Search

With recent advances in machine learning research, it is now possible for many simple tasks to be accurately modelled at human level performance. Modern machine learning is therefore increasingly focused on more complex tasks. Complex datasets in machine learning are of- ten presented as a monolithic whole, and are in turn treated as such. However, it is important to remember that as tasks become more complex, we begin to see interference between the subtasks which comprise them. This phenomenon is known as inter-task interference. If this inter-task interference is not addressed, the machine learning model will attempt to find a sin- gle approximation which performs equally well on many tasks, thereby lowering performance in comparison to the union of the tasks solved individually. In order to address this issue, this thesis argues that practitioners of machine learning should account for the tasks and subtasks within complex datasets by approaching them in a multi-task learning manner. Visual Question Answering exemplifies this complexity and, we hypothesise, constitutes a collection of shared tasks in a domain, rather than a single task as it has previously been treated. We argue for a new view of neural network training as a collection of subnetworks that can be generated and exploited as needed for a given task. This thesis presents three papers analysing the multi-task nature of the Visual Question An- swering task, proposing a new method inspired by Biased Competition by which this attribute can be taken advantage of, and finally demonstrating that this new methodology can apply to other less complex datasets with similar benefits. By empirical evidence, we demonstrate that Visual Question Answering can be viewed as a multi-task learning problem resulting in both improved performance and better generalisation. We then present the second of the three pa- pers, in which we make the primary contributions of this thesis. These contributions include an argument for learning complex tasks as a collection of component subtasks and the proposal of the Biased Competition model which takes advantage of this collection of subtasks. In addition, we present a new toy dataset for developing and analysing Visual Question Answering systems quickly while preserving the complexity of the original task in the reasoning component. We demonstrate that the Biased Competition model is capable of learning to separate tasks in an end-to-end manner without the requirement of additional task type labels, and in doing so, transfers knowledge about component subtasks between each task. This both significantly re- duces catastrophic interference between tasks and in some cases augments performance, as more tasks are added to the domain. Finally, we present a third paper in which we demonstrate that the Biased Competition model is applicable to simpler datasets, and showing that it acts to counteract overfitting and improve performance of both a Deep Convolutional Neural Network model and Deep Long-Short Term Memory model by learning to separate tasks.

The thesis is concerned with capacity and coverage enhancement of OFDMA heterogeneous cellular systems with a specific focus on fractional frequency reuse (FFR), femtocells and amplify-and-forward (AF) relay systems. The main aim of the thesis is to develop new mathematical analysis for the spectral efficiency and outage probability of multi-cells multi-tier systems in diverse traffic, interference and fading scenarios. In the first part of the thesis, a new unified mathematical framework for performance analysis of FFR and soft frequency reuse (SFR) schemes is developed. This leads to new exact expressions of FFR and SFR area spectral efficiency in downlink and uplink scenarios which account for a mixture of frequency reuse factors in a homogeneous cellular system. The mathematical framework is extended to include modelling and performance analysis of FFR systems with elastic data traffic. Further analysis is carried out in relation to the performance of FFR and/or SFR schemes, in terms of energy efficiency and base station cooperation. The new proposed analytical framework can lead to a better understanding and computationally efficient performance analysis of next generation heterogeneous cellular systems. Next generation cellular systems are characterized by an increase in the spatial node density to improve the spectral efficiency and coverage, especially for users at home and at the cell edges. In this regard, relays and femtocells play a major role. Therefore, relays and femocells are the focus of the second part of the thesis. Firstly, we present a new and unified spectral efficiency analysis in dual-hop fixed-gain AF relay systems over generalised interferences models. The generalised interference models are either based on the Nakagami-m fading with arbitrary distance or on spatial Poisson Point Process in case of randomly deployed heterogeneous interferers. The models have been considered separately in the open literature due to the complexity of the mathematical analysis. Secondly, the outage probability is utilised to deduce the femtocell exclusion region for FFR system and a new static resource allocation scheme is proposed for femtocells which improve the capacity. The work presented in the thesis has resulted in the publication of seven scientific papers in prestigious IEEE journals and conferences.

From aviation and safety critical applications to personal navigation; Global Navigation Satellite System (GNSS) technology has found a place in everyday civilian life and will only become more intertwined as the technology continues to improve. One of the issues with GNSSs is that they are susceptible to interference in the forms of intentional (i.e., jamming) or unintentional sources. As this interference is a threat to civil applications, this thesis focuses on providing the framework to understand the performance degradation of GNSS receivers in the presence of interference. More specifically, this paper focuses on the high accuracy user, and how interference affects a commercially available differential GNSS (DGNSS) receiver system. A comparison was done to investigate differences between two commercially available DGNSS receivers’ performances as well as analyze the different approaches manufactures take in the integration of multi-constellation GNSS signals. After testing two DGNSS receivers, two things were clear when analyzing the data. First, there is a significant benefit when using multi-constellation GNSS signals. Not only does this help improve position accuracies; it also allows for a more robust system in the presence of GNSS interference. Second, both of the receivers tested performed worse in the presence of a GPS L1 interference signal compared against a GPS L2 interference signal when in either a GPS only or all GNSS receiver configuration.

Interference alignment (IA) has been shown to provide all users of an interference channel with half the capacity achievable in an interference free point-to-point link resulting in linear sum capacity scaling with the number of users in the high SNR regime. The linear scaling is achieved by precoding transmitted signals to align interference subspaces at the receivers, given channel knowledge of all transmit-receive pairs, effectively reducing the number of discernible interferers. The theory of IA was derived under assumptions about the richness of the propagation channel; practical channels do not guarantee such ideal characteristics. This paper presents the first experimental study of IA in measured multiple-input multiple-output orthogonal frequency-division multiplexing (MIMO-OFDM) interference channels. We show that IA achieves the claimed scaling factors in a wide variety of measured channel settings for a 3 user, 2 antennas per node setup. In addition to verifying the claimed performance, we characterize the effect of several realistic system imperfections such as channel estimation error, feedback delay, and channel spatial correlation, on sum rate performance.

What should the role of government be in regulating the economy? Through this paper the reader will gain an understanding of the impact of government intervention, when intervention is justified, and what happens when the government over regulates. The research will also reveal some major similarities between today’s economic downturn and the financial debacle of the Great Depression. Finally, analysis will show that, as a result of this unnecessary government intervention, that there is a crisis that is not solely American, but one that has become a major contributor to the global downturn.

A fundamental problem in multi-hop wireless networks is to estimate their throughout capacity. The problem can be informally stated as follows: given a multi-hop wireless network and a set of source destination pairs, determine the maximum rate r at which data can be transmitted between each source destination pair. Estimating the capacity of a multi-hop wireless network is practically useful --- it yields insights into the fundamental performance limits of the wireless network and at the same time aids the development of protocols that can utilize the network close to this limit. A goal of this dissertation is to develop rigorous mathematical foundations to compute the capacity of any given multi-hop wireless network with known source-destination pairs. An important factor that affects the capacity of multi-hop wireless networks is radio interference. As a result, researchers have proposed increasingly realistic interference models that aim to capture the physical characteristics of radio signals. Some of the commonly used simple models that capture radio interference are based on geometric disk-graphs. The simplicity of these models facilitate the development of provable and often conceptually simple methods for estimating the capacity of wireless networks. A potential weakness of this class of models is that they oversimplify the physical process by assuming that the signal ends abruptly at the boundary of a geometric region (a disk for omni-directional antennas). A more sophisticated interference model is the physical interference model, also known as the Signal to Interference Plus Noise Ratio (SINR) model. This model is more realistic than disk-graph models as it captures the effects of signal fading and ambient noise. This work considers both disk-graph and SINR interference models. In addition to radio interference, the throughput capacity of a multi-hop wireless network also depends on other factors, including the specific paths selected to route the packets between the source destination pairs (routing), the time at which packets are transmitted (scheduling), the power with which nodes transmit (power control) and the rate at which packets are injected (rate control). In this dissertation, we consider three different problems related to estimating network capacity. We propose an algorithmic approach for solving these problems. We first consider the problem of maximizing throughput with the SINR interference model by jointly considering the effects of routing and scheduling constraints. Second, we consider the problem of maximizing throughput by performing adaptive power control, scheduling and routing for disk-graph interference models. Finally, we examine the problem of minimizing end-to-end latency by performing joint routing, scheduling and power control using the SINR interference model. Recent results have shown that traditional layered networking principles lead to inefficient utilization of resources in multi-hop wireless networks. Motivated by these observations, recent papers have begun investigating cross-layer design approaches. Although our work does not develop new cross-layered protocols, it yields new insights that could contribute to the development of such protocols in the future. Our approach for solving these multi-objective optimization problems is based on combining mathematical programming with randomized rounding to obtain polynomial time approximation algorithms with provable worst case performance ratios. For the problems considered in this work, our results provide the best analytical performance guarantees currently known in the literature. We complement our rigorous theoretical and algorithmic analysis with simulation-based experimental analysis. Our experimental results help us understand the limitations of our approach and assist in identifying certain parameters for improving the performance of our techniques.

The Bluetooth wireless transmission standard provides a low-power data link between electronic devices over relatively short ranges. These links, also known as piconets, transmit using frequency hopping spread spectrum (FHSS) to send information over the air. As more applications for Bluetooth technology become available, the number of Bluetooth devices will continue to increase. With this increase in use, there will be a decrease in performance that can be attributed to Bluetooth "inter-piconet" interference. To date, very little has been published on the subject of inter-piconet interference. Previous studies have derived mean packet error rates for an increase in the number of piconets present. To come up with the mean rate, many papers make the assumption that the probability of a Bluetooth device hopping to a channel is random. However, making this assumption does not explain what happens in real time. This research gives some insight into what really happens when multiple piconets are interfering in real time. Bluetooth devices actually use a frequency hopping algorithm to determine the hopping sequence. This algorithm has been implemented in software to test various aspects of inter-piconet interference. Previous studies have shown that synchronizing the clocks among neighboring piconets will result in an increase in performance. This study shows that there are cases where synchronization alone will not provide sufficient improvement. Experimental testing has been conducted to validate some of the simulated results. Adjacent channel interference was observed during experimentation. This contradicts previous research, which has assumed that adjacent channel interference is insignificant.