represent sales transactions as links connecting consumer and product nodes. In order to address the sparsity problem, I investigate the network spreading activation algorithms and a newly proposed link analysis algorithm motivated by ideas from Web graph analysis techniques. Experimental results with several e-commerce datasets indicated that both classes of algorithms outperform a wide range of existing collaborative filtering algorithms, especially under sparse data. Two graph-based models that enhance the simple consumer-product graph were proposed to provide unified recommendation frameworks. The first model, a two-layer graph model, enhances the consumer-product graph by incorporating the consumer/product attribute information as consumer and product similarity links. The second model is based on probabilistic relational models (PRMs) developed in the relational learning literature. It is demonstrated with e-commerce datasets that the proposed frameworks not only conceptually unify many of the existing recommendation approaches but also allow the exploitation of a wider range of data patterns in an integrated manner, leading to improved recommendation performance.In addition to the recommendation algorithm design research, this dissertation also employs the random graph theory to study the topological characteristics of consumer-product graphs and the fundamental mechanisms that generate the sales transaction data. This research represents the early step towards a meta-level analysis framework for validating the fundamental assumptions made by different recommendation algorithms regarding the consumer-product interaction generation process and thus supporting systematic recommendation model/algorithm selection and evaluation.

No / Ethnicity is one of the most salient clues to face identity. Analysis of ethnicity-specific facial data is a challenging problem and predominantly carried out using computer-based algorithms. Current published literature focusses on the use of frontal face images. We addressed the challenge of binary (British Pakistani or other ethnicity) ethnicity classification using profile facial images. The proposed framework is based on the extraction of geometric features using 10 anthropometric facial landmarks, within a purpose-built, novel database of 135 multi-ethnic and multi-racial subjects and a total of 675 face images. Image dimensionality was reduced using Principle Component Analysis and Partial Least Square Regression. Classification was performed using Linear Support Vector Machine. The results of this framework are promising with 71.11% ethnic classification accuracy using a PCA algorithm + SVM as a classifier, and 76.03% using PLS algorithm + SVM as a classifier.

Engineering problems have characteristics such as a large number of variables, non-linear, computationally expensive, complex and black-box (i.e. unknown internal structure). These characteristics prompt difficulties for existing optimisation techniques. A consequence of this is that the required optimisation time rapidly increases beyond what is practical. There is a needfor dedicated techniques to exploit the power of mathematical optimisation tosolve engineering problems. The objective of this thesis is to investigate thisneed within the field of automation, specifically for control optimisation ofautomated systems.The thesis proposes an optimisation algorithm for optimising the controlof automated interacting production stations (i.e. independent stations thatinteract by for example material handling robots). The objective of the optimisation is to increase the production rate of such systems. The non-separable nature of these problems due to the interactions, makes them hard to optimise.The proposed algorithm is called the Constructive Cooperative CoevolutionAlgorithm (C3). The thesis presents the experimental evaluation of C3, bothon theoretical and real-world problems. For the theoretical problems, C3 istested on a set of standard benchmark functions. The performance, robustness and convergence speed of C3 is compared with the algorithms. This shows that C3 is a competitive optimisation algorithm for large-scale non-separable problems.C3 is also evaluated on real-world industrial problems, concerning thecontrol of interacting production stations, and compared with other optimisation algorithms on these problems. This shows that C3 is very well-suited for these problems. The importance of considering the energy consumption and equipment wear, next to the production rate, in the objective function is also investigated. This shows that it is crucial that these are considered to optimise the overall performance of interacting production stations.

Power limits on future high-performance computing (HPC) systems will constrain applications. However, HPC applications do not consume constant power over their lifetimes. Thus, applications assigned a fixed power bound may be forced to slow down during high-power computation phases, but may not consume their full power allocation during low-power I/O phases. This paper explores algorithms that leverage application semantics-phase frequency, duration and power needs-to shift unused power from applications in I/O phases to applications in computation phases, thus improving system-wide performance. We design novel techniques that include explicit staggering of applications to improve power shifting. Compared to executing without power shifting, our algorithms can improve average performance by up to 8% or improve performance of a single, high-priority application by up to 32%.

Two dimensional magnetic recording (TDMR) achieves high areal densities by reducing the size of a bit comparable to the size of the magnetic grains resulting in two dimensional (2D) inter symbol interference (ISI) and very high media noise. Therefore, it is critical to handle the media noise along with the 2D ISI detection. In this paper, we tune the generalized belief propagation (GBP) algorithm to handle the media noise seen in TDMR. We also provide an intuition into the nature of hard decisions provided by the GBP algorithm. The performance of the GBP algorithm is evaluated over a Voronoi based TDMR channel model where the soft outputs from the GBP algorithm are used by a belief propagation (BP) algorithm to decode low-density parity check (LDPC) codes.