Heriot-Watt University
ROS: The Research Output Service. Heriot-Watt University EdinburghNot a member yet
4689 research outputs found
Sort by
A study of photooxidation reaction in a novel batch oscillatory baffled photoreactor
Full text linkThe work of this thesis focuses on a kinetic investigation into a photooxidation reaction
between α-terpinene and singlet oxygen (1O2), involves several initiatives, and has
generated a number of new results. Firstly, a novel batch Oscillatory Baffled Photo
Reactor (OBPR) was employed in this work, LEDs were planted evenly on the surfaces
of orifice baffles that move up and down the column, providing uniform light
distribution. In addition, the OBRP has excellent capabilities of multiphase mixing and
solid suspension. In this context, polymer supported Rose Bengal (Ps-RB) beads were
used as the photosensitizer, the conversion rate achieved in this study ranged between
20-30%, which is higher than similar studies that used a different type of photosensitizer.
Compared to existing studies, this work contributes novel insights into the use of
oscillatory baffled reactors for photooxidation reactions, especially in the context of
using a mobile photosensitizer. Secondly, the widely used NMR data analysis for
determining ascaridole and α-terpinene was examined by carrying out detailed kinetic
assessment of the photooxidation reaction, for the first time, conditions were identified
where the data analysis is valid or invalid.
Thirdly when the data analysis is valid, a methodology of quantifying the
concentration of singlet oxygen was, for the first time, proposed, the concentrations so
generated compared well with these from the kinetic analysis. Fourthly, definitions and
methods of evaluating the efficiency of singlet oxygen utilization and the efficiency of
photo conversion from molecule oxygen to singlet oxygen were, for the first time, proposed, this provides the insight of inefficiency of the reaction process. Fifthly,
parameters affecting kinetics were investigated, including the mass of the beads, mixing,
air flow rate, light intensity and wavelength. Finally, the durability of the Ps-RB beads
was examined, the decay of activity was evaluated
Across the counter : socially irresponsible human resource management in UK & Ireland betting firms
Full text linkSocially (ir)responsible human resource management (SIHRM) is an area of growing
interest to researchers and practitioners. However, SIHRM is in its infancy, indicating
a range of research gaps. The gap that this research fills considers how in certain
industries employers appear historically irresponsible toward their frontline employees.
Specifically, this thesis investigates the work environment experiences of front-line
staff within the retail betting industry, a sector rarely examined more widely across
HRM research.
This research investigates corporate rhetoric of betting firms and staff experiences
working within front-line betting firms, which aims to provide a nuanced understanding
of what it is like to work within the gambling industry in a customer-facing role. Drawing
on gender at work and feminist theory of violence against women this thesis
progresses the SIHRM framework by providing a gendered scope by utilising the
experiences of betting shop staff as an industry example of SIHRM practices. This
research investigates how gender at work and feminist theory of violence against
women can be used to analyse and critique the work environment experiences of frontline staff within the retail betting industry, a sector rarely examined more widely across
HRM research.
This research adopts a mixed method approach, using semi-structured interviews with
22 front-line staff and 292 staff completing a short survey from five different betting
firms in the UK. The data was analysed using thematic analysis, guided by the SIHRM
framework and the gender at work theory.
Building on the work of Richards & Sang (2019) this research contributes to the
existing counter-philosophy of SIHRM. The analysis demonstrates how betting firms
are socially irresponsible employers, and adds a gendered consideration to SIHRM
philosophy. Moreover, this thesis contributes to extending theories related to gender
and aggression, and providing a timely, unique and rare empirical accounts of betting
shop work.
The main findings of this thesis reveal that betting shop staff experience various forms
of customer misbehaviour and violence, which are influenced by gender dynamics and
the nature of the gambling industry. The thesis also shows that betting firms fail to protect and support their staff adequately, and often engage in practices that contradict
their own policies and regulations. The thesis concludes that betting firms are socially
irresponsible employers, and that SIHRM can be used as a framework to analyse and
critique their practices from a gendered perspective. The thesis also offers some
recommendations for improving the working conditions and well-being of betting shop
staff, as well as for advancing the SIHRM literature
Ultrafast laser fabrication of a K-band integrated optic 2-telescope beam combiner for astronomical interferometry
No full textThis dissertation explores the vast potential of the ultrafast laser inscription (ULI) fabrication technique for astrophotonics applications. A fibre-connectorised K-band 2-telescope
integrated optics (IO) beam combiner is designed and fabricated in commercial Infrasil®
glass (IG) to update the existing JouFLU beam combiner at the Center for High Angular Resolution Astronomy (CHARA) array. A liquid crystal on silicon phase-only spatial
light modulator (SLM) is integrated into a conventional ULI fabrication system to control
the laser phase profile at the objective plane. During the fabrication runs, the SLM is used
to provide dynamic control of the virtual numerical aperture of the writing objective. The
investigation begins by determining the ULI parameters which optimise the guiding properties of straight single mode waveguides. The insertion losses of the waveguides were
measured at 1.1 ± 0.1 dB over a 17 mm IG chip, which corresponds to 78 % of global
throughput over the whole K-band. The prototype IO beam combiner incorporates three
asymmetric directional couplers: a 3 dB coupler for interferometric measurements and
two photometric taps for calibration purposes. When tested in the lab, the interferometry
contrast of the ULI fabricated bare beam combiner provided a high figure of ≈ 87 % when
input polarisation was controlled. Furthermore, the fibre-connectorised beam combiner
prototype presented consistent interferometric performance under the same conditions
and produced an interferometric contrast of ≈ 92 %, marking a significant result in the
field of near-infrared (NIR) interferometry. This achievement opens new opportunities
for driving exoplanet research and expanding the capabilities of the K-band instruments.
The successful implementation of this fibre-connectorised IO beam combiner highlights
the potential of ULI to produce highly efficient and versatile devices for astronomical applications. The findings of this research contribute to the ongoing development of state-of-the-art astronomical instruments, uncovering new possibilities for exploring celestial
phenomena. The knowledge obtained from this study along with the beneficial partnership with our partners at, contributed to establishing the groundwork for a potential joint
project that seeks to further develop this capability
History matching and uncertainty quantification of reservoir performance with generative deep learning and graph convolutions
No full textThis research tackles the challenges faced in geological modelling under uncertainty to
flow profiling and history matching. Geological uncertainty encompasses various interpretations that may be consistent with the available data. When it comes to modelling,
these interpretations necessitate different modelling approaches and configurations. As a
result, it becomes challenging to effectively define the space of models and their parameters, as the problem’s dimensionality constantly changes. To overcome this challenge,
the study proposes an innovative approach that involves parameterisation through implicit
low-dimensional hidden spaces.
The study recognises the need to consider uncertainty in geological scenarios, structural
uncertainty, and petrophysical dependencies. These factors play a crucial role in accurately
representing and predicting the behaviour of geological objects. The chosen methodology
is the Graph Variational Autoencoder approach, which allows for the parameterisation of
the prior set of geological representations while considering various uncertainty.
The main idea behind this approach is to utilise an Encoder to map the original prior set into
a latent space that implicitly describes the prior. The latent space and the Decoder act as
a generator that can search for realisations that meet specific requirements. This methodology enables the estimation of uncertainty in dynamic response and history matching,
enhancing the overall understanding of geological systems.
The study justifies and presents a transition to graph-based generative modelling. I will
show that geometric deep learning, in particular graph convolutions, is the most convenient
method to account for geological representations with generative models. This transition
is motivated by the need to handle non-Euclidean data types, specifically those lacking
a strict structure. This enables the consideration of the structural and spatial features of
the reservoir by moving away from the classical lattice representation. The transition expands the applicability of generative models to a broader range of geological objects and
enhances the realism of the generated representations, as conventional approaches have
limitations and cannot describe the complex structural features of reservoirs or irregularities in flow behaviour.
The study employs advanced analytics tools to gain a deeper understanding of the hidden
spaces within the generative models. These tools provide valuable insights into the internal
structure of hidden spaces, allowing for a more informed analysis of the generative models’
capabilities and limitations.
Moreover, the study introduces a geodesic metric for efficient navigation in high-dimensional
hidden spaces. This metric enables more effective exploration and interpolation within
the hidden space, resulting in a more predictable behaviour than the standard Euclidean
metric. The geodesic metric also serves as the foundation for controlling the geological realism in the latent space, ensuring that the generated realisations maintain their geological
coherence.
To test the generative capabilities of a graph-based generative model, the study develops
three prior datasets of 3D geological objects, focusing on the uncertainty of geological
scenarios, structural uncertainty, and the semi-synthetic Brugge field dataset, which represents four different stratigraphic zones. These datasets serve as test cases to evaluate the
effectiveness and limitations of the proposed generative models
Essays in quantile regression with Growth-at-Risk and allied applications
Full text linkThe thesis presents new methods for quantile regression. Of key interest is the
performance of the methods in small sample settings, as my work here focuses on
implementation of these methods for macroeconometrics. In particular, the methods
are applied primarily to growth-at-risk, a macroeconometric framework designed to
capture downside (and upside) risk of GDP growth.
Chapter 2 extends the horseshoe prior to Bayesian quantile regression and provides a fast sampling algorithm for computation in high dimensions. Compared to
alternative shrinkage priors, our method yields better performance in coefficient bias
and forecast error, especially in sparse designs and in estimating extreme quantiles.
In a high dimensional Growth-at-Risk forecasting application, we forecast tail risks
and complete forecast densities using a database covering over 200 macroeconomic
variables. Quantile specific and density calibration score functions show that our
method provides competitive performance compared to competing Bayesian quantile
regression priors, especially at short and medium run horizons.
Bayesian quantile regression models with continuous shrinkage priors are known
to predict well but are hard to interpret due to lack of exact posterior sparsity.
Chapter 3 proposes a way to tackle this by decoupling shrinkage and sparsity. The
goal of this chapter is to obtain easily interpretable models with good fit. This
can be achieved by extracting variables important from the posterior of a shrunk
model using a decision theoretic framework. The proposed procedure follows two
steps: First, we shrink the quantile regression posterior through state of the art
continuous shrinkage priors; then, we sparsify the posterior by taking the Bayes
optimal solution to maximizing a policy maker’s utility function that considers the
predictive performance of the un-sparsified model as well as sparsity. In a large
Monte Carlo exercise we show that the sparsified model not only preserves the fit
but also improves it, while yielding interpretable results. We apply our approach to
a high dimensional growth-at-risk exercise in which we identify and communicate to
the policy maker which variables drive tail risks to the macroeconomy.
Quantile crossing has been an ever-present thorn on the side of quantile regression. This has spurred research into obtaining densities and coefficients that obey
the quantile monotonicity property. While important contributions, these papers
do not provide insight into how exactly these constraints influence the estimated
coefficients. Chapter 4 extends non-crossing constraints and shows that by varying a
hyperparameter one can obtain commonly used quantile estimators. Importantly, it
is shown that non-crossing constraints are simply a special type of fused-shrinkage.
Mixed frequency data improve the performance of growth-at-risk models. Nevertheless, in the literature most of the research has focused on only imposing structure
on the high-frequency lags when estimating MIDAS-QR (MIDAS (mixed data
sampling) quantile regression) models. In chapter 5 we extend the framework by
introducing structure on both the lag dimension and the quantile dimension. In
this way we are able to shrink unnecessary quantile variation in the high-frequency
variables. This leads to more gradual lag profiles in both dimensions compared to
the MIDAS-QR and UMIDAS-QR (unrestricted MIDAS quantile regression). We
show that this proposed method leads to further gains in nowcasting and forecasting
on a pseudo-out-of-sample exercise on US data.
Chapter 6 extends the at-risk framework to estimate inflation-at-risk. This
measure is particularly important as both tails, deflation and high inflation, are of
key concern to policymakers. The key insight of the chapter is that inflation is best
characterised by a combination of two types of nonlinearities: quantile variation, and
conditioning on the momentum of inflation.
Dynamic quantiles, or Conditional Autoregressive Value at Risk (CAViaR) models,
have been extensively studied at the individual level. However, efforts to estimate
multiple dynamic quantiles collectively have been limited. Existing approaches either
sequentially estimate fitted quantiles or impose restrictive assumptions on the data
generating process. Chapter 7 fills this gap by proposing an objective function for the
joint estimation of all quantiles, introducing a crossing penalty to guide the process.
Monte Carlo experiments validate the effectiveness of the method, offering a flexible
and robust approach to modelling multiple dynamic quantiles in time series data.
Chapter 8 concludes with a brief summary of the thesis, and provides an evaluation
of the limitations of the research. The critical assessment is coupled with proposals
for future research
Relationships between social identity, self-efficacy, other-efficacy, relation-inferred-self-efficacy and performance in sport
Full text linkSocial identification with one’s team predicts many positive outcomes including
motivation, group dynamics, and performance (Stephen et al., 2023). Despite initial evidence
that social identification is positively related to both self-efficacy and collective efficacy (Evans
et al., 2023; Fransen et al., 2014; Strachan et al., 2012), there has been little research to fully
investigate the link between social identification and efficacy beliefs in sport. Several studies
have demonstrated the relevance that relational efficacy beliefs (i.e., self-efficacy, other-efficacy, and relation-inferred self-efficacy) have upon psychological, behavioral, and
relational outcomes in sport. As such, the relationship between social identity and relational
efficacy is likely important for athlete success and performance. Bridging social identity and
efficacy theories offers a novel area of investigation which, to date, is under researched. The
purpose of this thesis was to investigate if athletes’ perceptions of social identification with
their team, training-group, or sport is associated with relational efficacy beliefs. Chapters 3, 4,
and 5 include original research that employs a wide range of designs including a cross sectional
study, a longitudinal study, and an experimental study. In Chapter 3, the relationship between
social identity and relational efficacy beliefs was examined using a cross-sectional design. In
Chapter 4, the relationship between social identity, self-efficacy and group-focused RISE over
time was examined using a sample of individual sport athletes (i.e., runners) using random-intercept cross lagged panel modelling. Finally, in Chapter 5, the interaction between social
identification and relational efficacy beliefs was examined using a performance task
experiment. Specific theoretical contributions from this thesis include: First, the role that RISE
serves as a potential mediator between social identification and self-efficacy, indicating its
pivotal role in connecting social identity and efficacy theories (Chapter 3). Second, the use of
random-intercept cross-lagged panel modelling allowed for within-person effects (changes
over time within individuals) to be examined, which enhanced the depth of comprehension
regarding the intricate and dynamic relationship between social identification, self-efficacy,
and RISE (Chapter 4). Third, a complex interplay in addition to predictive relationships may
exist between social identity and efficacy beliefs when predicting performance outcomes
(Chapter 5). Finally, results from this thesis provide an expansion of knowledge concerning the
influence of group-focused efficacy beliefs on individual and team outcomes (Chapters 3 and
4)
Random walks and quasi-isometries
Full text linkThis thesis deals with the study of random walks up to quasi-isometry. We
first show that some of the basic properties of random walks are not preserved
by applying a quasi-isometry of the group to a random walk. This leads to the
necessity of introducing a more general class of stochastic processes which are
preserved by quasi-isometries. To this end, we introduce tame Markov chains
and prove the first results about them. More specifically, we prove that for a
large class of groups acting acylindrically on a hyperbolic space, tame Markov
chains will make linear progress in the hyperbolic space. As a consequence, we
prove a Central Limit Theorem for random walks on groups quasi-isometric to
one of the groups in our class.
We push the study of these tame Markov chains further and introduce a new
quasi-isometry invariant, called random divergence and prove that in many cases,
the random divergence is equal to the usual divergence of the group. In an ongoing project, we study another quasi-isometry invariant called the Random Dehn
function of a group and bound this quantity for many groups.
This thesis is based on (parts of) the following articles: [GS21], [GS23],
[GHP+23] and [GZ23] as well as parts of an ongoing project: [CGMG]
Exploring novel biogeochemical approaches for de-risking energy exploration : understanding organic matter processes in the subsurface
Full text linkDissolved organic matter (DOM) is ubiquitous and exists in different subsurface environments
such as sedimentary basins, oil fields and coal systems. It is critical in controlling biogeochemical
cycles and carbon flux in surface and sub-surface ecosystems. However, the source, composition,
and fate of DOM within shallow and deep (>500 meters below ground level; mbgl) engineered
reservoirs, used for energy production or storage, is rather unknown or attracts limited attention.
The injection of fluids associated with drilling may not only leach DOM and other inorganic
constituents from the rock surfaces but also alter other rock geochemical properties. Furthermore,
these fluids may introduce other microbial taxa or add existing ones into the wellbore, kick starting deep microbiological processes. This PhD research project aims to determine the
mechanisms by which DOM is released during fluid flow in the subsurface and explore whether
the chemically released (leached) “aged” DOM can stimulate microbial biodegradation which
could lead to the generation of climate-active gases (i.e., CO2, CH4, etc.). This novel concept
requires a pairing of geochemical and microbial approaches to characterize shale/coal and its
respective leachates. The rocks and fluids geochemical characterization were combined to study
the source, mobilization, and fate of DOM in the UK Geo-Energy Observatory Site (UKGEOS),
Glasgow cores. The rock’s pyrolytic parameters were analysed using Rock-Eval Pyrolysis. The
maceral compositions in coal were examined by petrographic analysis. The δ13Corg isotopes were
analysed using elemental analyser Isotope ratio mass spectrometry (EA-IRMS). The DOM in the
fluids and leachates were analysed using a novel liquid chromatographic organic carbon detector
(LC – OCD). The CO2 and CH4 in the headspace of the microcosms were measured using a
greenhouse gas analyser (GHG). The study proves that DOM leached from the Cretaceous shale
could be bio-transformed and mineralized to generate active greenhouse gases (CO2 and CH4). It
showed that C cycling might be orchestrated by microbial activities that utilized specific C pools
available from DOM, suggesting that such interactions between DOM and microorganisms might
dictate the ultimate fate of C in shallow (<500 mbgl) engineered environments. The extraction of
SOM altered the rock’s properties by declining TOC, S2, and HI and, changed δ
13Corg isotope
compositions and organic matter quality which reduced the shale's potential for hydrocarbon
generation. The release of DOM is significantly influenced by the pyrolytic parameters (S2, TOC,
HI, and OI). The study showed that liptinite macerals influence the amount of DOM more than
vitrinite and inertinite macerals (Liptinite > vitrinite > inertness). This study confirms that the
sub-surface's drilling processes and flushing fluids can leach and mobilise DOM from the bedrock
lithologies. These findings have important implications for carbon dynamics, water quality, and
energy recovery in subsurface systems. However, complementing these studies with inorganic
studies will help to a more refined understanding of fluid chemistry. In combination, this may
guide improved subsurface energy exploitation and de-risk the operation processes
Novel approaches in macro-level neural ensemble architecture search
Full text linkThe recent successes of deep learning come partly from sophisticated deep neural
network architectures. However, designing such architectures is slow and costly, requiring many experiments and iterations by human experts. The problem motivated
the emergence of neural architecture search (NAS), which automatically discovers
highly performing neural network architectures for a particular task. Unlike most
recent work in the field that focus on small parts of the architectures, this PhD
explores novel ways of designing the overall structure, or “macro-architecture”.
In the first phase of this project, a population-based algorithm called artificial
immune system (AIS) is introduced for the first time in NAS in a framework that we
dub “ImmuNES” (Immune-inspired Neural Ensemble Search). It generates a competent and diverse population of candidate architectures, which are ensembled to
produce highly competitive results in several image classification tasks, within reasonable computational budgets, and without complex pre- and post-processing. Notably, it performs on par with a highly influential larger-scale evolutionary method
in macro-architecture search while reducing the search time by two to three orders
of magnitude. Finally, its solutions are shown to generalise across tasks without
requiring a new search.
Recognising that candidate evaluation remains the major performance bottleneck in NAS, the second phase of the research proposes a framework that addresses
this issue by predicting the performance of untrained architectures. The predictions
are obtained by encoding the search space in a context-free grammar, shaping a continuous latent space of architectural components with an autoencoder, and training
a performance predictor in that space. This Grammar+Autoencoder-based predictor is tested in various experiments, including full AIS search runs with the algorithm
designed in the first phase of research, where the predictions are used in lieu of evaluation by training. These experiments show that this approach achieves comparable
or better performance with significantly less network evaluations: around half the
budget on the initial run, and four orders of magnitude less on subsequent runs, as
the predictor’s training set generated in the initial run can be re-used. This leads
to dramatic search time amortisation, further reinforced by the resulting ensemble’s
transferability to different tasks.
The combination of the AIS-based neural ensemble search algorithm and the autoencoder-based performance predictor leads to state-of-the-art results on a challenging image classification benchmark for a fraction of the compute budget of
previous methods
An analysis of leader and follower behaviours in the context of new ways of working (NWW) : evidence from the German IT industry and recommendations for companies
Full text linkInformation technology (IT) enables certain sectors of a workforce to work wherever
and whenever they want. This new flexibility has led to New Ways of Working
(NWW): a working style that emphasises mutual trust, empowerment and the freedom
to work anywhere and anytime. Pre-COVID-19, Germany ranked considerably below
the European average for remote and time-independent working. The pandemic
substantially disrupted the German workforce and led to almost 18 million German
employees working from home from one day to the next. Predictions foresee that NWW
will become the new standard instead of the exception; thus, companies and their
leaders will need to adapt rapidly to these new working conditions.
This research will, therefore, address the following questions: What are behaviours that
are shown by leaders and followers in NWW? Will daily interactions between leaders
and followers have an impact on their mutual trust relationship? How do leaders engage
in behaviours that empower their followers? Are there any other significant leader
behaviours in NWW?
To answer these questions, this study will be conducted in the German IT industry.
Qualitative diary research is followed to provide relevant insights into the behaviour of
leaders and followers. The findings are analysed and later triangulated with semi-structured interviews.
The results add to the existing literature on NWW and provide guidance on what
constitutes appropriate leadership behaviour in this new work environment. The key
outcome is a new framework of leader behaviours and follower outcomes (LBFO) in
NWW. This framework provides insights into the expected leader behaviours and the
resulting affective state of the follower. From this, three fields were identified to give
concrete, easily implemented guidelines for companies which adopt NWW: (I)
guidelines for leader behaviour, (II) guidelines on an organisational level and (III)
considerations for organisational culture.
This research project adds to and expands on current theory, suggests practical
implications, and provides a new modern framework which companies can implement
at various levels. It also paves the way for future research into this field
Access Repository Dashboard
Do you manage ROS: The Research Output Service. Heriot-Watt University Edinburgh ? Access insider analytics, issue reports and manage access to outputs from your repository in the CORE Repository Dashboard!