18397 research outputs found
Sort by
Joint Learning of Neural Networks
No full textThe recent advances in deep learning bring many opportunities and challenges to apply this technique into some special areas, such as healthcare. This thesis focuses on developing new joint learning methods to address three fundamental challenges in deep learning, including data limitation, model over-parameterization, and model
uncertainty. By addressing these challenges, we can obtain better neural models to build more efficient automatic diganosis systems.</p
Stock Price Prediction Models using Neural Networks
No full textDue to the nonlinearity and high volatility of stock prices, it is challenging to predict stock prices. Certainty in investment decisions is the main tool for evaluating stock markets. Hence, a reliable and accurate model for predicting stock prices is desperately needed since it could inform investors about stock prices, which could ultimately result in profitable investments. This thesis introduces three novel neural networks models: BiCuDNNLSTM-1dCNN, BiCuDNNGRU-1dCNN and BiCuDNN(SLSTM-GRU)-1dCNN for stock price prediction. The models' competitive performance will make them useful for predicting time series and used as an instance for most investors to constructively evade financial hazards in investments.</p
Evaluating the impact of renewable energy policy instruments
No full textThis thesis comprises three separate studies, employing difference-in-differences estimators to evaluate the causal impact of renewable energy (RE) policy instruments on: (i) RE production - 116 countries; (ii) RE technology innovation - 26 OECD and 6 BRIICS countries; and (iii) carbon emissions reduction - 114 countries, from 1992 to 2016, in each of the wind, solar and bioenergy sectors. The findings will interest public policymakers, regulators, energy producers and their employees and consumers, lobby groups, climate change activists, investors, and societies. The findings may help these people to determine which RE policy instruments can help their country to attain the Sustainable Development Goals.</p
Do mindfulness-based interventions change brain function in people with substance dependence? A systematic review of the fMRI evidence
No full textBackground: Substance use disorders (SUDs) affect ~ 35 million people globally and are associated with strong cravings, stress, and brain alterations. Mindfulness-based interventions (MBIs) can mitigate the adverse psychosocial outcomes of SUDs, but the underlying neurobiology is unclear. Emerging findings were systematically synthesised from fMRI studies about MBI-associated changes in brain function in SUDs and their associations with mindfulness, drug quantity, and craving. Methods: PsycINFO, Medline, CINAHL, PubMed, Scopus, and Web of Science were searched. Seven studies met inclusion criteria. Results: Group by time effects indicated that MBIs in SUDs (6 tobacco and 1 opioid) were associated with changes in the function of brain pathways implicated in mindfulness and addiction (e.g., anterior cingulate cortex and striatum), which correlated with greater mindfulness, lower craving and drug quantity. Conclusions: The evidence for fMRI-related changes with MBI in SUD is currently limited. More fMRI studies are required to identify how MBIs mitigate and facilitate recovery from aberrant brain functioning in SUDs
The shear mechanical behaviour of hard undulating discontinuities under unloading normal stress
No full textThis study investigated the shear mechanical behaviour of hard undulating discontinuities with different undulation shapes. A series of direct shear tests with unloading normal stress were conducted at different initial shear stresses. The failure, deformation, and strength characteristics were analyzed. The shear strength criterion governed by the undulating angle, initial stress, and failure pattern under unloading normal stress was proposed. Analyzing the experimental results and establishing modified Patton and Barton strength criteria provided theoretical support for understanding the rock mass failure mechanism and stability evaluation in excavating slopes and underground caverns under normal unloading conditions.</p
Laboratory Evaluation of Laser Cladding on Railway Wheel Steels
No full textRegular assessments of in-service wheel/rail systems are vital to manage operational costs and preserve rail infrastructure. Train wheels often suffer wear and damage, necessitating repair or replacement through conventional but inefficient welding processes. Achieving specific surface attributes like wear, corrosion, and fatigue resistance without affecting bulk properties poses challenges. Laser cladding, known for precise protective coatings, displays potential in enhancing wear and fatigue resistance in rail components. This study investigates laser cladding parameters for metro and freight train wheels using statistical analysis and deposit examinations, evaluating phase formation and mechanical properties for durability improvements.</p
Deep Learning Technologies for Apparel Design and Display
No full textCurrently, remote online apparel design and display are more widely used than the traditional offline mode. To provide suitable and aesthetically pleasing apparel for online customers, remote measurement of body dimensions, online selection and customization of apparel styles, and multiple viewpoints observation are crucial, but they are challenging now. Additionally, the complex real backgrounds of images used for anthropometry seriously hinder the improvement of measuring body dimensions' accuracy. Similarly, they are also not conducive to generating novel views from different perspectives and protecting personal privacy. In order to overcome these problems, deep Learning collaborative technologies are explored in this thesis.</p
A Comparative Study of Physicochemical and Antioxidant Properties Between Stingless Bee Honey from Sarawak and Honey from Other Origins
No full textStingless bee honey (SBH), known as 'Kelulut' honey, holds untapped commercial potential but lacks comprehensive research compared to well-established Apis spp. Acacia and Manuka honeys. This study investigates the physicochemical attributes of selected SBH in Sarawak, contrasting them with Apis spp. Acacia honey from Sarawak and Apis spp. Manuka honey from Australia/New Zealand. By evaluating 20 samples, including properties, phenolic composition, and antioxidant activities, we enhance SBH categorization and quality grading. The findings underscore moisture-driven variations in physicochemical traits and reveal key insights into honey quality. These insights aid SBH's commercial journey and future standardization efforts
Development of Novel Heat Resistive and High Thermal Energy Storage Lightweight Concrete Panels for Energy Efficient Buildings
No full textThis study investigates developing and characterizing a novel heat-resistive and high thermal energy storage lightweight concrete panels for buildings to reduce energy use and thermal discomfort in a residential buildings at temperate oceanic climates. It also describes life cycle assessment of developed panels to assess and compare their environmental impact. The developed energy-saving panels can be used as both load-bearing and non-load-bearing walls in buildings. The outcome will help to promote a circular economy for sustainable built environment design and contribute to achieving net-zero carbon economy
Enhancing the Properties of Polymer Composites by Stable Dispersion and Magnetic Alignment of Graphene Nanoplatelets
No full textPolymer composites with graphene fillers is prepared. Excellent dispersion of high concentration graphene in composites is achieved. The fabricated foam sensor is electrically conductive, highly sensitive and exhibit sensing characteristics at low pressure <1kPa. Sensors are calibrated, and mechanical properties is studied. Composite materials with graphene orientation are prepared using static magnetic field. Nearly 50% of THz radiation is absorbed by aligned graphene composite films. The mechanism of graphene orientation under magnetic field, the degree of alignment is studied mathematically and by SANS technique. This work opens avenues for graphene inks/paints, sensing using composite materials and composites for future technologies