Learning from Geometry

Subspaces and manifolds are two powerful models for high dimensional signals. Subspaces model linear correlation and are a good fit to signals generated by physical systems, such as frontal images of human faces and multiple sources impinging at an antenna array. Manifolds model sources that are not linearly correlated, but where signals are determined by a small number of parameters. Examples are images of human faces under different poses or expressions, and handwritten digits with varying styles. However, there will always be some degree of model mismatch between the subspace or manifold model and the true statistics of the source. This dissertation exploits subspace and manifold models as prior information in various signal processing and machine learning tasks.

A near-low-rank Gaussian mixture model measures proximity to a union of linear or affine subspaces. This simple model can effectively capture the signal distribution when each class is near a subspace. This dissertation studies how the pairwise geometry between these subspaces affects classification performance. When model mismatch is vanishingly small, the probability of misclassification is determined by the product of the sines of the principal angles between subspaces. When the model mismatch is more significant, the probability of misclassification is determined by the sum of the squares of the sines of the principal angles. Reliability of classification is derived in terms of the distribution of signal energy across principal vectors. Larger principal angles lead to smaller classification error, motivating a linear transform that optimizes principal angles. This linear transformation, termed TRAIT, also preserves some specific features in each class, being complementary to a recently developed Low Rank Transform (LRT). Moreover, when the model mismatch is more significant, TRAIT shows superior performance compared to LRT.

The manifold model enforces a constraint on the freedom of data variation. Learning features that are robust to data variation is very important, especially when the size of the training set is small. A learning machine with large numbers of parameters, e.g., deep neural network, can well describe a very complicated data distribution. However, it is also more likely to be sensitive to small perturbations of the data, and to suffer from suffer from degraded performance when generalizing to unseen (test) data.

From the perspective of complexity of function classes, such a learning machine has a huge capacity (complexity), which tends to overfit. The manifold model provides us with a way of regularizing the learning machine, so as to reduce the generalization error, therefore mitigate overfiting. Two different overfiting-preventing approaches are proposed, one from the perspective of data variation, the other from capacity/complexity control. In the first approach, the learning machine is encouraged to make decisions that vary smoothly for data points in local neighborhoods on the manifold. In the second approach, a graph adjacency matrix is derived for the manifold, and the learned features are encouraged to be aligned with the principal components of this adjacency matrix. Experimental results on benchmark datasets are demonstrated, showing an obvious advantage of the proposed approaches when the training set is small.

Stochastic optimization makes it possible to track a slowly varying subspace underlying streaming data. By approximating local neighborhoods using affine subspaces, a slowly varying manifold can be efficiently tracked as well, even with corrupted and noisy data. The more the local neighborhoods, the better the approximation, but the higher the computational complexity. A multiscale approximation scheme is proposed, where the local approximating subspaces are organized in a tree structure. Splitting and merging of the tree nodes then allows efficient control of the number of neighbourhoods. Deviation (of each datum) from the learned model is estimated, yielding a series of statistics for anomaly detection. This framework extends the classical {\em changepoint detection} technique, which only works for one dimensional signals. Simulations and experiments highlight the robustness and efficacy of the proposed approach in detecting an abrupt change in an otherwise slowly varying low-dimensional manifold.

Uncomfortable 'truths' and the centenary anniversary of the 1916 Rising in Ireland

How to deal with uncomfortable ‘truths’ from the past has long posed problems for historians and politicians alike and this is exemplified by attempts to ‘deal with’ the centenary anniversary of the 1916 Easter Rising in Ireland. How do we recognise the revolutionary ‘heroes’ of the past and their contribution to the building of the new ‘nation’ state to which we may pledge allegiance, without exposing the contradictions inherent in the way that ‘nation’ state has transformed, subverted and indeed corrupted many of the ideas for which they fought? More controversially, how do we honour the actions of revolutionaries in the past which led to death and destruction in pursuance of a grand ideal, while at the same time condemning others today who claim to have been likewise engaged, using similar methods, during the recent ‘Troubles’ (1969-98 and counting)? Attempts by the Irish state to deal with the centenary seem to illustrate the point.

Multi-Party Trust Computation in Decentralized Environments

In this paper, we describe a decentralized privacy-preserving protocol for securely casting trust ratings in distributed reputation systems. Our protocol allows n participants to cast their votes in a way that preserves the privacy of individual values against both internal and external attacks. The protocol is coupled with an extensive theoretical analysis in which we formally prove that our protocol is resistant to collusion against as many as n-1 corrupted nodes in the semi-honest model. The behavior of our protocol is tested in a real P2P network by measuring its communication delay and processing overhead. The experimental results uncover the advantages of our protocol over previous works in the area; without sacrificing security, our decentralized protocol is shown to be almost one order of magnitude faster than the previous best protocol for providing anonymous feedback.

The redemption of the past. On a main motif in Walter Benjamin’s work

The following paper examines Walter Benjamin’s reflection on the category of “redemption”, mainly developed in the theses On the concept of History. To this end, we will try firstly to reconstruct Benjamin’s critique of “fate”, as it unfolds in the twenties on the field of right, economy and, especially, history. The critique of the expiatory logic of “fate” – developed in essays such as Fate and Character, Critique of violence or Capitalism as religion – will then allow us to disclose the “dialectical” structure of redemption, whereby Benjamin mobilizes his previous theory of knowledge against the doctrine of progress.

Secure Multiple Amplify-and-Forward Relaying with Co-Channel Interference

We investigate the impact of co-channel interference on the security performance of multiple amplify-and-forward (AF) relaying networks, where N intermediate AF relays assist the data transmission from the source to the destination. The relays are corrupted by multiple co-channel interferers, and the information transmitted from the relays to destination can be overheard by the eavesdropper. In order to deal with the interference and wiretap, the best out of N relays is selected for security enhancement. To this end, we derive a novel lower bound on the secrecy outage probability (SOP), which is then utilized to present two best relay selection criteria, based on the instantaneous and statistical channel information of the interfering links. For these criteria and the conventional maxmin criterion, we quantify the impact of co-channel interference and relay selection by deriving the lower bound on the SOP. Furthermore, we derive the asymptotic SOP for each criterion, to explicitly reveal the impact of transmit power allocation among interferers on the secrecy performance, which offers valuable insights into practical design. We demonstrate that all selection criteria achieve full secrecy diversity order N, while the proposed in this paper two criteria outperform the conventional max-min scheme. 

Equality

Equality is a political and moral ideal that refers to some universal condition thought to be shared by human beings. Since this inherent equality is often thought to have been corrupted by a self-interested secular world, this essay shifts the emphasis from equality as a timeless concept to equalization as a historical process.

Thermal Tracking and Estimation for Microprocessors

Due to increasing integration density and operating frequency of today's high performance processors, the temperature of a typical chip can easily exceed 100 degrees Celsius. However, the runtime thermal state of a chip is very hard to predict and manage due to the random nature in computing workloads, as well as the process, voltage and ambient temperature variability (together called PVT variability). The uneven nature (both in time and space) of the heat dissipation of the chip could lead to severe reliability issues and error-prone chip behavior (e.g. timing errors). Many dynamic power/thermal management techniques have been proposed to address this issue such as dynamic voltage and frequency scaling (DVFS), clock gating and etc. However, most of such techniques require accurate knowledge of the runtime thermal state of the chip to make efficient and effective control decisions. In this work we address the problem of tracking and managing the temperature of microprocessors which include the following sub-problems: (1) how to design an efficient sensor-based thermal tracking system on a given design that could provide accurate real-time temperature feedback; (2) what statistical techniques could be used to estimate the full-chip thermal profile based on very limited (and possibly noise-corrupted) sensor observations; (3) how do we adapt to changes in the underlying system's behavior, since such changes could impact the accuracy of our thermal estimation. The thermal tracking methodology proposed in this work is enabled by on-chip sensors which are already implemented in many modern processors. We first investigate the underlying relationship between heat distribution and power consumption, then we introduce an accurate thermal model for the chip system. Based on this model, we characterize the temperature correlation that exists among different chip modules and explore statistical approaches (such as those based on Kalman filter) that could utilize such correlation to estimate the accurate chip-level thermal profiles in real time. Such estimation is performed based on limited sensor information because sensors are usually resource constrained and noise-corrupted. We also took a further step to extend the standard Kalman filter approach to account for (1) nonlinear effects such as leakage-temperature interdependency and (2) varying statistical characteristics in the underlying system model. The proposed thermal tracking infrastructure and estimation algorithms could consistently generate accurate thermal estimates even when the system is switching among workloads that have very distinct characteristics. Through experiments, our approaches have demonstrated promising results with much higher accuracy compared to existing approaches. Such results can be used to ensure thermal reliability and improve the effectiveness of dynamic thermal management techniques.

The Anagó Language of Cuba

This thesis investigates the socio-linguistic factors that led to the emergence of a new language in Cuba known as Anagó. This language emerged from contact between multiple dialects of the West African Yoruba language and Spanish. Language contact between the Yoruba language and Spanish took place in Cuba beginning in the nineteenth century after the introduction of large numbers of Yoruba speakers into Cuba during the trans-Atlantic slave trade. This thesis argues against the opinion that Anagó is simply a corrupted and imperfect form of Yoruba. Instead, it maintains that Anagó is a new language that emerged in Cuba and became a functional vehicle for the transmission of ideas and culture. Additionally, this study will present evidence that the Anagó speaking community was a constituent part of Cuban society since the nineteenth century, and is therefore an inextricable part of Cuban cultural patrimony. Twentieth century examples of Anagó language are examined as evidence of a vital Anagó speaking transnational community.

Inside standard and hyperspectral low-dose CT variational imaging: parameter identification and material decomposition

The main contribution of this thesis is the proposal of novel strategies for the selection of parameters arising in variational models employed for the solution of inverse problems with data corrupted by Poisson noise. In light of the importance of using a significantly small dose of X-rays in Computed Tomography (CT), and its need of using advanced techniques to reconstruct the objects due to the high level of noise in the data, we will focus on parameter selection principles especially for low photon-counts, i.e. low dose Computed Tomography. For completeness, since such strategies can be adopted for various scenarios where the noise in the data typically follows a Poisson distribution, we will show their performance for other applications such as photography, astronomical and microscopy imaging. More specifically, in the first part of the thesis we will focus on low dose CT data corrupted only by Poisson noise by extending automatic selection strategies designed for Gaussian noise and improving the few existing ones for Poisson. The new approaches will show to outperform the state-of-the-art competitors especially in the low-counting regime. Moreover, we will propose to extend the best performing strategy to the hard task of multi-parameter selection showing promising results. Finally, in the last part of the thesis, we will introduce the problem of material decomposition for hyperspectral CT, which data encodes information of how different materials in the target attenuate X-rays in different ways according to the specific energy. We will conduct a preliminary comparative study to obtain accurate material decomposition starting from few noisy projection data.