21684 research outputs found
Sort by
SAMFusion3D: Self-adaptive multi-modality fusion for 3D object detection in autonomous driving
Get PDFAutonomous vehicles rely on a diverse array of sensors to achieve comprehensive visual perception of their surroundings. Consequently, the integration of multimodal data, aimed at harnessing the complete spectrum of features from each sensor’s Bird’s Eye View (BEV) information, has emerged as a pivotal area of interest for numerous researchers. Currently, the research community is dedicated to enhancing the accuracy of detection models. However, given that the visual perception systems of autonomous vehicles are typically compact to medium-sized mobile platforms, computational complexity and efficiency are paramount. As the surrounding environment of an autonomous vehicle can fluctuate rapidly at times, maintaining a static sampling rate in such varied contexts results in suboptimal computational efficiency. Furthermore, as each modality’s features are processed through Vision Transformers, particularly in the self-attention mechanism where the attention values for features are computed, it has been observed that adhering to the conventional pipeline approach results in elevated computational complexity and diminished efficiency. For the self-adaptive sampling mechanism, we adeptly extract depth information from camera features by utilizing point cloud data. Then, the fusion rate, which functions as a regulatory factor, dynamically adjusts the size of the effective sampling intervals, significantly impacting the computational load of the feature integration process. We also adopted the structure of the iTransformer that masterfully inverts the dimensions of the embedding. Our experiments conducted on the nuScenes dataset prove that our model can perform with reduced computational complexity while maintaining results comparable to those of the baseline model
Automatic detection of laughter in spontaneous conversations
Get PDFLaughter is an important expression used to communicate in a variety of important ways. It is used to signal enjoyment and humour, to control and maintain the flow of conversations, to help mediate the discussion of controversial conversation topics and is used to help speakers bond. Given laughter’s wide range of uses, if they are to engage in effective human-computer interactions, it is vital for computers to be able to detect laughter. However, laughter is not homogeneous. There are two broad types of laughter: voiced and unvoiced. In addition, many individuals have different and unique ways of laughing. The pitch, volume, length and frequency of laughter has a wide divergence across speakers. Furthermore, it is used infrequently. These factors make the application of machine learning approaches, to the automatic detection of laughter, difficult.
This thesis initially shows, through a literature review, that the task of laughter detection has been widely addressed previously. However, the field has placed constraints upon the task of laughter detection. These constraints are shown to split the field into three broad tasks. Type 1 classification tasks involve short clips of between 1 and 3 seconds and contain only one kind of speech event (i.e., laughter, speech, sighs or fillers) being classified. Type 2 tasks make use of medium length clips of between 3 and 11 seconds. Each clip of this type contains multiple speech events: however, laughter can constitute a large amount of the total audio of each clip, i.e., between 10-30%. Finally, type 3 tasks employ long form conversations that are between 10 minutes to an hour. Laughter makes up less than 10% of the audio in this case: there is no guarantee of any laughter being present. Initially, it is shown that these three types of tasks vary in difficulty. Evidence of this is given by examining the F1 score achieved by the same methodology when applied to the three tasks. Scores vary from 80-100% in type 1 tasks and 50% in type 2 tasks to 25% in type 3 tasks. Furthermore, a disparity in the effectiveness of laughter detection methods, as estimated by different evaluation metrics, is found. This is shown to lead to an over-estimation of the effectiveness of state-of-the-art methods in types 2 and 3 laughter detection tasks.
This thesis replicates the state-of-the-art research on a publicly available type 2 corpus, achieving a frame level F1 of 40% and an event level F1 of 52%. It then applies these methods to the SSPNet Mobile Corpus, a private type 3 dataset, and shows the same methods achieve a frame level F1 of 15% and an event level F1 of 26%. An extensive performance analysis illustrates that the longer length of the audio introduces a large number of false laughter detections that are centred on speech. It is then demonstrated that methods that specifically target the removal of these false detections, by leveraging automatic speech recognition, are able to achieve a frame level F1 of 30% and an event level F1 of 45%. This enables an almost two-fold increase in performance over the state-of-the-art approaches for type 3 tasks.
Transformers are then applied to the task. It is demonstrated that these transformers, pretrained on audio tasks such as automatic speech recognition, can be used to extract attention embeddings in terms of low level descriptors of the audio data. Such embeddings are shown to be more effective than hand-crafted features for training laughter detectors. This method is then demonstrated to achieve a frame level F1 of 60% and an event level F1 of 80%, i.e., the best results achieved in type 3 laughter detection. The effectiveness of this approach is then replicated on the SSPNet Vocalisation corpus, which achieves a frame level F1 of 77% and an event level F1 of 88%. Furthermore, it is shown to be as effective at the task of automatic filler detection by achieving a frame level F1 of 70% and an event level of 80%. The final section applies a selection of the laughter detection systems to detect differences in laughter behaviour due to the gender composition of the speakers in a conversation. This demonstrates an initial use-case of automatic speaker information extraction. Overall, this thesis accomplishes effective laughter detection in a type 3 task
Use of synthetic mRNA transfection encoding telomerase to improve the in-vitro lifespan of dental pulp stem cells
Get PDFDental pulp stem cells (DPSCs), like most stem cells, have the potential to be used in regenerative medicine. However, one of the limitations to the usage of DPSCs is their inability to maintain a long term, in-vitro lifespan due to replicative senescence. A key contributing factor to this senescence is telomere erosion.
Our research aimed to generate and use a synthetic messenger RNA (mRNA) encoding the human telomerase, a key protein responsible for telomere maintenance. mRNA was chosen instead of reprogramming cells using DNA constructs as the latter carries the risk of unwanted genomic integrations. Successes with the highly effective COVID-19 vaccines and other RNA drugs, such as small interfering RNA (siRNA) drugs, has driven substantial improvements in RNA delivery technologies and greatly increased the potential of mRNA approaches.
After correcting a mutation found in the telomerase mRNA sequence, we synthesised telomerase mRNA from the corrected plasmid using CleanCapAGTM to add cap1 structure to the 5’ end, substituting the normal uridine triphosphate (UTP) with the N1-methylpseudouridine triphosphate and polyadenylating the final product. We used eGFP-encoding mRNA made the same way as a transfection control. We showed that both DPSCs and a control cancer cell line, TR146, could be transfected with enhanced green fluorescent protein (eGFP) mRNA at high efficiency with a simple procedure and an inexpensive commercially available RNA transfection reagent.
Telomere lengths (TL) in DPSCs (both wild type and transfected) were quantified as a relative measurement using the quantitative polymerase chain reaction (qPCR) method originally designed by Richard Cawthon and later modified by O’Callaghan and Fenech. We found that, indeed, telomere erosion occurs in DPSCs as the cells age by increasing passage numbers. However, DPSCs with shorter telomere lengths transfected with the telomerase mRNA showed no sign of telomere extension.
In contrast, TR146 cancer cells which had shorter telomeres than DPSCs had their telomeres significantly extended after transfection with the same synthetic telomerase mRNA indicating the mRNA was functional. It is possible that the synthetic telomerase mRNA is not translated in DPSCs, and Western blot evaluation is needed to test this. Initial experiments suggested other genes required for telomere extension were transcribed in DPSCs. However further works is also needed to test if these are expressed strongly enough
An investigation into the role of tumour cell Rab27a-dependent small extracellular vesicles in modulating the tumour immune microenvironment
No full textAbstract not currently available
Impact of doping and interfacial band bending of charge transport layer in inverted perovskite solar cells
Get PDFThis thesis investigates the n-doping reaction mechanism and interfacial band bending of charge transport layers (CTLs) in inverted perovskite solar cells (PSCs). CTLs, comprising electron transport materials (ETMs) and hole transport materials (HTMs), play a crucial role in determining the efficiency of PSCs by facilitating efficient charge extraction and transport while minimising recombination losses. However, device performance is often hindered by challenges such as low intrinsic conductivity of organic ETMs. To address these challenges, this work explores the n-type doping in non-fullerene organic ETMs, with a focus on improving conductivity and studies self-assembled monolayers (SAMs) as HTMs, understanding the influence of fermi levels on high efficiency of inverted PSCs.
The study begins by examining functionalized bisflavin (BF) derivatives and naphthalenediimide (NDI) derivatives with glycol and alkyl side-chains as nonfullerene ETMs due to bio-inspired nature and more straightforward synthetic process, compared to conventional ETM such as [6,6]-phenyl-C61-butyric acid methyl ester (PCBM). Due to inherently lower conductivity of these pristine derivatives, n-type doping was performed to enhance the conductivity using the dopant to generate free radical, as confirmed through electron paramagnetic resonance (EPR) measurements. UV-vis absorption spectroscopy and conductivity studies revealed that derivatives with polar glycol side-chains (BFG and NDI-G) facilitated faster doping reactions compared to the non-polar alkyl counterparts (BFA and NDI-EtHx). This behaviour was attributed to the polarity compatibility between the glycol side-chains and the dopant, which promoted molecular interactions and enhanced the doping efficiency.
Interestingly, the BF and NDI systems exhibited distinct responses to doping effects. While the doped BF derivatives show limited improvement in charge transport, the doped NDI derivatives demonstrated significant conductivity enhancements. Optimised NDI-G doped materials achieved a conductivity exceeding 10-2 S/cm, resulting in improved photovoltaic performance. Density functional theory (DFT) calculations explained these observations by highlighting the formation of charge transfer complexes (CTCs) with strong binding energies. The alignment of energy levels between CTC and neutral molecules was found to be critical for effective electron transfer and the generation of free charges. Based on these findings, a detailed doping mechanism is proposed in this work.
Additionally, bulk defects such as ion vacancies caused the surface recombination in the PSC system, it is necessary to decouple the charge accumulation from recombination. In here, we investigate using a novel stabilization and pulse (SaP) measurement technique to decouple the ionic feature with electronic effect, studying the influence of SAMs on Fermi-level alignment in PSCs. SAMs with varying dipole moments (MeO-2PACz, Me-4PACz, and 2PACz) were studied, revealing distinct flat ion potentials (Vflat) that affected charge extraction efficiency. An optimal Vflat of approximately 0.8 V was identified, while higher values were associated with interfacial barriers and reduced performance. Supporting evidence from Kelvin probe microscopy (KPFM) and time-resolved photoluminescence (TRPL) further confirm this hypothesis.
In summary, this thesis contributes insights into the charge transport and recombination through the n-type doping of non-fullerene organic ETMs and the interfacial band bending of SAM-based HTMs in inverted PSCs. The findings underline the strategic importance the doping mechanism and the critical role of interfacial engineering in enhancing photovoltaic performance. These results have broader implications for advancing efficient perovskite-based solar technologies
Digital innovation to support reperfusion therapies in acute ischaemic stroke
No full textAbstract not currently available
The role of E-cadherin loss in transgenic mouse skin carcinogenesis
Get PDFThe purpose of this study is to examine the role of E-cadherin in cell-cell adhesion failure and altered signalling within a transgenic mouse skin carcinogenesis model that models human carcinogenesis. The main aim is to understand how the loss of E-cadherin collaborates with the activation of ras and fos oncogenes, along with PTEN loss, in driving the transformation of benign tumours into malignancies and their subsequent progression to aggressive squamous cell carcinomas (SCC). These are considered the most critical events in the progression of carcinogenesis from a patient’s perspective.
E-cadherin is a highly conserved adhesion molecule essential for maintaining epithelial integrity through its interaction with β-catenin. Loss of E-cadherin disrupts adhesion, promoting invasion and progression. Although many studies support this view, emerging evidence, such as that from Padmanaban et al. (2019), suggests that E-cadherin may also be necessary for certain types of invasion, emphasising the complexity of its role and the importance of tissue context in carcinogenesis.
To address these challenges, the approach involved a well-characterised transgenic mouse skin carcinogenesis model that expresses a combination of ras and fos activation, via a modified human keratin 1 vector (HK1.ras; HK1.fos), ensuring that ras and fos oncogenes are exclusively expressed in the proliferative basal layers of the epidermis and in malignant tumours. This results in hyperplasia and papillomatogenesis but shows no evidence of spontaneous malignant conversion. The stability of this phenotype makes it ideal for investigating the roles of oncogene cooperation in the development of benign tumours and their progression to malignancy. To explore the specific roles of E-cadherin in skin carcinogenesis models, the RU486-inducible cre/lox method was utilised to exclusively knock out E-cadherin via exon 6-10 ablation (K14creP/Δ-6-10Ecadflx).
These transgenic mice expressed activated ras and/or fos oncogenes (HK1.ras, HK1.fos). Inducible mutation of PTEN-regulated AKT activation via exon 5 ablation (K14creP/Δ5PTENflx) was also incorporated into this model by using the RU486-inducible cre/loxP method. Previous analysis of endogenous E-cadherin expression in the HK1.ras.fos-K14creP Δ5PTENflx/flx transgenic skin mouse model showed a reduction in membranous E-cadherin expression at the invasive front of well-differentiated SCC (wdSCC) following p53 loss.
In bi-genic HK1.ras-K14creP/Δ6-10Ecadflx mice, the synergistic effect between E-cadherin loss and wound-promotion sensitive (ear tagging) HK1.ras1205 line was observed to induce malignant conversion. The results showed that reduced functional E-cadherin in HK1.ras-K14creP/Δ6-10Ecadflx/het mice led to hyperplasia and papillomas similar to HK1.ras mice, but with intercellular gaps in the basal keratinocytes and carcinoma in situ. Functional E cadherin ablation in HK1.ras-K14creP/Δ6-10Ecadflx/flx initially caused malignant transformation into well-differentiated squamous cell carcinoma (wdSCC) that invaded in a grouped, collective manner but rapidly progressed into poorly differentiated squamous cell carcinoma (pdSCC), consistent with cell-cell adhesion failures and invasion by the more aggressive individual mode. These tumours correlated with cell-cell adhesion failure associated with p53 loss and nuclear β-catenin expression.
The heterozygous HK1.fos-K14creP/Δ5PTENflx/flx/Δ6-10Ecadflx/het mice develop keratoacanthomas (KAs) similar to HK1.fos-K14creP/Δ5PTENflx/flxmice, exhibiting characteristic micro-cysts that indicate accelerated and premature differentiation. No malignant transformation was observed because of strong membranous basal E-cadherin and basal membranous β-catenin expression, which triggers nuclear p53 expression. However, functional ablation of E-cadherin in HK1.fos-K14creP/Δ5PTENflx/flx/Δ6-10Ecadflx/flx KAs led to malignant conversion into invasive wdSCC. The HK1.fos-K14creP/Δ5PTENflx/flx/Δ6-10Ecadflx/flx mice initially develop KA-like tumours, with minimal effects on early pre-KA hyperplasia stages, similar to observations in HK1.ras mice. Over time, these tumours progress to invasive wdSCC with scattered areas of SCC. This progression ultimately results in the development of aggressive SCC. The gradual loss of p53, combined with a decrease in membranous and an increase in nuclear β-catenin expression, along with the progressive loss of Δ6-10E-cadherin, indicates a shift towards a more advanced SCC phenotype. These molecular changes suggest an increased potential for invasion and highlight the crucial role of these factors in driving malignant transformation and progression.
The HK1.ras1276.fos-K14creP/Δ5PTENflx/flx/Δ6-10Ecadflx model was also utilised, and in this model, the tumours were independent of wound promotion. This was evident as tumours developed in both TGE and NTGE of RU486-treated HK1.ras1276.fos-K14creP/Δ5PTENflx/flx/Δ6-10Ecadflx mice. The emergence of these tumours demonstrated a synergic interaction, resulting in a rapid malignant conversion and progression to SCC/pdSCC soon after p53 loss, consistent with cell-cell adhesion failure and deregulated signalling to β-catenin. Interestingly, these tumours were associated with loss of β-catenin, suggesting a critical link between E-cadherin loss and the dysregulation of β-catenin signalling. The loss of E-cadherin, a key component of adherens junctions, might disrupt the normal Wnt/β-catenin signalling dynamics, particularly affecting its nuclear translocation. This disruption may lead to inadequate accumulation of nuclear β-catenin, promoting its degradation rather than facilitating its role in gene transcription regulation. The consequent reduction in β-catenin expression could further exacerbate the loss of cell-cell adhesion, thereby accelerating the malignant progression of these tumour models.
In summary, this study demonstrates that E-cadherin loss has a minimal impact on papillomatogenesis but promotes hyperplasia and, in cooperation with Ras activation and endogenous p53 loss, facilitates malignant conversion. Furthermore, in the context of ras and fos activation with PTEN loss, E-cadherin loss accelerated malignant progression, consistent with disruption of cell-cell adhesion. Altogether, these models recapitulate key features of human SCC progression and provide insight into the molecular mechanisms linking E-cadherin loss, β-catenin dysregulation, and tumour invasiveness. Understanding these interactions not only deepens the comprehension of SCC pathogenesis but also highlights potential therapeutic targets aimed at restoring adhesion or modulating Wnt/β-catenin signalling to delay or prevent malignant progression
Design and optimization of transceiver design for semantic communication
Get PDFSemantic communication (SemCom) has emerged as a transformative paradigm that transmits the meaning of information instead of raw data, offering significant potential to break through the limitations of traditional communication systems. By focusing on semantics, SemCom can dramatically reduce data transmission requirements, enhance communication efficiency, and enable intelligent decision-making in complex environments. Its applications range from terrestrial wireless networks to extraterrestrial missions, where it can significantly reduce communication latency and ensure reliable information exchange. However, the practical deployment of SemCom requires overcoming multiple challenges to ensure its universal applicability and robust performance across diverse scenarios.
These challenges can be broadly categorized into two key areas. First, when introducing SemCom within traditional communication frameworks, selecting an appropriate semantic coding model (SCM) remains difficult due to the diversity of source information, user background knowledge (BK), and dynamic channel conditions. Efficiently managing computing resources and bandwidth is also essential, as large-scale semantic coding models consume significant resources. Furthermore, constructing effective background knowledge for reasoning in SemCom is complex, particularly when training data is insufficient or incomplete. Addressing these challenges is critical to achieving the general applicability and reliability of SemCom. Second, SemCom’s potential can be further unlocked in specific engineering applications where its advantages are particularly evident. For example, in autonomous lunar landing missions and UAV/UGV cooperative operations, SemCom’s capability to extract and transmit only the most relevant semantic information becomes essential. However, these scenarios introduce additional obstacles such as channel instability, limited computational capacity, and dynamic communication conditions. Developing tailored SemCom frameworks is necessary to ensure robust performance and reliable communication in these demanding environments.
To ensure the reliable and efficient operation of SemCom, my work proposes a series of solutions. First, a Background knowledge Aware SCM SElection (BASE) scheme is developed to tackle the SCM selection problem. BASE leverages graph theory to model relationships between different BKs and employs a deep learning algorithm to predict the performance of semantic coding models. This approach achieves higher information recovery accuracy and improves the likelihood of selecting optimal models compared to traditional methods. Second, a joint computing resource and bandwidth allocation framework is proposed to optimize resource management in SemCom networks. Formulated as a deep reinforcement learning task, this problem is addressed using a multi-agent proximal policy optimization algorithm, which maximizes semantic accuracy under resource-constrained conditions. Third, to enhance the reliability of SemCom transceivers, a GAI-assisted SemCom framework (Gen-SC) is introduced. By utilizing Generative Artificial Intelligence (GAI) to generate high-quality training samples tailored to user contexts, Gen-SC improves the reasoning capabilities of semantic coding models. A discriminator module further ensures that generated samples align with actual data distributions, enabling higher semantic accuracy, especially in scenarios with limited training data.
Building on these foundations, the effectiveness of SemCom is demonstrated in challenging application scenarios. For extraterrestrial missions, a novel SemCom framework is designed to support autonomous lunar landing. This framework facilitates the transmission of essential image features from the lander to satellites running remote landing control algorithms. By employing adaptive semantic encoding, it enhances landing accuracy, reduces end-to-end latency, and ensures robust performance in harsh lunar environments. Additionally, a control aware SemCom framework is proposed for UAV and UGV cooperative path planning. Instead of transmitting raw sensory data, this framework extracts and communicates only the critical semantic information relevant to path planning. This approach effectively addresses communication challenges caused by channel fading, interference, and occlusion. The proposed transceiver design ensures accurate and timely coordination between UAVs and UGVs, improving path planning efficiency and mission success rates. Through these contributions, my work advances the understanding and application of SemCom, providing comprehensive solutions for its practical deployment. The proposed methodologies enhance resource efficiency, ensure reliable transceiver operation, and demonstrate robust performance in both terrestrial and extraterrestrial environments. These findings offer valuable insights into the development of intelligent and resilient communication systems for future applications
Understanding shoulder surfing and informing the design of protection mechanisms
Get PDFShoulder surfing, the act of looking at the screen of someone’s device without their consent, is a ubiquitous threat when accessing information on personal devices like smartphones. With the rapid increase in the use of smartphones, the threat of shoulder surfing is also increasing. This thesis first contributes a systematic literature analysis that focuses on the resources required for targeted attacks against mobile devices and finds that shoulder surfing, which belongs to the novice attacks category, is one of the most accessible attacks. This is because it requires no sophisticated setup. An attacker must only be near a user to observe the device’s screen. Considering the ease of execution of shoulder surfing, we investigated shoulder surfing more in-depth through two studies, which are this thesis’s second and third contributions. First, we conducted a one-month diary study to understand how shoulder surfing happens in the real world. We found that shoulder surfing can happen anywhere, anytime, without the users realising it. Further, our results showed that content such as text and photos are shoulder surfed more frequently than authentication credentials. Second, to examine the impact and importance of addressing shoulder surfing, we conducted an online survey asking participants how it impacted their social lives, perceptions of privacy, and interactions with their mobile devices. We discovered that shoulder surfing is a deep concern among users, affecting their perception of privacy. It was seen as the gateway to threats like identity or device theft. Based on the empirical discoveries around how shoulder surfing happens and impacts users’ privacy perceptions, the fourth contribution of this thesis looks into uncovering a user-centred approach to designing protection mechanisms. For this, we designed and validated a scientific instrument, the Out-of-Device Privacy Scale (ODPS), to measure users’ privacy regarding threats in the physical world. ODPS fills the gap between protection mechanisms and users’ perceptions of privacy. The fifth contribution presents an exploratory study to explore correlations between personal attributes such as ODPS and user preferences for privacy mechanisms extracted from the literature. The results proved that user preferences for protection mechanisms highly correlate with ODPS. Overall, the results help understand the relationship between a user’s perception of privacy against device-external threats and the design of protection mechanisms. We conclude by discussing design recommendations to assist in developing novel protection mechanisms. Based on a series of empirical investigations, this thesis presents a user-tailored privacy investigation of shoulder surfing and informs the design of protection mechanisms
Data sources:
Access Repository Dashboard
Do you manage Open Research Online? Become a CORE Member to access insider analytics, issue reports and manage access to outputs from your repository in the CORE Repository Dashboard! 👇