Master Theses & Projects (FSCI)

Permanent URI for this collectionhttps://hdl.handle.net/10155/387

Browse

Now showing 1 - 20 of 344

A few-shot learning method for single-object visual anomaly detection
(2024-04-01) Ejaz, Neha; Qureshi, Faisal
We propose a few-shot learning method for visually inspecting single objects in an industrial setting. The proposed method is able to identify whether or not an object is defective by comparing its visual appearance with a small set of images of the “working” object, i.e., the object that passes the visual inspection. The method does not require images of defective objects. Furthermore, the method does not need to be “trained” when used to inspect new, previously unseen, objects. This suggests that the method can be easily deployed in industrial settings. We have evaluated the method on three visual anomaly detection benchmarks—1) MVTec, 2) MPDD, and 3) VisA. On the first two datasets the proposed method achieves performance that is comparable to state-ofthe- art methods that require access to object-specific training data. Model performance on VisA is poor; however, it is to be noted that the model was never trained on VisA dataset. We also show that the proposed model boasts fast inference times, which is a plus for industry applications. This project is funded in part by Axiom Plastics Inc., and we have evaluated the proposed method on a proprietary dataset provided by Axiom. The results confirm that the proposed method is well-suited for single-object visual anomaly detection in industry settings.
A study of meta-learning methods on the problem of video matting
(2024-04-01) Tabaraki, Negin; Qureshi, Faisal; Pu, Ken
Applying image matting techniques directly to video matting presents challenges, primarily due to the complex temporal dynamics inherent in video data. In this work, we studied two Meta Learning approaches—Boosting with Adapters (BwA) and Boosting using Ensemble (BuE)—to tackle the task of video matting using pre-trained image matting models. BwA refines (image matting) alpha mattes by fine tuning pre-trained segmentation models, which we refer to as adapters, using video frames. BuE, additionally, combines multiple fine-tuned adapters using a convolutional neural network. We introduced a meta-learning architecture that incorporates both adapters and ensemble boosting through an iterative process of expert selection and fine tuning. Based on our evaluation on benchmarks based on a standard video matting dataset (VideoMatte240K), we confirm that the proposed scheme improves the performance of image matting models on the task of video matting. In addition, the proposed approach also improves the performance of VMFormer (c. 2022), a recent video matting method.
A targeted reverse mapping machine learning approach for non-dominated solutions in multi-objective optimization
(2024-08-01) Kermani Poor, Masoud; Ibrahimi, Mehran; Rahnamayan, Shahryar
Multi-objective optimization problems aim to identify solutions that maximize or minimize conflicting objectives. Population-based multi-objective algorithms, inspired by biological populations, are effective but often provide limited solutions within the decisionmakers’ region of interest (ROI) on the Pareto front. Recent advancements in machine learning have shown promise in generating solutions, yet they suffer from a lack of control and require knowledge of objective function attributes. This study proposes a framework using Gaussian process regression and artificial neural networks to generate innovative solutions in the ROI. By employing diverse sampling techniques and integrating long term memory, the framework can produce more than twice as many solutions in the ROI, as demonstrated in experiments with real-world problems and various benchmark functions.
Achieving real-time video summarization on commodity hardware
(2018-04-01) Taylor, Wesley; Qureshi, Faisal
We present a system for automatic video summarization which is able to operate in real-time on commodity hardware. This is achieved by performing segmentation to divide a video into a series of small video clips, which are further reduced or eliminated with the assistance of highly efficient low-level features. A numerical score is then assigned to each segment by our model trained using a set of highperformance hand-crafted features. Finally, segments are selected based on their score to generate a final video summary. On our benchmark dataset, we achieve results competitive to other methods. In cases where our accuracy is lower than competitive methods, we achieve significantly higher performance. We additionally present methods for generating additional summaries almost instantly, and for learning user preferences over time—two processes which are often overlooked in work on video summarization, but essential for real-world use
An adaptive crowdsourced investigation of word abbreviation techniques for text visualizations
(2017-04-01) Shimabukuro, Mariana Akemi; Collins, Christopher
A known problem in information visualization labeling is when the text is too long to fit in the label space. There are some common known techniques used in order to solve this problem like setting a very small font size. On the other hand, sometimes the font size is so small that the text can be difficult to read. Wrapping sentences, dropping letters and text truncation can also be used. However, there is no research on how these techniques affect the legibility and readability of the visualization. In other words, we don’t know whether or not applying these techniques is the best way to tackle this issue. This thesis describes the design and implementation of a crowdsourced study that uses a recommendation system to narrow down abbreviations created by participants allowing us to efficiently collect and test the data in the same session. The study design also aims to investigate the effect of semantic context on the abbreviation that the participants create and the ability to decode them. Finally, based on the study data analysis we present a new technique to automatically make words as short as they need to be to maintain text legibility and readability.
Adaptive learning game to personalize occupational health and safety training
(2017-11-01) Chodan, Cameron; Mirza-Babaei, Pejman; Sankaranarayanan, Karthik
In 2012, the Association of Workers’ Compensation Boards of Canada recorded approximately a quarter-million workplace injuries, a staggering figure keeping in mind that some incidents go undocumented. It is important that organizations continue make Occupational Health and Safety (OHS) one of their top priorities. In this thesis, we discuss an implementation of an adaptive personalized learning support system within a game that is centered on health and safety training to promote the understanding of health and safety material. The design of the game incorporates a feedback loop that constantly evaluates the player’s performance while they complete learning challenges. As the players proceed within the game's environment their profile is constantly updated thus providing an insight into their strengths and weaknesses. The game is designed to adjust the challenges given to the player to focus on improving the player’s underperforming skills. The goal of this game is to promote health and safety in small and medium enterprises. Through this game we created a motivational designed application that helps to teach targeted health and safety information to the workers. The game was made in collaboration with the Public Services Health and Safety Association based in Toronto. The game aims to better the player’s health and safety performance in the Organizational Performance Metric and hone their underlying health and safety skills.
Aerodynamic and thermal analysis of a heat source at the underside of a passenger vehicle
(2014-12-01) Khasow, Rocky; Agelin-Chaab, Martin
The first part of this thesis involves full experimental and numerical studies to understand the effects of cross-winds on the automotive underbody aero-thermal phenomena using a 2005 Chevrolet Aveo5 with a heat source affixed to it to create a baseline. The results show that irrespective of the yaw angle used, only temperatures in the vicinity of the heat source increased. The rear suspension also deflected the airflow preventing heat transfer. The second part of this thesis investigated using a diffuser to improve hybrid electric battery pack cooling. It was found that the diffuser led to more consistent temperatures on the diffuser surface, suggesting the same for the battery.
Aiding the experts: how artificial intelligence can augment expert evaluation with PathOS+
(2022-07-01) Nova, Atiya Nowshin; Mirza-Babaei, Pejman
Within games user research (GUR) predictive methods like expert evaluation are good for getting easy insights on a game in development but may not accurately reflect the player experience. On the other hand, experimental methods like playtesting can accurately capture the player experience but are time consuming and resource intensive. AI agents have been able to mitigate the issues of playtesting, and the data generated from these agents can supplement expert evaluation. To that end we introduce PathOS+. This tool allows the simulations of agents and has features that allows users to conduct their evaluations in the same place as the game, and then export their findings. We ran a study to evaluate how PathOS+ fares as an expert evaluation tool with participants of varying levels of UR experience. The results show that it is viable to use AI to identify design problems and offer more validity to expert evaluation.
An investigation of the environmental factors that affect water quality and the occurrence of harmful cyanobacteria in stormwater management ponds
(2024-04-01) Horton, Kaitlyn; Kirkwood, Andrea
Stormwater management ponds (SWMP) have an important role in flood mitigation and basic water quality treatment via sedimentation. As aquatic ecosystems, less is known about their role as habitats for aquatic organisms and their potential to transform pollutants. This study focused on twelve SWMP in Oshawa, Ontario, Canada to assess the impacts of flow conditions and pond characteristics on SWMP water quality treatment and algal growth, including toxic cyanobacteria. Net release of total phosphorus (TP) and net retention of total nitrogen (TN) by SWMP were observed. Flow conditions had little affect on the overall functioning of the study SWMP. Large pond designs and recent dredging were observed to positively influence the reduction of total suspended solids (TSS), TN, and TP net release from SWMP. Algae and cyanobacteria were observed to be generally N-limited. The presence of the cyanobacterial toxin gene mcyE was positively associated with chloride and heterocyst-forming cyanobacteria.
Analysing atomic structure in nano-scale imagery
(2017-01-01) Nemirovsky, David M.; Qureshi, Faisal Z.; Tamblyn, Isaac
This thesis presents methods for both detecting atoms and reconstructing single layer molecular geometry depicted within high resolution nanoscale imagery. Nanoscale imaging is a common method of analysing molecular structures produced through self-assembly. The second derivatives of the nanoscale images are then computed and used to aid in analysis of the depicted molecules. Peaks from the derivative images are used as the locations of the atoms and using those locations we derive the molecular geometry. In order to efficiently determine the performance of our proposed methods, a system to synthesize HRTEM imagery and ground truth data was developed. We demonstrate and compare the effectiveness of our system to other proposed systems using both experimentally produced nanoscale images and the synthetic data that has been produced.
Analysis of peptide production by Lactobacillus species and evaluation of their antihypertensive and immunomodulatory activities
(2015-08-01) Adams, Christina; Strap, Janice L.
Cardiovascular disease (CVD) is the leading cause of death among Canadian adults. Research has demonstrated an inverse relationship between the consumption of fermented dairy products and a decreased risk of CVD due to lactic acid bacteria used in the fermentation process which liberate small bioactive peptides from larger milk proteins (eg. casein). We observed that supplementation with 0.1% casein significantly increased the growth rate of L. helveticus R0389 and L. rhamnosus R0011 and increased the ACE-inhibitory activity of their secreted peptide fractions. Peptide-containing supernatants of L. rhamnosus R0011 show comparable ACE inhibition to known antihypertensive peptides, VPP and IPP. Supernatants of milk ferments induced the production of the regulatory cytokine, IL-10, by THP-1 monocytes. Novel antihypertensive and immunomodulatory activities of individually synthesized peptides were also reported. By investigating the relationship between these bioactive properties, we can improve upon the use of probiotic organisms to confer maximal health benefits to Canadians.
Anodic alcohol fuel cell reactions at platinum active sites on doped metal oxide supports
(2022-12-01) Black-Araujo, Keenan; Easton, Brad
Fuel cells produce electrical energy via chemical reactions, which take place on the carbon supported platinum catalyst (Pt/C). As attractive as these devices may seem, their practicality is often limited by the functionality of the carbon support, which can be prone to corrosion. To limit the corrosion associated with a carbon-based catalyst, metal oxide supports are an area of interest as they display high stability in the harsh conditions associated with fuel cell operations. In this work, the employment of various doped metal oxide supports; silicon & molybdenum doped titanium oxide (TOMS), silicon doped titanium oxide (TOS), and silicon doped niobium oxide (NbOS) were investigated to replace carbon as a supporting material in alcohol fuel cells. In an alkaline environment, Pt/TOMS displayed faster kinetics and less poisoning during the methanol and ethanol oxidation reactions compared to the commercial Pt/C. When compared in an acidic ethanol and methanol oxidation, Pt/TOS and Pt/NbOS exhibit smaller activation energies than Pt/C, however, only Pt/NbOS produces smaller charge transfer resistances and higher peak oxidation currents. These studies highlight the promise of replacing the carbonaceous material presently found in fuel cells with metal oxide, or doped metal oxide, materials.
The anti-estrogenic and liver metabolic effects of DHAA in rainbow trout (oncorhynchus mykiss)
(2011-08-01) Pandelides, Zacharias; Holdway, Douglas
Recent studies have shown that dehydroabietic acid (DHAA), a resin acid present in pulp and paper mills, may have anti-estrogenic effects in fish. A chronic-exposure toxicity experiment using immature rainbow trout (Oncorhynchus mykiss) was conducted in order to assess the endocrine disrupting and liver metabolic effects of the wood extractives DHAA and β- sitosterol (BS) regularly present in pulp and paper mills and the model estrogen 17β-estradiol (E2). It was found that exposure to 5 ppm of E2 significantly increased hepatosomatic index (HSI), vitellogenin (VTG) and plasma sorbitol dehydrogenase (SDH). This effect was reduced by mixing E2 with DHAA, indicating that DHAA does not cause its anti-estrogenic effects indirectly due to liver damage. Exposure to 5 ppm of DHAA caused a significant increase in liver citrate synthase (CS), and liver ethoxyresorufin-O-deethylase (EROD) activity after 7 days, however, the fish recovered by 28 days. This study also determined the effect of 14 different pulp and paper mill effluent extracts on liver enzyme metabolism through alterations in the activity of liver lactate dehydrogenase activity (LDH) and CS. This activity varied greatly between mills but most showed an induction of CS after 28 days exposure through i.p. injection. The results of the study indicate that DHAA may alter energy metabolism as well as cause anti-estrogenic effects in female juvenile rainbow trout.
Applications and numerical investigation of differential-algebraic equations
(2010-05-01) Milton, David Ian Murray; Berg, Peter; Staley, Mark
Differential-algebraic equations (DAEs) result in many areas of science and engineer- ing. In this thesis, numerical methods for solving DAEs are compared for two prob- lems, energy-economic models and traffic flow models. An energy-economic model is presented based on the Hubbert model of oil production and is extended to include economic factors for the first time. Using numerical methods to simulate the DAE model, the resulting graphs break the symmetry of the traditional Hubbert curve. For the traffic flow models, a numerical method is developed to solve the steady-state flow pattern including the linearly unstable regime, i.e. solutions which cannot be found with an initial value solver.
Arthropod successionin Whitehorse, Yukon Territory and compared development of protophormia terraenovae (R. -D.) from Beringia and the Great Lakes Region
(2012-07-01) Bygarski, Katherine; LeBlanc, Helene
Forensic medicocriminal entomology is used in the estimation of post-mortem intervals in death investigations, by means of arthropod succession patterns and the development rates of individual insect species. The purpose of this research was to determine arthropod succession patterns in Whitehorse, Yukon Territory, and compare the development rates of the dominant blowfly species (Protophormia terraenovae R.-D.) to another population collected in Oshawa, Ontario. Decomposition in Whitehorse occurred at a much slower rate than is expected for the summer season, and the singularly dominant blowfly species is not considered dominant or a primary colonizer in more southern regions. Development rates of P. terraenovae were determined for both fluctuating and two constant temperatures. Under natural fluctuating conditions, there was no significant difference in growth rate between studied biotypes. Results at repeated 10°C conditions varied, though neither biotype completed development indicating the published minimum development thresholds for this species are underestimated.
Artificial intelligence agent for assessing game difficulty
(2024-04-01) Sansalone, Stevie C. F.; Mirza-Babaei, Pejman
Balancing game difficulty is a key element of developing games which engage players without causing frustration. Difficulty balancing is an iterative process and game developers frequently rely on user tests to guide them, but the cost of user testing can be prohibitive, particularly for independent developers. In response to this, I seek to determine whether AI can be used to test the difficulty of developing games before setting up a full playtest. For this purpose, I developed ARTemiS, a prototype tool for AI-based playtesting focused on game difficulty arising from precision input tasks such as combat. I then performed a user study with 10 participants who used the tool to assess and modify three demo levels. This research seeks to assess the applicability and utility of AI-based playtesting for difficulty balancing and lay a foundation for the development of tools for assessing other aspects of difficulty with AI.
Assessing the influence of geological and land-use gradients on zooplankton and phytoplankton biodiversity in the land between ecotone
(2021-11-01) Hassal, Emily; Kirkwood, Andrea
The Land Between (TLB) is an ecotone in south-central Ontario that represents a transition from limestone-dominated bedrock to granite-dominated bedrock, creating notable calcium and land-use gradients across this unique geographic region. The goal of this study was to investigate water quality and plankton community patterns in lakes across TLB as a function of geological and spatial drivers. Lake water quality profiles were differentiated based on calcium concentrations, and increased watershed land-use was related to increased lake nutrient levels. Plankton communities were not spatially structured. Phytoplankton communities were influenced by environmental drivers including nutrients and water temperature, as well as the inferred effect of zooplankton grazing. Zooplankton communities were influenced by environmental drivers including calcium concentration. Additionally, plankton communities appeared to be influenced by species-specific and trophic interactions. Overall, my research has provided insight into the important drivers of lake water quality and plankton communities in TLB.
Assessing the memorability of familiar vocabulary for system assigned passphrases
(2021-08-01) Jagadeesh, Noopa; Vargas Martin, Miguel
Text-based secrets are still the most commonly used authentication mechanism in information systems. Initially introduced as more secure authentication keys that people could recall, passphrases are tokens consisting of multiple words. However, when left to the choice of users, they tend to choose predictable natural language patterns in passphrases, resulting in vulnerability to guessing attacks. System-assigned authentication keys can be guaranteed to be secure, but this comes at a cost to memorability. In this study we investigate the memorability of system-assigned passphrases from a familiar vocabulary to the user. The passphrases are generated with the Generative Pre-trained Transformer 2 (GPT-2) model trained on the familiar vocabulary and are readable, pronounceable, sentence like passphrases resembling natural English sentences. Contrary to expectations, following a spaced repetition schedule, passphrases as natural English sentences, based on familiar vocabulary performed similarly to system-assigned passphrases based on random common words.
Assessing the use of DNA expert evidence, by justice system participants, in Ontario criminal courts
(2017-08-01) Cohen, Dawn; Hageman, Cecilia
What happens to forensic DNA opinion evidence when the expert witness is not present in the courtroom? Research addressing this issue has largely been focused on polling lawyers regarding their perceptions of DNA evidence, as well as studies of juror understanding of DNA expert evidence in real and mock court situations. This thesis attempted to address the question in a different way, by analyzing transcripts of expert DNA evidence, opening & closing addresses, and judges’ instructions to juries, for cases that have passed through the Ontario criminal courts within the past fifteen years. This project is the first assessment of Canadian criminal court case transcripts, comparing expert DNA evidence with the (largely) non-scientist attorneys’ and judges’ inferred understanding and use of that evidence. Trial transcripts from cases involving DNA expert evidence were located by keyword searching Ontario Court of Appeal decisions via the Canadian Legal Information Institute (CanLii) public online database. This research question was approached from a social science methodology, making use of both qualitative and quantitative analyses. Quantitative analysis was conceptualized first, as a survey that was developed to track topics of interest in Interval Ratio and Nominal variable form. Qualitative Data Analysis (QDA) Miner 4 Lite™ was used to code sections of these transcripts and complete the survey. Coded sections involved random match probabilities (RMPs), likelihood ratios (LR), mitochondrial & Y-STR lineage confidence intervals, as well as body fluid attribution statements by attorneys and judges. These transcript excerpts were compared to each case’s respective DNA expert testimony. This enabled the application of qualitative analysis of question and response exposition within the expert testimonies. The survey data were inputted into IBM Statistical Package for the Social Sciences (SPSS) Statistics 24 for pattern analysis and descriptive statistics. For example, the sets of cases studied (N=32), contained 101 autosomal random match probability statements provided by DNA experts. Many times these RMPs did not enter into the crown summations (only 48.5%), defence summations (only 31.7%) or judges’ instructions (only 57.4%). When attorneys and judges did discuss and review the DNA statistical evidence, mistakes and misstatements occurred in the majority of instances – these mistakes included statistical fallacies and numerical misstatements. This research suggests a lacunae of knowledge with respect to the meaning of DNA evidence, and in particular, the correct understanding and communication of the RMP estimate of statistical weight of DNA profile comparison evidence. Further research is recommended, to address the use of transfer and persistence expert testimony, as well as testimony regarding complex mixture profile interpretations and comparisons
Assessment of wastewater algae for use in biofuel production
(2013-08-01) Stemmler, Kevin; Kirkwood, Andrea
Although new technologies have allowed the attainment of previously untapped fossil fuels these practises are unsustainable and harmful to the environment. Part of the solution to ease the fuel burden is through renewable fuels derived from microalgae as they are a carbon neutral source of fuel. The aim of this research was to assess if algae derived from municipal wastewater sources could be potential biofuel feedstocks by assessing their growth and fatty acid accumulation. When comparing wastewater derived algae to culture collection strains there was no significant difference (p>0.05) in terms of growth rates under photoautotrophic and mixotrophic conditions. The strain Botrydiopsis B2N under mixotrophic (14mM glucose) possessed the highest growth rate (2.7x104 cells•L-1•day-1) of all the strains tested under the various conditions. It was noted that under mixotrophic growth (14mM glucose) non-axenic algae accumulated significantly higher concentrations of neutral lipids compared to the same algal strains under axenic conditions. The result of which is thought to be caused by bacteria creating a nutrient deprived media causing the algae to become stressed and accumulate fatty acids. Under mixotrophic growth (14mM glucose and 3mM acetate) the organic carbon in the media appeared to shift the composition of fatty acids in most cases increasing the likelihood of an even blend of saturated to unsaturated fatty acids.

Browse

Browsing Master Theses & Projects (FSCI) by Title