Management Science and Engineering

Permanent URI for this collectionhttps://uwspace.uwaterloo.ca/handle/10012/9910

This is the collection for the University of Waterloo's Department of Management Science and Engineering.

Research outputs are organized by type (eg. Master Thesis, Article, Conference Paper).

Waterloo faculty, students, and staff can contact us or visit the UWSpace guide to learn more about depositing their research.

Browse

Now showing 1 - 20 of 367

The use of data analytics to analyze various behaviors during the COVID- 19 pandemic
(University of Waterloo, 2024-05-24) AbdulHussein, Ali; Dimitrov, Stanko; Cozzarin, Brian
This dissertation outlines findings based on three journal manuscripts. The publications have a common theme: the use of data analytics techniques, economic theory, and knowledge of online shopping to analyze and assess change in behavior after the COVID-19 pandemic. I employed a variety of empirical analysis tools, including different regression methods, statistical analysis, and validity testing. I also utilized well-established theory to add another lens to my findings. Such frameworks include the Transaction Cost Economics (TCE) and the Technology Acceptance Model (TAM). Furthermore, I applied my tools to two domains to further widen my knowledge scope: e-commerce and public health. The first manuscript offered insight into consumer behavior after the COVID-19 pandemic. It analyzed the change in online shopping activity in 12 different product categories and offered an empirical association between it and various demographic factors. The second manuscript builds the first with more focus on online grocery shopping. As a result, the findings offered a managerial perspective on how various customer segments changed their shopping behaviors online, providing insight to guide marketing and merchandising efforts. The third paper presented an association between demographic and occupational factors with worsened mental health conditions of healthcare workers (HCWs) after the pandemic. The finding presented an insight into how different HCW groups reacted to the pandemic and hence aid in providing more effective mental health programming targeting specific groups in future events.
Unmanned Aerial Vehicle Traffic Network Design with Risk Mitigation
(University of Waterloo, 2024-02-01) Nicholson, Jeremy; Gzara, Fatma
As unmanned aerial vehicle (UAV) technology becomes more robust and widespread, more and more retail companies are seeing UAVs as a suitable alternative to ground-based transportation to deliver their packages. As a result, there has been an abundance of OR research focused on UAV utilization for last-mile delivery. Due to the size and mobility of UAVs, most of this research considers UAV movement within a shortest path or Euclidean shortest path context. While this may be plausible if drone usage remains sparse, this framework will not be possible as drone utilization ramps up to the levels required to satisfy the levels of package demand expected in the coming decades. Furthermore, none of this prior research (to our knowledge) suggests using risk inherent with UAV travel to influence their proposals from a logistical and/or modelling perspective. As a solution to this problem, our industry partner AirMatrix proposes that UAV travel be restricted to transportation networks situated above the streets of population centres. We propose a bi-objective network selection model for drone delivery which minimizes risk while maximizing the amount of satisfied demand subject to budgetary constraints. We discuss the factors that affect UAV risk and what metrics can be used to effectively reduce those factors from a modelling perspective. We propose a two-stage stochastic variant of the model and additional problem requirements to reflect practical operational requirements and design goals. Using sample average approximation, we show that a deterministic solution is effectively as good as an associated stochastic solution. We conduct testing on a region of suburban Miami to evaluate how different risk objectives perform with respect to network, path, arc, and performance metrics.
Data-driven Models for Inferring the Patient Scheduling Policies via Inverse Reinforcement Learning
(University of Waterloo, 2024-01-23) Moradi, Parham; Abouee Mehrizi, Hossein
In this work, we study multi-class patient scheduling with stochastic daily patient arrivals. Different classes of patients are characterized by different service times, waiting cost parameters, and rejection cost parameters. Our primary objective is to infer the policy used by the decision-makers, who schedule patients over a finite time horizon, based on their historical decisions. To achieve this, we first develop a mathematical model that captures the complexities of patient scheduling and is representative of the problem that decision-makers may consider to scheduling patients. Then, we utilize the Riccati and Hamiltonian approaches to estimate the cost parameters that have influenced the scheduling decisions made by the decision-maker. The Riccati approach begins by estimating the expert's policy, which is then used to determine the cost parameters. Conversely, the Hamiltonian approach derives the cost parameters through the optimality conditions of a path trajectory without needing to estimate the expert's policy. Using a simulation model, we demonstrate the efficiency and robustness of the proposed methods. Furthermore, we apply Riccati and Hamiltonian approaches to MRI data from two hospitals to estimate the cost parameters used in their scheduling decisions. Utilizing the estimated cost parameters, we analyze the root causes of the observed outcomes and examine the impact of these underlying factors on the scheduling process. Finally, through counterfactual analysis, we propose two alternative scheduling policies that reduce the total cost, even with the original cost parameters used by the decision-makers.
The Impacts of Climate Change via Robust Optimization: Two Applications in Land Investment and Electricity Storage Systems
(University of Waterloo, 2024-01-03) WU, ZHENGGAO; Dimitrov, Stanko; Pavlin, Michael
Effectively adapting to a changing climate involves making appropriate operational decisions based on long-term climate forecasts. This dissertation presents a comprehensive framework that combines climate data, regression models, and robust optimization models to examine the decision-making process for adapting to climate change over long time horizons. The research includes two projects: one focuses on studying land investment decisions, and the other investigates the operations of electricity storage systems, both considering the impacts of climate change. Project 1: Climate change affects agricultural inputs, like temperature and precipitation, and further affects the economic output of farmland. In this study, we focus on formulating effective policies to aid various stakeholders, including investors and farmers, in adapting to the climate-induced impacts on farmland investment in the Mississippi River Basin (MRB) by using well-known climate models. Each climate model generates a unique climate forecast, and based on these forecasts, we compute a range of farmland values for the MRB. Utilizing these ranges, we apply a robust optimization model to study the optimal investment policies under varying levels of conservatism, representing the extent to which farmland assets are constrained to adopt worst-case values. We show that the optimization model can be linearized and can scale to long time frames, about 50-plus years, and sets of assets. The case study of investment in the MRB covers the years 2023-2090 and uses trajectories of land values determined for each climate scenario using a regression model. Our empirical study shows that there is a disagreement between popular climate forecasts that influence land investment and may affect the most profitable land investments. Project 2: The effects of climate change on energy markets are diverse, encompassing changes in demand patterns and supply dynamics, particularly concerning the increasing penetration of renewable energy. These changes impact the dynamics of energy supply from renewable sources, such as wind and solar, leading to increased intermittency. Battery energy storage systems (BESSs) present a promising solution to effectively manage this intermittency from renewable energy sources. However, their profitability and incentive to participate in markets under climate change are susceptible to both the magnitude and frequency of price variation. This project investigates the impact of climate change on a BESS operating in a North American deregulated electricity market. We propose a robust optimization model to determine the operating policy of a BESS over 80 years (from 2021 to 2100) under different climate projections. We reformulate the robust optimization model to an equivalent linear program that allows us to numerically explore the operations of the BESS over the time horizon. Our empirical study analyzes the optimal arbitrage operations of the BESS in the Midcontinent Independent System Operator market in the United States, using the proposed robust model and trajectories of electricity prices determined for each climate scenario by a regression model. Additionally, we introduce a downscaling method to adjust climate scenarios to the desired resolutions for predicting electricity prices through the regression model. The results of the robust model reveal significant variations in the operating incomes of the BESS across different geographical locations and climate scenarios, highlighting the need for tailored strategies adapting to climate-induced variations in energy markets. The findings from both projects underscore the critical significance of considering a wide range of climate scenarios, encompassing detailed temporal and spatial data when assessing climate adaptation decisions.
Blockchain Recommender Systems using Blockchain Data
(University of Waterloo, 2023-12-18) Khatiri, Sean; Golab, lukasz
Blockchain systems allow a network of pseudo-anonymous users (identified only by their public key) to maintain a secure transaction ledger in a decentralized manner. Transactions are executed and recorded on the ledger by programs called smart contracts. Decentralized applications (dApps) can be built on top of blockchains, for tasks such as exchanging cryptocurrencies and other digital assets, without the need for trusted third parties such as banks. As is the case with traditional Web applications, personalization is key to user acquisition and retention in decentralized systems. We therefore ask the following question in this thesis: how can we build effective blockchain recommender systems? To answer this question, we turn to collaborative filtering, a popular recommendation approach that captures similarities among users in terms of their transaction histories. For example, if two users liked movies a, b, c, and d, and the first user additionally liked movie e, then collaborative filtering may suggest movie e to the second user. The main technical challenge we address is how to map smart contract code to the underlying items or concepts that may be recommended, e.g., a smart contract that facilitates an in-game purchase using Bitcoin may map to the “gaming” concept. Using this mapping and real-world data from the Ethereum network, which is the largest smart-contract-enabled platform, we test two collaborative filtering systems: a simple and fast Matrix Factorization (MF) algorithm and a more complex one based on Graph Neural Networks (GNN). Our empirical results show that GNN outputs more effective recommendations, at the expense of latency. We conclude with an overview of a blockchain-native implementation of our framework as a decentralized recommendation service, and we discuss the corresponding practical challenges such as incentive mechanisms (tokenomics).
Optimal Order Batching for Automated Warehouse Picking
(University of Waterloo, 2023-09-18) Kucuksari, Zeynep; Elhedhli, Samir
With the unexpected increase in demand and the need to minimize human interaction during the Covid-19 pandemic, companies have been forced to accelerate the transition from traditional to robotic mobile fulfillment systems. The key to a successful warehouse management system, whether traditional or automated, is an efficient order-picking process. In this study, we focus on the order batching problem, where items and orders are grouped into batches for simultaneous picking in automated warehouses that use autonomous picking carts. We propose five different mathematical models, including a generalized quadratic assignment model. We focus on the latter as it provides the best results and propose a Lagrangian relaxation to obtain lower bounds and an iterative Simulated Annealing (SA) algorithm that generates an initial solution using a K-means clustering algorithm. We carry out testing using an open-source dataset to assess the iterative SA algorithm in minimizing congestion and travel distance in an automated warehouse. We find that it finds solutions of good quality as measured by Lagrangian relaxation and is capable of solving large realistic instances. The solutions successfully minimize travel distance and reduce congestion by limiting path intersections.
A Tractable Approach To Inverse Optimization Under Euclidean Norm
(University of Waterloo, 2023-09-11) Ebrahimkhani, Sara; Mahmoudzadeh, Houra; Abouee Mehrizi, Hossein
The conventional optimization assumes that the problem and its parameters are known, and it utilizes this information to determine the optimal solution. Inverse optimization works in reverse by determining different parameters of an optimization model such that a given dataset of observed decisions from the past becomes optimal for the model. The parameters imputed through inverse optimization can be in the objective function and/or the constraints of the model. When inferring the constraint parameters, the choice of objective for the inverse optimization problem can result in different inverse optimal solutions. However, it is unclear which solution provides the best fit to the data. In this study, a goodness-of-fit measure is first introduced to evaluate the fit between the model and data and determine the quality of the inferred feasible region based on the distances of data points from its boundary. Next, employing this measure as the objective function, a multi-point inverse optimization problem under the Euclidean norm is proposed to infer the feasible region of a linear optimization model. Given the nonlinear nature of the Euclidean norm, a relaxation technique using the non-smooth L1 penalty function is proposed for the inverse optimization problem. This reformulates the non-convex mixed-integer quadratically-constrained programming problem into a mixed-integer quadratic programming problem which is more tractable. Then, an exact heuristic method and a greedy heuristic method are introduced to alleviate the computational challenges of the problem. Finally, two application examples to illustrate the practicality and effectiveness of our proposed model and solution approach are presented. In the first application, our model determines the implicit criteria based on which a patient is identified as an outpatient without requiring hospital supervision. The second application focuses on improving the recommended diets by uncovering hidden preferences and suggesting meal plans based on individuals' past food choices.
Open-Ended Problem Solving in Groups
(University of Waterloo, 2023-09-08) Alattas, Hanan; Duimering, Rob
Previous experimental research on problem-solving has predominantly investigated well-structured problems with predefined solutions. Studies of ill-structured, open-ended problem-solving have primarily employed observational and case study methods. This study used a controlled experiment in which groups solved ill-structured categorization problems to investigate effects of problem open-endedness on problem-solving behaviors and solution outcomes. The experimental design enables precise measurement and tracking of open-ended problem-solving behaviors. In the experiment, N=48 four-person groups solved three categorization problems, in which they grouped 16 randomly selected pictures into 4 categories of 4 pictures each. Task goals and participant beliefs were varied to create three levels of problem open-endedness. In two tasks, participants grouped pictures based on similarity, and their open-endedness beliefs were altered based on instructions suggesting either that a single best solution identified by experts should be found (“Expert”; least open-ended), or that multiple solutions were available, and a “good” solution should be found (“Good”; more open-ended). In a third task, participants grouped pictures by creating 4 simple stories involving the items (“Story”; most open-ended). The experiment investigated effects of the degree of problem open-endedness on several indicators of problem-solving behavior and properties of the solution, including problem-solving difficulty, the variability of solutions produced by different problem-solving groups, the influence of initial conditions on solutions (path dependency), the strength of concept association in solutions, structural moves toward solutions, and the variability of problem-solving search behavior. ANOVA results across the three levels of open-endedness confirmed hypothesized negative effects of problem open-endedness on task difficulty and variability in problem-solving behavior, as well as positive effects on solution variability, path dependency, and the strength of solution association. The results also provided evidence that solutions to open-ended problems are non-random. Post-hoc pairwise comparisons between open-endedness levels partially supported our hypotheses. Differences between the similarity and story tasks strongly supported hypotheses; however, differences between the two (least open-ended vs. more open-ended) similarity tasks were mainly non-significant although the distribution means varied in the predicted directions. Regarding structural progress towards a solution, participants in the least open-ended “Expert” condition first formed categories based on the strongest associations between items, then moved to progressively weaker associations. This effect was less prominent in the more open-ended “Good” condition and absent in the most open-ended “Story” condition. A verbal protocol analysis conducted on nine experimental tasks provided further insights into the problem-solving process across three conditions. A prominent pattern of behavior observed in all conditions was iterative conflict recognition and resolution until groups reached a satisfactory solution. In the “Expert” condition, groups exhibited more conflict recognition and resolution iterations, more emphasis on the logic behind requested picture exchanges and more resistance to accepting proposed resolutions, compared to the Good and Story conditions. Individual group members tended to develop partial solutions independently and simultaneously in the Story condition, whereas partial solutions were developed collectively in a sequential manner in the similarity conditions.
Social Media Influencers- A Review of Operations Management Literature
(University of Waterloo, 2023-08-29) Matthias, Mikhailla; Dimitrov, Stan; Mantin, Binyamin
This literature review provides a comprehensive survey of research on Social Media Influencers (SMIs) across the fields of SMIs in marketing, seeding strategies, influence maximization and applications of SMIs in society. Specifically, we focus on examining the methods employed by researchers to reach their conclusions. Through our analysis, we identify opportunities for future research that align with emerging areas and unexplored territories related to theory, context, and methodology. This approach offers a fresh perspective on existing research, paving the way for more effective and impactful studies in the future. Additionally, gaining a deeper understanding of the underlying principles and methodologies of these concepts enables more informed decision-making when implementing these strategies
Models of Deterministic and Stochastic Comparison: Two Studies in Applied Operations Research
(University of Waterloo, 2023-08-29) Burgess, Kiefer Joe; Dimitrov, Stanko; He, Qi-Ming
This dissertation includes two essays on applications of management science methods to modelling service systems and developing novel improvements to sports team ranking systems. The first essay proposes a novel approach to modelling changes in business procedures that have neither explicitly positive nor explicitly negative effects on operational performance, but are changes to operating rules; we call these procedure changes Operational Protocol Modifications (OPMs). Our approach is to model these OPMs via distributional censoring. Using the scenario of a technical support employee at a SaaS firm, we model changes in OPMs as censoring effects on the distributions of both service quality and service time. We demonstrate the nonlinear effects OPMs can have on the optimal service contract and the employer's (principal's) expected utility in hiring the technical support employee (agent), under certain distributional assumptions. This modelling approach arms operations management analysts with a new tool to better capture the impact of OPMs and their non-linear impacts on operational performance. The second essay proposes a number of additions to both static and dynamic network ranking models for professional soccer teams. We introduce ways to incorporate relevant home/away game status and goal difference information. Further, we introduce a collection of methods to measure the competitive similarity between teams, which we then integrate into the ranking systems. We demonstrate, using a large collection of data on five of the top European professional soccer leagues, that our methods produce superior empirical performance when compared to comparable approaches. Importantly, our work is the first to integrate the competitive similarity notion directly into network ranking models, providing the first direct link between two related bodies of literature.
Perishable Inventory Routing Problem under Uncertainty
(University of Waterloo, 2023-08-24) Khalili, Ghazaleh; Abouee-Mehrizi, Hossein; Ghadimi, Saeed
In an Inventory Routing Problem (IRP), a decision-maker decides the number of units delivered to each retailer and determines delivery routes, which becomes increasingly challenging as the network expands. Incorporating uncertainty and perishability into the IRP gives rise to a more complex problem known as the stochastic Perishable Inventory Routing Problem (PIRP). Traditional approaches, such as dynamic programming, often struggle to efficiently solve this problem. This is due to the curse of dimensionality, which grows exponentially with the number of retailers and the product's shelf life. In this work, we decompose the PIRP into a Perishable Inventory Problem (PIP) and a Vehicle Routing Problem (VRP) and address them sequentially in two distinct phases. By successfully determining the replenishment quantities first, we then solve the VRP using state-of-the-art algorithms. Consequently, our primary focus lies in identifying the optimal replenishment quantities for perishable products. To address the complexities of this problem, we propose a Direct Lookahead Approximation (DLA) policy designed for sequential decision-making problems under uncertainty. Specifically, we employ a two-stage approximation method that considers a limited number of sample paths while still achieving promising results. The problem is formulated as a mixed-integer programming (MIP) model with the objective of minimizing holding, shortage, wastage, and replenishment costs. In this context, a fixed cost is employed as an approximation for the routing costs of the second phase. To enhance the implementation of the DLA policy, we conduct a comprehensive analysis and recommend techniques such as incorporating linear cuts into the MIP model. To evaluate the effectiveness of the policy, we examine a blood supply chain focusing on perishable platelet units. Through extensive experiments, we demonstrate that the proposed policy can significantly outperform several known algorithms in the literature.
Localization and Counting of Indoor Populations on a University Campus using Wi-Fi Connection Logs and Floor Plans
(University of Waterloo, 2023-04-17) Carroll-Woolery, Lannois; Golab, Lukasz
The localization and counting of persons in indoor spaces is an area of extensive research. Indoor population metrics can inform energy conservation, health and safety, security, resource optimization, and location-aware services such as marketing and navigation. Building utility is impacted by the number of persons in each space, and the management of person flows into and out of building spaces is a critical consideration of space design, and the COVID-19 pandemic elevated the need to accurately measure and monitor indoor populations. Indoor populations’ size, movement and location can be ascertained by a variety of automatic means, but scalability, repeatability and cost are limiting factors. One low-cost technique is the use of wireless logs from Wi-Fi-enabled devices, which provide precise counts but inaccurate locations due to Access Points’ widely varying coverage areas. Population locations, as estimated by wireless logs, are usually defined at a floor, or building level. In this paper, I propose a generalized technique for more precise identification of indoor populations’ location, using wireless logs. It is based on the merging of connection logs with floor layout plans, to define floor zones, representing the general area(s) of wireless coverage provided by each wireless AP, including areas served by more than one AP. The combined information allows for more precise location and counting of indoor populations. This analysis could be useful across multiple functional domains, including sustainability management, resource optimization, and capacity monitoring. The technique can be implemented in any environment where there is an extensive wireless network, widespread usage of the network, and reliable data records. It is non-invasive and does not require the purchase or installation of new equipment. As a case study, we applied the technique to data from a mid-sized university. Spatial and temporal population analyses were completed using wireless logs collected over a 6-week period prior to the COVID pandemic. The logs included unique User Ids and Device Ids; The floor layout plans included the installed locations of AP devices. Facilities management records included building, floor, and room metrics. Population analyses were completed by building, room types, work weeks, and duration of wireless connections. The population estimations for size and location were compared to expected indoor populations, based on student class enrolments and employee work schedules, to gauge accuracy and utility. Linear Correlation Coefficients were calculated for measured vs. expected population counts. The results indicated that the definition of Building Floor Zones provided more accurate indoor population location values than floor-level estimates, across a variety of building types and room types. Facilities management definitions for Building Floors allowed generic description of campus spaces that could be applied to any environment with varying building usage and occupant activity. The merged data allowed the estimation of indoor populations’ size and location at various levels of aggregation: zones, floors, and buildings; and allows for comparisons of activity in similar environments in differing locations. Possible research and/or application areas include: the use of indoor spaces outside of business hours, occupancy/utility rates, and the measurement of indoor crowd densities.
Data-Driven Inverse Optimization with Applications in Electricity Markets
(University of Waterloo, 2023-01-19) Rafieepouralavialavijeh, Ali; Pirnia, Mehrdad; Pavlin, Michael
Due to the increasing penetration of renewable resources and demand response instruments in the electricity markets, generation planning models have become more complex and require detailed information on the inherent structure of the system, including generator and demand parameters. Demand should be met by cost-effective, adaptable, and efficient power plants to ensure that it is met even in the worst-case scenarios, such as an unanticipated peak or the failure of a critical generating unit. On the other hand, there is a need to consider short-term details in the Planning problems to address the needed system flexibility due to sudden changes in demand and renewables generation. Such short-term details increase the size of the models and their related computations. As a result, there is a trade-off between the complexity of the computation and the level of short-term operational details, which should be considered. Accessing electricity infrastructure data in North America is often difficult due to the lack of open data standards and the proprietary nature of much of the data. The regulations and policies surrounding the data also vary significantly from province to province, making it difficult to access the data uniformly. Additionally, privacy and security considerations can limit access even further. Despite these limitations, there are indirect methods such as inverse optimization(IO) to derive the market parameters using publicly available data; examples of these parameters include generator costs of generation, their capabilities, etc. The discovery of unobservable information via IO could aid energy models to account for operational details without increasing the complexity of their problem. Furthermore, this information can inform policymakers on potential interventions to improve the efficiency of the electricity market. In this research, a MIP model is developed to incorporate capital and operational costs associated with long-term planning problems. The operating costs of each technology are assumed to be approximated by a series of step-wise functions so that model outcomes, such as generation output, are as close as possible to real-world electricity market generation. The proposed method employs a two-stage algorithmic framework using data-driven inverse optimization and regression. In the first stage, constraints are generated based on relationships between cost and electricity prices. In the second stage, these constraints on costs are added to a problem that finds and reconciles the parameters of the cost functions. To evaluate the performances of the proposed IO-based method, it was applied to a DC-OPF model using the IEEE 24-bus system, which helped eliminate power flow constraints. This approach was then applied to a long-term planning model using Ontario's electricity market data. The results indicate that the proposed approach could find a close solution to the conventional models. In the long-term planning model, the IO-based approach showed more moderate investment policies, while the traditional methods tend to over or under-invest.
Towards Measuring Coherence in Poem Generation
(University of Waterloo, 2023-01-11) Mohseni Kiasari, Peyman; Vechtomova, Olga
Large language models (LLM) based on transformer architecture and trained on massive corpora have gained prominence as text-generative models in the past few years. Even though large language models are very adept at memorizing and generating long sequences of text, their ability to generate truly novel and creative texts including poetry lines is limited. On the other hand, past research has shown that variational autoencoders (VAE) can generate original poetic lines adhering to the stylistic characteristics of the training corpus. Originality and stylistic adherence of lines generated by VAEs can be partially attributed to the fact that, firstly, VAEs can be successfully trained on small highly curated corpora in a given style, and secondly, VAEs with a recurrent neural network architecture has a relatively low memorization capacity compared to transformer networks, which leads to the generation of more creative texts. VAEs, however, are limited to producing short sentence-level texts due to fewer trainable parameters, compared to LLMs. As a result, VAEs can only generate independent poetic lines, rather than complete and coherent poems. In this thesis, we propose a new model of coherence scoring that allows the system to rank independent lines generated by a VAE and construct a coherent poem. The scoring model is based on BERT, fine-tuned as a coherence evaluator. We propose a novel training schedule for fine-tuning BERT, during which we show the system different types of lines as negative examples: lines sampled from the same vs. different poems. The results of the human evaluation show that participants perceive poems constructed by this method to be more coherent than randomly sampled lines.
Answering Consumer Health Questions on the Web
(University of Waterloo, 2022-12-21) Vakili Tahami, Amir; Smucker, Mark
Question answering is an important sub task in the field of information retrieval. Question answering has typically used reliable sources of information such as the Wikipedia for information. In this work, we look at answering health questions using the web. The web offers the means to answer general medical questions on a variety of topics but comes with the downside of being rife with misinformation and contradictory information. We develop our techniques using the TREC health misinformation tracks that use consumer health question as topics and web crawls as their document collection. In this work, we implement a document filtering technique based on topic-sensitive PageRank that uses a web graph of the hosts in common crawl. We develop a new passage extraction technique that performs query-based contextualized sentence selection. We test this technique on a multi-span extractive question answering dataset. We also develop an answer aggregation technique that can combine language features and manual features to predict answers to these consumer health questions. We test all of these approaches on the TREC Health Misinformation Track. We show that these techniques in the majority of cases provide an uplift in performance.
Factors of Haptic Experience across Multiple Haptic Modalities
(University of Waterloo, 2022-12-08) Anwar, Ahmed; Schneider, Oliver
Haptic Experience (HX) is a proposed set of quality criteria useful to haptics, with prior evidence for a 5-factor model with vibrotactile feedback. We report on an ongoing process of scale development to measure HX, and explore whether these criteria hold when applied to more diverse devices, including vibrotactile, force feedback, surface haptics, and mid-air haptics. From an in-person user study with 430 participants, exploratory factor analysis (EFA), and confirmatory factor analysis (CFA), we extract an 11-item and 4-factor model (Realism, Harmony, Involvement, Expressivity) with only a partial overlap to the previous model. We compare this model to the previous vibrotactile model, finding that the new 4-factor model is more generalized and can guide attributes or applications of new haptic systems. This can inform designers about the right quality criteria to use when designing or evaluating haptic devices.
Gender Differences in Engineering: A Data-Driven Study
(University of Waterloo, 2022-09-27) Chopra, Shivangi; Golab, Lukasz
The gender gap in Science, Technology, Engineering, and Mathematics (STEM) is well known. Not only do fewer women apply to and earn engineering degrees, but also more women leave engineering programs and careers. Past studies have identified various reasons that affect female students' decisions to join engineering programs. Some of them include STEM interest, access to role models, and high school context. Further, labour market studies, which focus on later career stages, have found workplace experiences of engineering graduates to differ based on gender and drive female attrition. While the majority of studies on STEM recruitment are qualitative in nature or are based on small datasets collected using surveys and interviews, this thesis takes a data-driven approach towards studying gender differences in engineering. Moreover, since early career experiences can greatly affect subsequent career choices, this thesis investigates gender differences in early engineering careers, specifically in the co-operative education (co-op) form of work-integrated learning. Our analysis is enabled by unique datasets from a large North American university with renowned engineering programs and mandatory co-op. We use standard statistical and text analysis tools to measure gender differences in (a) motivations, interests, and backgrounds of 33,763 applicants, and (b) co-op work experiences of 8,956 students in terms of their choices, opportunities, evaluations, and satisfaction. The goal of this thesis is to quantify the gender gap in engineering and provide data-driven insights into closing it. While analyzing students' motivations behind joining co-op engineering programs, we find that female applicants are more likely to mention personal influences, a desire to contribute to society, and access to real-world work experiences. In addition, the unique characteristics of high schools that produce more female engineering applicants include: a) on average, female students from these schools outperform male students on standardized math tests, and b) applicants from these schools report more personal influence and a wider variety of interests, encompassing technology, arts, community, and travel. However, these applicants participate in fewer collaborative and competitive STEM activities. Our analysis of students' co-op experiences shows that female students tend to apply to and fill slightly different positions than male students. While male and female students appear equally likely to obtain interviews and secure placements, female students seem to take more risks when ranking potential job opportunities and receive slightly higher performance appraisals. Nevertheless, male students appear to be perceived as more agentic and are more satisfied than female students from the very beginning of their careers. The data-driven findings presented in this thesis may encourage female students to apply to engineering, as well as provide actionable insights to academic institutions and employers wishing to diversify their talent pool.
Radiotherapy Patient Scheduling During Pandemics
(University of Waterloo, 2022-09-26) Raeisi, Shamim; Mahmoudzadeh, Houra
With the Covid-19 outbreak happening worldwide, clinically vulnerable people should be of concern, as they are more likely to be exposed to the virus. Cancer patients with weak immune systems are a group of aforementioned people that often have to undergo radiotherapy treatment sessions every day for several weeks. Therefore, special measures are to take place for more protection. During the treatment process, they will be assigned to Linear Accelerator (LINAC) machines that are located in separate rooms of the radiotherapy center. During each visit, they are in close contact with other patients that are assigned to the same LINAC, but for different time slots. Our research focuses on scheduling radiotherapy patients, using two mixed-integer linear programming models, to minimize the total number of potential interactions between patients. A secondary objective is then proposed to choose among the set of optimal solutions, and the models' complexity growth is discussed. Then, we introduce a heuristic algorithm to increase the efficiency of the proposed model for large instances and use a visual step-by-step example to further elaborate the algorithm details. Finally, small numerical examples are used to demonstrate the effectiveness of the models, followed by larger instances from our partner clinic, the Grand River Regional Cancer Center (GRRCC). The results show that implementing the proposed model and the heuristic will decrease the number of interactions up to 75%, compared to the centre's original schedule.
Dynamic Robust Multi-Class Advance Patient Scheduling
(University of Waterloo, 2022-09-01) Khajeh Arzani, Hamidreza; Abouee Mehrizi, Hossein
In this work, we study an advance patient scheduling problem where patients of different classes have different service times and incur different waiting costs to the system. It is known in the literature that multi-class advance dynamic patient scheduling is a challenging problem due to the high variability in the daily arrival process of patients, as well as the high dimensionality of the problem. To overcome these challenges, we develop a novel dynamic optimization framework where the multi-class advance scheduling problem can be approximately decomposed to multiple single-stage stochastic programs. Furthermore, we develop a distributionally robust formulation and quantify uncertainty in arrivals by applying a risk-averse optimization approach. Exploiting patient-level offline data, we develop a data-driven algorithm to minimize the worst-case outcome that may happen due to the high variability in arrivals. We examine the performance of the proposed robust algorithm by leveraging the MRI data from hospitals in Ontario and show that the dynamic robust model outperforms the dynamic stochastic approach significantly. We also observe that the proposed robust model performs well compared to an offline policy, which is based on the full knowledge of the future arrivals.
Constraint-Guided Machine Learning for Solving Optimal Power Flow Problem
(University of Waterloo, 2022-08-31) Lotfi, Amir; Pirnia, Mehrdad
Due to the nonlinear and non-convex attributes of the optimization problems in power systems such as Optimal Power Flow (OPF), traditional iterative optimization algorithms require significant amount of time to converge for large electric networks. Therefore, power system operators seek other methods such as DC Optimal Power Flow (DCOPF) to obtain faster results, to obtain the state of the system. However, DCOPF provides approximated results, neglecting important features of the system such as voltage and reactive power. Fortunately, recent developments in machine learning have led to new approaches for solving such problems faster, more flexible, and more accurate. In this research, a Deep Neural Network-based Optimal Power Flow (DNN-OPF) algorithm is implemented on small to large case studies to show the accuracy and efficiency of the ML-based algorithms. Since the ML methods such as NN are considered black-box approaches, the system operators are not satisfied with solving power system models using them, as such methods do not explain the reasoning behind the generated solutions. Moreover, there is no guarantee that the obtained solutions would be converging and close to optimality. To overcome such issues this research provides a novel approach to first classify the converging and non-converging ACOPF problems, and then suggests a constraint-guided method, based on normalizing outputs and using particular activation functions to satisfy the technical limits of the generators such as maximum and minimum generation. Furthermore, a post-processing approach is incorporated to check for the convergence of the power flow equations which are in form of equality constraints. The suggested method is applied on IEEE24-bus, IEEE 300 bus, and PEGASE 1354 bus systems and the results show significant improvement on execution time, comparing to traditional gradient-based methods, such as Newton-Raphson and Gauss–Seidel methods. Also, the approach has been benchmarked against DCOPF model and it is shown that the proposed DNN-OPF not only provides faster speed, but also ensures higher accuracy on the final results. Furthermore, since is a need to run ACOPF problem using different scenarios, to account for continuous changes in the demand, the suggested DNN-OPF is solved for various scenarios from 1 to 10,000 to appreciate the improved execution time obtained from the ML-based approaches. Our results show that DNN can improve execution time a factor of 400 to 800 for large to small networks.

Browse

Recent Submissions