Search

Natural Language Processing is an important area of artificial intelligence concerned with the interactions between computers and human language. Semantic matching requires accurately modeling the relevance between two portions of text and is widely used in various natural language processing tasks, such as paraphrase identification, machine translation, and question answering. In the past years, several works have continually progressed towards improving the ability to capture and analyse the text matching information (e.g. lexical, syntactic, etc.) through the application of various techniques. In this thesis, we examine the problem of developing efficient and reliable semantic matching methods for question answering and exploring the effects of the different deep learning algorithms. We split our work into three pieces, where each part refers to a different sentence matching structure. In the first part of the thesis, we propose a deep semantic similarity model to learn a distributed similarity representation for sentences pairs. Text matching can use lexical information without any consideration of semantics. Semantic similarities can be determined from human-curated knowledge, but such knowledge may not be available in every language. The novelty of the proposed architecture lies in an abstract representation of the pairwise similarities created by deep denoising stacked auto-encoders. Model training is accomplished through a greedy layer-wise training scheme, that incorporates both supervised and unsupervised learning. The proposed model is experimentally compared to state-of-the-art approaches on two different dataset types: the TREC library and the Yahoo! community question datasets. The experimental results show the proposed model outperforming other approaches. In the second part of the thesis, we focus on designing a new question-answer matching model, built upon a cross-sentence, context-aware, bi-directional long short-term memory architecture. Semantic matching between question and answer sentences involves recognizing whether a candidate's response is relevant to a particular input question. Given the fact that semantic matching does not examine a question or an answer individually, context information outside the sentence should be considered with equal emphasis to the within-sentence syntactic context. An interactive attention mechanism is proposed which automatically select salient positional answer representations, that contribute more significantly towards the relevance of an answer to a given question. In the experiments, the proposed method is compared with the existing models, using four public community datasets. The results state that the proposed model is very competitive. In particular, it offers 1.0%-2.2% improvement over the best performing model for three out of four datasets, while for the remaining one performance is around 0.5% of the best performer. In the last part of this thesis, a much more complex deep memory network structure is considered. In particular, we aim at developing a novel memory network for storing and reading the relevant question answer pairs from the internal corpus. Traditionally, to provide more related sentence information, the external resources (e.g. knowledge base, related documents) are widely used in question answering leading to large computational costs. Thus, our goal in this work is to build a reliable model that can utilize the input corpus to build the memory network without the knowledge-based resources. In the experiments, our proposed method indeed improves the matching performance on three library/community question answering datasets, when compared with those methods relying on memory structures with document resources, it achieves better performances compared with the state-of-arts.

Two novel methods in strong-field and attosecond physics are studied in this thesis: Bohmian trajectories and initial-value representations (IVRs). Both methods describe the system dynamics in terms of trajectories of electrons, and include the binding potential and the laser field. The full solution of the time-dependent Schrodinger equation (TDSE) is used as a benchmark. First, Bohmian trajectories are applied to high-order harmonic generation (HHG). We find that the trajectory located at the innermost region of the atomic core is the most important for HHG and yields spectra with a well defined plateau and cutoff. Using time-frequency analysis, we associate this trajectory to an ensemble of unbound classical trajectories leaving and returning to the core, and observe arch-like patterns that build up in the phase of the wave function. Furthermore, we relate the Bohmian trajectories to the short and long trajectories encountered in the strong-field approximation (SFA) and show that the central Bohmian trajectory overestimates the contribution of the long SFA trajectory. Second, IVRs, namely the Herman-Kluk (HK) propagator and the coupled coherent states (CCS) method, are employed in the study of strong-field ionization dynamics. This study is performed in phase space, where signatures of tunneling and over-the-barrier ionization are identified for an initially bound wave packet. Our results indicate that the HK propagator does not fully account for tunneling and over-the-barrier refections, but partly reproduces features associated with the wave packet crossing classically forbidden regions. In contrast, the CCS method accurately reproduces the results from the time-dependent Schrodinger equation (TDSE). The original CCS, compared with the TDSE, works well in short time propagation. If reprojection of the wave function to the initial grid is employed quantitative agreement between these two methods, and the first ever CCS computation of HHG spectra, with a plateau and sharp cutoff, is obtained in long time propagation.

Interstrand crosslinking (ICL) agents such as nitrogen mustards and platinum compounds are mainstays of cancer chemotherapy. SJG-136 is a highly potent ICL agent currently in clinical trials. γ-H2AX foci form at the sites of ionizing radiation-induced double strand breaks (DSBs) and their persistence is an indication of radiosensitivity. The aim of this study was to establish if γ-H2AX is a suitable marker of ICL-associated damage. γ-H2AX foci can be detected at pharmacologically relevant doses of ICL agents (0.5 μM HN2, 5 μM cisplatin, 0.01 nM SJG-136), lower than those used to detect ICLs in the modified comet assay (5 μM HN2, 10 μM cisplatin, 1 nM SJG-136). The peak of foci formation followed the peak of ICL formation, and persistence of foci was observed in CHO and human cells defective in either ICL unhooking or homologous recombination (HR). Persistence of foci was also detected in A2780 ovarian cancer cells compared to the paired A2780 cisplatin resistant cells. γ-H2AX was found to be a sensitive marker of ICL-associated damage independent of DSB formation and a potential tool to distinguish ICL-sensitive cells in vitro. γ-H2AX has also been investigated as a pharmacodynamic endpoint in clinical studies. Maximal foci formation was observed at 24 hours post-infusion of SJG-136 in peripheral blood lymphocytes obtained from all 16 patients in a phase I clinical trial. Increased foci induction was also detected in two tumour biopsies obtained from patients treated with SJG-136. In contrast to γ-H2AX, a higher level of RAD51 foci was observed in ICL unhooking defective CHO cells compared to the wild type, whereas no foci were observed in HR defective cells following cisplatin treatment. Therefore, both γ-H2AX and RAD51 foci could be useful markers of ICL-associated damage in clinical settings.

People think and reason about situations using past experience. Often, past experience consists of stereotypes and exemplars that depict common situations. In order for a computer to reason in a similar way, the identification and representation of exemplars is required. This research aims to investigate and develop a new model, called FReBE, that uses Family Resemblance and Bayesian networks for developing an Exemplar based model. In the thesis, the broad area of research is stated and the areas of background work to be studied are identified and reviewed. An exemplar based model based on family resemblance, clustering and Bayesian networks is developed and implemented. An empirical evaluation of the model is carried out using fifteen well-known benchmark datasets, such as Breast Cancer, Monks, and Heart Disease. The data sets selected include those with discrete attributes, numerical attributes and even those with missing values. The thesis includes a critical comparison with related systems such as PROTOS, PEBM and AUTOCLASS. The results show that: (a) the model is able to identify exemplars that lead to a classification accuracy comparable to published results with other methods on most of the chosen datasets; (b) most of the datasets can be represented well by a relatively small number of exemplars which are identified by FReBE; (c) FReBE shows that family resemblance can be used as a principle for finding good exemplars.

I. Chromosome Behaviour in Three Genera of Grasshoppers • II. Analysis of the Genetical Differentiation in the Species of Drosophila pseudo-obscura • III.Six New Mutations and their Linkage Relationship in Drosophila pseudo-obscura.

The concept of introducing an intermediate band to overcome the efficiency limit of single-bandgap solar cells was proposed by Luque and Martí in 1997. It is predicted that utilising the intermediate band for multi-photon absorption can significantly improve the photocurrent generation without accompanying output voltage loss. Amongst several approaches to develop an intermediate band solar cell, quantum dots have drawn much attention as intermediate band due to their three-dimensional quantum confinement and bandgap tunability. However, despite the effort expended so far, there still remains several major challenges that prevent the successful implementation of quantum dot intermediate band solar cells. The work reported in this thesis aims to provide solutions to the main challenges in implementing high-efficiency quantum dot solar cells. The work involves the design, epitaxial growth by molecular beam epitaxy, device processing, and characterisation of QDSCs. This thesis first investigates the influence of direct Si doping on InAs/GaAs quantum dot solar cells with AlAs cap layers. Si doping in QDs leads to state filling of the intermediate band, which is one of the key requirements for a high-efficiency intermediate band solar cell. Moreover, the introduction of moderate amount of Si dopants leads to passivation of defect states, and hence prolongs the carrier lifetime and increases the open-circuit voltage. Secondly, type-II InAs/GaAsSb quantum dot solar cells are studied. Increased photocurrent contribution from the quantum dot region is observed due to the prolonged carrier lifetime associated with the type-II band alignment. Lastly, different types and positions of quantum dot doping methods are investigated. The photoluminescence spectra indicate that using delta or modulation doping in quantum dots can reduce the degradation of crystal quality, and hence decrease the number of non-radiative recombination centres, when compared with using direct doping.