21793 research outputs found
Sort by
Understanding and Locating Quality Issues in the Database Access Code of Database-Backed Applications
Get PDFDatabase-backed applications interact with the database management system (DBMS), such as MySQL, for persistent data storage. These database accesses play a central role in such applications and are crucial for their maintenance and quality. Developers build database-backed applications to access relational databases using object-oriented programming languages such as Java, Python, C#, PHP, and C++. Since object-oriented programming is a different paradigm compared to relational databases, developers use various technologies to ease database access by abstracting persistent data as objects. Specifically, developers often rely on two main access technologies: (i) executing a Structured Query Language (SQL) query and manually converting the results to objects; and (ii) using Object-Relational Mapping (ORM) frameworks, which automatically generate SQL queries and convert the results to objects based on various object-database mapping configurations. However, developers may face different database access challenges when using different technologies. Moreover, due to the abstraction of ORM frameworks, developers may face challenges when debugging database access problems. ORM automatically generates SQL queries based on various ORM configurations (e.g., the relationship among object types) and the invoked ORM APIs. As a result, developers do not have direct control over how ORM generates SQL queries. If there is a database access issue associated with a problematic-generated SQL query, developers may have difficulties knowing how and where the SQL query is generated in the application code, causing challenges in debugging database access problems.
Motivated by the importance and challenges of database access, in this thesis, we first conduct an empirical study of database access bugs in seven large-scale Java open-source applications that use relational database management systems. Specifically, by manually examining the bug reports and commit histories ranging from 5 to 16 years, we investigate and derive the characteristics such as categories, root cause, impact, and occurrence of database access issues when using popular database access technologies. Our empirical study provides motivations and guidelines for future research to help avoid, detect, and test database access bugs in database-backed applications. To assist developers in debugging database access problems, we propose an approach for locating the origin (i.e., the control flow path containing a sequence of method calls) that generates a given SQL query. It achieves state-of-the-art localization accuracy and improves Top@5 accuracy by 225% and 333% compared to the baseline approach when using SQL session logs and individual query logs, respectively. We also find that our approach can help developers locate data access issues that generate problematic SQL queries (i.e., slow SQL queries and database deadlocks). In conclusion, this thesis uncovers the root causes of database access issues and demonstrates that leveraging both static analysis and information retrieval techniques can help developers debug database access issues associated with problematic SQL queries. It also paves the way for future research on the development and automatic generation of tests for database access code to improve the quality of database-backed applications
Automated Fiber Placement for Dome-Type Structures Using Dual Robots
Get PDFThe growing complexity and precision required in modern manufacturing processes have led to an increased use of automation and robotics, particularly in advanced composite material fabrication techniques such as Automated Fiber Placement (AFP). This thesis explores the ways to improve the fiber placement process and quality through the use of dual robot systems. The principal objective of this research is to investigate the potential of dual-robot systems to improve the precision and efficiency of fiber placement on dome-type geometries with singularity avoidance, particularly within the aerospace industries.
In this study, a dual robot system consisting of two industrial robots, i.e., Fanuc M20-iA with an automated fiber placement (AFP) head and a second robot are used to simulate the fiber placement process on a curved, dome-type surface. A typical example of this type of geometry in aerospace structures is a fuselage pressure bulkhead. RoboDK, an offline robot simulation and programming tool, is employe
REFINING OPTIMIZATION METHODS FOR TRAINING MACHINE LEARNING MODELS: A CASE STUDY IN ROBOTIC SURGICAL PROCEDURES
Get PDFMachine learning is a technology that builds predictive models from data, allowing generalization to unseen cases. At the core of every learning problem lies an optimization challenge, and solving these problems reliably is crucial to resolving the obstacles surrounding machine learning. Primarily, conventional optimization algorithms employed for training machine learning frequently are often ill-suited for various applications. Concerted efforts are needed to refine and optimize various components of machine learning training. This thesis explores fundamental optimization algorithms across various machine learning applications. By enhancing optimization schemes, including optimizers and model compression techniques, the resilience and effectiveness of machine learning applications can be improved.
The first segment of the thesis introduces an innovative low-rank matrix factorization scheme aimed at enhancing the scalability of machine learning. Gaussian Process Regression is used as the machine learning
model to scale with low rank matrix factorization in this section of the thesis due to its lightweight nature, which enables the incremental updating of model parameters online prior to prediction. A nonconvex
formulation of a low-rank matrix factorization (SRLSMF) with convex formulation of a low-rank matrix factorization initialization (ℓ1-SRLSMF), is advocated to scale Gaussian Process Regression (GPR). Thus, by
employing convex nonconvex low rank matrix factorization to scale a given the Gaussian Process Regression model, the model can avoid local minima and converge to a solution with smaller recovery residuals.
Also, the running time of convex nonconvex low rank matrix factorization is expected to be smaller than that of applying nonconvex low rank matrix factorization alone under the same stopping criterion. To the best of our knowledge, the machine learning method proposed in this thesis is the first to exploit nonconvex formulation of a low-rank matrix factorization (SRLSMF) with convex formulation of a low-rank matrix factorization initialization (ℓ1-SRLSMF) to scale machine learning in machine learning domain. Recognizing the cost-prohibitive nature of standard eigen decomposition for online Gaussian Process Regression covariance update, we implement incremental eigen decomposition within the ℓ1-SRLSMF and SRLSMF Gaussian Process Regression methodologies. Finally, an illustration of the potential applications in suturing, knot-tying and needle passing task using kinematic dataset is provided.
In the latter part of the thesis, a novel adaptive stochastic gradient descent (ASGD) method, which leverages the non-uniform p-norm concept to train machine learning is presented. The proposed ASGD assigns distinct categories of coordinate values with varying base learning rates, thereby enabling the training of machine learning models. Additionally, theoretical guarantees for the efficacy of the proposed ASGD method in convex and nonconvex settings is discussed. The ability of the proposed ASGD approach to detect suturing gestures within the remote surgical gesture recognition task is discussed.
Finally, potential avenues for future research are outlined
Mujeres Colombianas y Arte Denunciante: Collective Resistance, Community Safeguarding and Memory-Making (2010-2024).
No full textHow has community-based art led by women creators contributed to confronting violence and promoting peace in Colombia, a country recovering from a civil war? (1948-2016). This dissertation highlights the often-overlooked role of women in peacebuilding at a local level through their artistic endeavours. It focuses on the period from 2010 to 2024 when art practices adopted a communal approach, primarily driven by women from rural and peripheral communities. The thesis highlights how their creative work has become crucial in reconciliation outside official frameworks, allowing them to denounce and challenge violence and engage with their communities through art.
The study establishes the concept of "arte denunciante" (art that denounces), encompassing art forms that confront violence, historical oppression, and systemic injustices in multiple contexts. Through the lenses of race and gender, it addresses how women's practices of "arte denunciate" have evolved from traditional approaches in museums and galleries to more inclusive, community-driven processes of social change. Thus, the research frames community-based “arte denunciante” as "invented spaces" and "sites of re-existence," showcasing their role in providing platforms for historically excluded voices and fostering communal reconciliation. By focusing on the creative strategies of Colombian women and their impact on local peace-building, the research underscores the broader relevance of “arte denunciante” as an approach applicable to global contexts experiencing conflict and violence.
The methodology of this reserach, which incorporated a "SentirPensar" (feeling-thinking) approach, was based on genuine connections with artists, active listening, site visits, and critical analyses of case studies. As such, the research was developed through oral history interviews with eight artists creating “arte denunciate" and site visits conducted in four peripheral regions in Colombia, including two urban centers and two rural towns.
Reliability Analysis of Low-Cost Underwater Wireless Sensor Networks
No full textUnderwater Wireless Sensor Networks (UWSNs) are one of the critical technologies that play a key role in environmental monitoring, ocean exploration, and marine security. UWSNs are deployed in harsh environments face aggressive conditions, limited energy, and bound communication bandwidth; maintenance is very expensive since ship missions are required to recover to deployed nodes. However, it reduces efficiency, reliability, and lifetime. This thesis presents the issues of reliability and lifetime issues of UWSNs by proposing an integrated method using fault-tree analysis (FTA), mean time between failures (MTBF), Failure Mode and Effects Analysis (FMEA), and reliability metrics. The proposed framework analyzes and compares the reliability and lifetime of existing UWSNs.
Furthermore, the thesis delves into the critical aspect of reliability analysis for low-cost underwater sensor nodes, which are becoming increasingly prevalent due to their affordability. A model is proposed for estimating the reliability of such nodes from individual component reliabilities. Case studies on existing low-cost underwater sensor node designs are used to validate this model.
The thesis integrates these methods to devise a framework that achieves substantial enhancement in reliability and lifetime, especially for cost-effective node-based UWSNs. That is to say, this will go a long way in establishing Underwater Sensor Networks as much more reliable and efficient in their operations for different applications. Moreover, the improvement in circuit reliability helped come up with a better design that improved upon its reliability and extended its lifetime
Learning Contextual Vision Representations via Masking
Get PDFSupervised learning on large-scale labeled datasets has been critical to the success of computer vision, with widespread applications in robotics, healthcare, security, sports, and retail. To address challenges of over-dependence on labeled data, self-supervised learning aims to learn from data without annotations. However, new problems arise, such as the difficulty in defining appropriate pretext tasks, increased computational demands from multiple stages, and the need for large amounts of unlabeled data. In this thesis, we introduce a learning paradigm that models global and local context for semantic segmentation. The proposed method effectively captures pixel relationships, improving performance in ambiguous regions and better segmenting minority classes through masking. We show that our approach achieves better performance than state-of-the-art single and multi-task learning baselines in both binary and multi-class semantic segmentation tasks, particularly in tackling small, ambiguous regions in medical images and minority class instances in cluttered scenes. Motivated by the intuition that occluded objects are partial inputs, we propose a single-stage, model-agnostic approach for multi-label image recognition. The proposed method learns contextualized representations using a masked branch and models label co-occurrence through label consistency. Experimental results demonstrate the simplicity, applicability, and, more importantly, the competitive performance of our approach against previous state-of-the-art methods, especially in identifying small and occluded objects. Additionally, we propose an efficient unsupervised object localization method that can segment unfamiliar objects in images without the need for additional training, particularly when they are small, reflective, or poorly illuminated. The proposed method learns context-based representations at both the pixel- and shape-level using only a single learnable convolutional layer decoder and a frozen encoder. We demonstrate on six benchmarks datasets the simplicity, efficiency and competitive performance of our approach in both single object discovery and unsupervised salient object detection, outperforming existing methods that require intensive computational resources, extensive training, and large data volumes
Computational Geometry and Online Algorithms
Get PDFThis thesis explores several problems in computational geometry and online algorithms,
focusing on efficient algorithms for geometric optimization and real-time decision-making. We
begin by addressing the offline computational geometry problem of the Maximum Weighted
Convex Polytope (MWCP), followed by two online algorithmic problems: Online Non-Crossing
Matching and Online Interval Scheduling.
In the MWCP problem, the goal is to find a convex polytope within a set of n weighted (positive
and negative) points in R^d that maximizes the total weight of points inside or on the boundary.
We present a new simple algorithm for the two-dimensional case, achieving the same time
complexity of O(n^3) as previous methods. We also prove that MWCP is NP-hard in dimensions
three or higher, even when weights are restricted to +1 and -1, and that in dimensions four or
higher, the problem is NP-hard to approximate within a factor of n^{1/2-epsilon}.
We next focus on the Online Non-Crossing Matching problem, where points arrive sequentially
in the plane and must be irrevocably matched to previously arrived points such that the resulting
matched pairs form non-crossing line segments. We introduce the weighted version of this
problem, aiming to maximize the total weight of matched pairs. We show that deterministic
algorithms can have arbitrarily bad competitive ratios due to adversarial weight assignments. To
address this, we consider weights within the range [1, U] and provide an algorithm with a
competitive ratio of Omega (2^{-2 sqrt{ log U }}), along with an upper bound of O(2^{-sqrt{log
U}}) for any deterministic algorithm. In the setting that allows revoking, we develop an algorithm
that achieves a competitive ratio of approximately 0.28, and prove that no deterministic
algorithm can exceed a competitive ratio of 2/3, even in the unweighted case. Additionally, we
propose a randomized algorithm with a competitive ratio of 1/3, and show that no randomized
algorithm can surpass a competitive ratio of 8/9 in the unweighted case.
We also study the advice complexity of this problem. We establish a lower bound of (n/3) - 1 on
the advice complexity and provide an algorithm that uses approximately 2n bits of advice,
matching log C_n, where C_n is the n th Catalan number. Furthermore, we derive a lower
bound on the advice complexity required to achieve a competitive ratio in (16/17, 1). In the
bichromatic version of this problem, where a set of n blue points are given offline and red points
arrive online to be matched only to blue points, we correct previous errors in the literature and
establish a lower bound of log C_n on the advice complexity. We also present an algorithm that
uses log C_n bits of advice when points arrive in convex position.
In the final part, we consider the Online Interval Selection problem when the interval graph of
the input is a simple path, allowing revoking of previous decisions. We show that under the
random-order model, a deterministic memoryless algorithm achieves a competitive ratio of
approximately 0.78. We also established an upper bound of 3/4 for any deterministic revoking
algorithm on a simple chain in the adversarial model, and a lower bound of n/4 for the advice
complexity of Online Interval Selection, marking the first lower bound for this problem.
Throughout this work, we develop novel algorithmic techniques, establish tight complexity
bounds, and provide new insights into advice complexity for key problems in computational
geometry and online algorithms. These contributions aim to advance the theoretical foundations
of these fields and have potential applications in real-time systems and geometric data analysis
‘Thick description’ of Graphic truths with ethno-graphy: a story of ALK- positive
Get PDFCancer is a condition commonly known as a “life snatcher”. It is frequently assumed that cancer is more likely to affect those with weak bodies and unhealthy lifestyles, as well as older individuals. These and other incorrect assumptions about cancer significantly impact on the experience of a cancer diagnosis. On receiving an ALK-positive cancer diagnosis, I turned to the internet for further information. While some of the content that is available online is accurate, it is often low-resolution, and in some instances, factually incorrect or misleading. The quality of accessible information about cancer impacts on the lives of patients, medical professionals, and caregivers alike. From this understanding, three questions emerged from the study and personal experience: (1) Why is online information about health conditions so generic, and what do researchers acknowledge about the shortcomings of publicly accessible content? (2) What observations can I make by analyzing a handful of websites? (3) Can creating content/ story/ information through "thick description" using ethno-graphy help consolidate an understanding of ALK-positive?
Having experienced the quality of accessible information and with literature review, it was evident that healthcare content needs to be relooked. As a patient, I reflected upon myself and reasoned a way of bringing the experience as research- creation. It is an autoethnography that maps various events in the journey of being a patient in the form of ethno-graphy. The narrative is not centered only around the condition but there are several aspects that are interconnected with each other. It is a cluster of coordination between various systems or groups of professionals who became a part of the day-to-day life to keep the disease monitored. Chapter 1 shows the life around me gradually transcending to a diagnosis that needed further investigation. Chapter 2 and Chapter 3 are not in the scope of the thesis. These two chapters cover various systems coming together to put the condition into control. The thesis also breaks down the method of research-creation
Modelling, Design Optimization and Control of Magnetorheological Dampers for Prosthetic Applications
No full textWalking is a fundamental aspect of human life. However, for individuals who have lost one or
both legs, walking unaided is unfeasible. As a result, extensive research has been conducted on
various types of prosthetic devices. Most of traditional prostheses for above-knee (AK) amputees
rely on passive mechanisms which often cause to an awkward gait due to the lack of actuators and
external controllers. Conversely, active-mode prostheses can adjust the knee joint angle to mimic
a natural leg. However, these devices are costly, heavy, and require a substantial amount of energy
to function. Alternatively, semi-active prostheses with smart materials have demonstrated potential
to achieve stable motion without the need for expensive sensors and actuators.
The present research investigates the modeling and control of a double-ended single coil
magneto-rheological (MR) fluid damper with annular gap for a prosthetic leg application. Based
on Bingham plastic characteristics for MR fluids a quasi-static modeling of MR damper is
developed to estimate the generated damping force of the damper. Analytical magnetic circuit
analysis based on Ampere’s law is conducted to predict the magnetic flux density in the annular
gap of the MR damper. A design optimization problem is formulated to identify the optimal
geometrical design variables of the MR damper to maximize the dynamic range (ratio of the
maximum to minimum damping force of the MR damper) subject to geometrical constraints. The
optimization problem is solved using combined Genetic Algorithm (GA) and Sequential Quadratic
Programming (SQP) methods. Using the optimal parameters, a magneto-static finite element (FE)
model of the magnetic circuit is developed in open-source Finite Element Method Magnetics
(FEMM) software. Results from FE model for the magnetic flux density in the MR damper gap is
compared to those obtained from the analytical approach.
Dynamic modeling of the MR damper based on Bouc-Wen model is also investigated to
simulate the inherent hysteresis phenomenon in MR dampers. The characteristic parameters of
Bouc-Wen model are calculated using simple close-form mathematical expressions. A new
algorithm is developed to automatically extract model parameters from the experimental dataset
allowing faster parameter estimation than conventional optimization approaches.
A dynamic model of the prosthetic leg integrated with MR damper is developed using the
Lagrange method, followed by an analysis of the desired hip and knee angles throughout the gait
cycle. A polynomial approximation is then formulated to estimate the knee angle based on the hip
joint angle. Finally, simulation experiments are conducted to evaluate the performance of three
control methods (feed-forward control, robust inverse dynamic control, and adaptive inverse
dynamic control) in supporting the amputee to achieve a natural gait
Probing microstructure and optical properties of organic semiconductor thin films
Get PDFThe immense success of conjugated organic molecules (COMs) in the multi billiondollar OLED industry has equally raised the interest in conjugated polymers (CPs) for
organic electronics. This is largely due to their flexibility and scalable, solution-based
processing using cost-efficient methods. In most applications, the utility of such organic
semiconductors (OSCs) in electronics relies heavily on tuning their electronic properties
via doping. Unlike COMs, which can easily form highly ordered films via vacuum deposition, CPs are thermally fragile, thus, typically requiring solution processing. This
often results in films with less-defined microstructures influenced by various experimental conditions. Further, CPs are susceptible to structural and chemical defects that reduce
their conjugation length. This thesis employs the unconventional method of high-vacuum
electrospray deposition (HV-ESD) to create poly(3-hexylthiophene) (P3HT) films under
minimal solvent influence elsewise only applicable for COMs. Conventional solutionprocessed P3HT is typically semi-crystalline, comprising crystalline and amorphous regions as shown by X-ray diffraction. In contrast, it is found in this work that HV-ESD
yields highly amorphous films. This technique allows therefore studying CP films without the complexity of mixed crystalline and amorphous regions, which is important in
analyzing properties like the doping of P3HT, where these regions respond differently
to doping and information gathered thereon is generally convoluted. Further, this thesis focuses on COMs exploring the optical and structural properties of blends created
by the rod-like molecules α-sexithiophene (6T) and para-sexiphenyl (6P), which are prototypical OSCs. Vacuum co-deposited films yield a mixed crystal structure, confirmed
by grazing-incidence X-ray diffraction, and unique optical properties. The blends show
green emission alongside the characteristic red and blue emissions of pure 6T and 6P explained by J- and H-type aggregation. Additionally, this work includes studies addressing
the molecular doping of oligomers to examine transitions between distinct doping phenomenologies based on the conjugation length, and the yet to be fully understood doping
of P3HT by Lewis acids as alternative dopants. These studies are nearing publication.
Overall, by commissioning HV-ESD as a tool to deliberately establish amorphous OSC
films, and by assessing OSC blends and molecularly doped OSCs in thin films, this thesis
contributes a broad view on the interplay between structural and optoelectronic properties of this important material class