CS Theses & Dissertations 2018

For 2018 graduation dates (in alphabetical order by last name):

Enabling configuration self-adaptation using machine learning
Araujo, Rodrigo Silva
Degree : Master of Science – MSc
Graduation Date : 2018-11
Supervisor : Dr. Holmes

Context-aware conversational developer assistants
Bradley, Nicholas
DOI : 10.14288/1.0370955
URI : http://hdl.handle.net/2429/66724
Degree : Master of Science – MSc
Graduation Date : 2018-11
Supervisor : Dr. Holmes

Inter-process communication in disaggregated datacenters
Carbonari, Amanda
DOI : 10.14288/1.0365936
URI : http://hdl.handle.net/2429/65548
Degree : Master of Science – MSc
Graduation Date : 2018-05
Supervisor : Dr. Beschastnikh

Towards automatic broadcast of team sports
Chen, Jianhui
DOI : 10.14288/1.0371096
URI : http://hdl.handle.net/2429/66791
Degree : Doctor of Philosophy - PhD
Graduation Date : 2018-11
Supervisor : Dr. Little

Sports is the social glue of society as it allows people to interact with each other and appreciate games irrespective of their social status, age and ethnicity. Automatic sports broadcasting produces stream videos from vision sensors without human intervention. The goal is to predict where cameras should look and which camera should be on air. The technique can benefit millions of people as most viewers participant in sports by watching TV or Internet broadcasting. The target team sports include basketball, soccer and ice hockey in which team members quickly move their positions in the game, excluding sports like baseball and cricket in which team members have relatively stable positions. Automatic sports broadcasting covers areas of statistics, commentary, camera control and so on. We provide solutions for automatically setting camera parameters such as camera orientation angles and locations using computer vision. We restrict our attention to static pan-tilt-zoom (PTZ) cameras for television or live Internet broadcasting. We propose three essential components of autonomous broadcasting: camera calibration, planning and selection. By learning from human demonstrations, our work can predict camera angles for single camera systems and camera viewpoints for multi-camera systems. We obtain human demonstrations from existing videos that are generated by professional camera operators. These videos contain camera angles and camera IDs if there are multiple cameras. Because camera angles are not directly available, we first propose two novel camera calibration methods. We evaluate and compare our methods with previous algorithms. Our methods are more accurate and faster than previous algorithms. With labeled data from human operators, we develop two methods for smooth camera planning which predict camera pan angles. The first method directly incorporates temporal consistency into a data-driven predictor. The second method optimizes the camera trajectory in overlapped temporal windows. We show they outperform previous methods in the literature. We also propose two methods for selecting a broadcast camera view from multiple candidate camera views. The first method uses deep features for camera selection. The second method augments the training data with Internet videos. We demonstrate comparable results with selections from human operators in soccer games.

Segmentifier: interactively refining clickstream data into actionable segments
Dextras-Romagnino, Kimberly
DOI : 10.14288/1.0365820
URI : http://hdl.handle.net/2429/65515
Degree : Master of Science – MSc
Graduation Date : 2018-05
Supervisor : Dr. Munzner

Dancing in the dark: private multi-party machine learning in an untrusted setting
Fung, Clement
DOI : 10.14288/1.0372888
URI : http://hdl.handle.net/2429/67623
Degree : Master of Science – MSc
Graduation Date : 2018-11
Supervisor : Dr. Beschastnikh

Deep learning with exchangeable tensors
Graham, Devon R.
DOI : 10.14288/1.0370999
URI : http://hdl.handle.net/2429/66762
Degree : Master of Science – MSc
Graduation Date : 2018-11
Supervisor : Dr. Ravanbakhsh

Inferring and asserting distributed invariants
Grant, Stewart Steven
DOI : 10.14288/1.0371258
URI : http://hdl.handle.net/2429/66965
Degree : Master of Science – MSc
Graduation Date : 2018-11
Supervisor : Dr. Beschastnikh

OnionCoin: peer-to-peer anonymous messaging with incentive system
Gu, Tianri
DOI : 10.14288/1.0371952
URI : http://hdl.handle.net/2429/67136
Degree : Master of Science – MSc
Graduation Date : 2018-11
Supervisor : Dr. Feeley

Visual techniques for exploring alternatives and preferences in Group Preferential Choice
Hindalong, Emily
DOI : 10.14288/1.0363067
URI : http://hdl.handle.net/2429/64387
Degree : Master of Science – MSc
Graduation Date : 2018-05
Supervisor : Dr. Carenini

Understanding the sources of error for 3D human pose estimation from monocular images and videos
Hossain, Mir Rayat Imtiaz
DOI : 10.14288/1.0361162
URI : http://hdl.handle.net/2429/63808
Degree : Master of Science – MSc
Graduation Date : 2018-05
Supervisor : Dr. Little

Summarization of partial email threads: silver standards and bayesian surprise
Johnson, Jordon Kent
DOI : 10.14288/1.0365780
URI : http://hdl.handle.net/2429/65468
Degree : Master of Science – MSc
Graduation Date : 2018-05
Supervisors : Dr. Carenini, Dr. Ng

Enhancing user privacy in web services
Légaré, Jean-Sébastien
DOI : 10.14288/1.0358013
URI : http://hdl.handle.net/2429/63664
Degree : Doctor of Philosophy - PhD
Graduation Date : 2018-05
Supervisors : Dr. Warfield, Dr. Aiello

FoldSketch: enriching garments with physically reproducible folds
Li, Minchen
DOI : 10.14288/1.0365821
URI : http://hdl.handle.net/2429/65532
Degree : Master of Science – MSc
Graduation Date : 2018-05
Supervisor : Dr. Sheffer

Detecting dementia from written and spoken language
Masrani , Vaden Wybert
DOI : 10.14288/1.0362923
URI : http://hdl.handle.net/2429/64313
Degree : Master of Science – MSc
Graduation Date : 2018-05
Supervisor : Dr. Carenini

Measurement and Animation of the Eye Region of the Human Face
Neog , Debanga Raj
DOI : 10.14288/1.0368664
URI : http://hdl.handle.net/2429/66300
Degree : Doctor of Philosophy - PhD
Graduation Date : 2018-11
Supervisors : Dr. Pai, Dr. Woodham

The goal of this dissertation is to develop methods to measure, model, and animate facial tissues of the region around the eyes, referred to as the eye region. First, we measure the subtle movements of the soft tissues of the eye region using a monocular RGB-D camera setup, and second, we model and animate these movements using parameterized motion models. The muscles and skin of the eye region are very thin and sheetlike. By representing these tissues as thin elastic sheets in reduced coordinates, we have shown how we can measure and animate these tissues efficiently. To measure tissue movements, we optically track both eye and skin motions using monocular video sequences. The key idea here is to use a reduced coordinates framework to model thin sheet-like facial skin of the eye region. This framework implicitly constrains skin to conform to the shape of the underlying object when it slides. The skin configuration can then be efficiently reconstructed in 3D by tracking two dimensional skin features in video. This reduced coordinates model allows interactive real-time animation of the eye region in WebGL enabled devices using a small number of animation parameters, including gaze. Additionally, we have shown that the same reduced coordinates framework can also be used for physics-based simulation of the facial tissue movements and to produce tissue deformations that occur in facial expressions. We validated our skin measurement and animation algorithms using skin movement sequences with known skin motions, and we can recover skin sliding motions with low reconstruction errors. We also propose an image-based algorithm that corrects accumulated inaccuracy of standard 3D anatomy registration systems that occurs during motion capture, anatomy transfer, image generation, and animation. After correction, we can overlay the anatomy on input video with low misalignment errors for augmented reality applications, such as anatomy mirroring. Our results show that the proposed image-based corrective registration can effectively reduce these inaccuracies.

Greed is Good: Greedy Optimization Methods for Large-Scale Structured Problems
Nutini , Julie Ann
DOI : 10.14288/1.0367781
URI : http://hdl.handle.net/2429/66137
Degree : Doctor of Philosophy - PhD
Graduation Date : 2018-11
Supervisor : Dr. Schmidt

This work looks at large-scale machine learning, with a particular focus on greedy methods. A recent trend caused by big datasets is to use optimization methods that have a cheap iteration cost. In this category are (block) coordinate descent and Kaczmarz methods, as the updates of these methods only rely on a reduced subspace of the problem at each iteration. Prior to our work, the literature cast greedy variations of these methods as computationally expensive with comparable convergence rates to randomized versions. In this dissertation, we show that greed is good. Specifically, we show that greedy coordinate descent and Kaczmarz methods have efficient implementations and can be faster than their randomized counterparts for certain common problem structures in machine learning. We show linear convergence for greedy (block) coordinate descent methods under a revived relaxation of strong convexity from 1963, which we call the Polyak-Lojasiewicz (PL) inequality. Of the proposed relaxations of strong convexity in the recent literature, we show that the PL inequality is the weakest condition that still ensures a global minimum. Further, we highlight the exploitable flexibility in block coordinate descent methods, not only in the different types of selection rules possible, but also in the types of updates we can use. We show that using second-order or exact updates with greedy block coordinate descent methods can lead to superlinear or finite convergence (respectively) for popular machine learning problems. Finally, we introduce the notion of “active-set complexity”, which we define as the number of iterations required before an algorithm is guaranteed to reach the optimal active manifold, and show explicit bounds for two common problem instances when using the proximal gradient or the proximal coordinate descent method.

Computational imaging with diffractive optics
Pang, Yifan
DOI : 10.14288/1.0365608
URI : http://hdl.handle.net/2429/65337
Degree : Doctor of Philosophy - PhD
Graduation Date : 2018-05
Supervisor : Dr. Heidrich

Diffractive optical elements (DOEs) have been studied extensively in optics for decades, but have recently received a lot of renewed interests in the context of computational imaging and computational display because they can drastically reduce the size and weight of devices. However, the inherent strong dispersion is an obstacle that limits the use of DOEs in full spectrum imaging, causing unacceptable color fidelity loss in the captured or reconstructed images. Despite the benefits of facilitating compact form factors, DOEs have sufficient degrees of freedom that one can manipulate to encode the desirable light modulation functionality. In this dissertation we theoretically and experimentally investigate the practicability of introducing numerical optimization into the design procedure of optics, to enable a variety of diffractive optics subject to different application scenarios. Regarding imaging applications, we first validate the practicality of introducing diffractive-refractive hybrid elements as simplified optics in conventional computational imaging. We re-implement a cross-channel based deconvolution to correct the chromatic aberration. The full fabrication cycle of photolithography technique is developed, serving as the basis for all designs. Then, the desirable focal powers in both spectral and spatial domain are encoded onto the DOEs. Precisely, we develop the diffractive achromat that balances the focusing contributions in full visible spectrum. The color fidelity can thus be well preserved. Meanwhile, we develop the encoded lenses that provide focus tunable imaging ability and multi-focal sweep imaging compromise. The trade-off light loss and residual aberrations are tackled by a deconvolution step. In addition, we extend the design paradigm from image capture to image display, where the holograms of visualization of multiple narrowband spectra are encoded onto a pair of ultrathin diffractive phase plates. The mix-and-match scheme of designing high degree-of-freedom diffractive optics is exploited via the complex matrix factorization. The essential random appearance of holograms in fact benefits to enabling a wide range of visualization applications. We envision the work on computational imaging, involving both capture and display end, provide new insights on incorporating optics and computation algorithms to better record, understand and deliver desirable visual information under the constraint of current data bandwidth of hardware.

[no title]
Reza, Adnan
Degree : Master of Science – MSc
Graduation Date : 2018-05
Supervisor : Dr. Poole

Computational Single-Image High Dynamic Range Imaging
Rouf , Mushfiqur
DOI : 10.14288/1.0368721
URI : http://hdl.handle.net/2429/66350
Degree : Doctor of Philosophy - PhD
Graduation Date : 2018-11
Supervisors : Dr. Little, Dr. Ward (EECE)

This thesis proposes solutions for increasing the dynamic range (DR)—the number of intensity levels—of a single image captured by a camera with a standard dynamic range (SDR). The DR in a natural scene is usually too high for SDR cameras to capture, even with optimum exposure settings. The intensity values of bright objects (highlights) that are above the maximum exposure capacity get clipped due to sensor over-exposure, while objects that are too dark (shades) appear dark and noisy in the image. Capturing a high number of intensity levels would solve this problem, but this is costly, as it requires the use of a camera with a high dynamic range (HDR). Reconstructing an HDR image from a single SDR image is difficult, if not impossible, to achieve for all imaging situations. For some situations, however, it is possible to restore the scene details, using computational imaging techniques. We investigate three such cases, which also occur commonly in imaging. These cases pose relaxed and well-posed versions of the general single-image high dynamic range imaging (HDRI) problem. The first case occurs when the scene has highlights that occupy a small number of pixels in the image; for example, night scenes. We propose the use of a cross-screen filter, installed at the lens aperture, to spread a small part of the light from the highlights across the rest of the image. In post-processing, we detect the spread-out brightness and use this information to reconstruct the clipped highlights. Second, we investigate the cases when highlights occupy a large part of the scene. The first method is not applicable here. Instead, we propose to apply a spatial filter at the sensor that locally varies the DR of the sensor. In post-processing, we reconstruct an HDR image. The third case occurs when the clipped parts of the image are not white but have a color. In such cases, we restore the missing image details in the clipped color channels by analyzing the scene information available in other color channels in the captured image. For each method, we obtain a maximum-a-posteriori estimate of the unknown HDR image by analyzing and inverting the forward imaging process.

Time-travel Programming: Programming Language Support for Interacting with Past Executions
Salkeld , Robin
DOI : 10.14288/1.0367029
URI : http://hdl.handle.net/2429/66058
Degree : Doctor of Philosophy - PhD
Graduation Date : 2018-11
Supervisor : Dr. Kiczales

Because software so often behaves unexpectedly or fails only in production environments, several recent tools from both industry and academia record data about execution for the benefit of post-hoc analysis. Debugging on these data instead of a live program is much more difficult, however, because the semantic abstractions provided by the programming language are no longer available. Many post-hoc analysis tools process this data through additional reflection-based code or domain-specific query languages, but do not recover the expressive power of the original programming language. This thesis proposes the concept of time-travel programming, which we define as simulating the execution of additional code in the same programming language as if it were present in the past environment of recorded data. Furthermore, we show that the aspect-oriented programming (AOP) paradigm provides a natural mechanism for specifying this additional execution, and allows us to reuse established semantics and implementations. We provide evidence of this technique’s flexibility, feasibility and effectiveness through two implementations: one an interpreter for an extremely simple AOP language in the style of a core calculus, and one for the AspectJ programming language. We evaluate flexibility via applying the implementations to multiple execution recording formats, feasibility by showing the AspectJ implementation is performant enough for post-hoc analysis, and effectiveness by demonstrating that evaluating new and existing aspects retroactively can be used to address common post-hoc analysis tasks.

[no title]
Song, Julin Asiiah
Degree : Master of Science – MSc
Graduation Date : 2018-05
Supervisor : Dr. Little

Exploiting temporal structures in computational photography
Su, Shuochen
DOI : 10.14288/1.0365768
URI : http://hdl.handle.net/2429/65446
Degree : Doctor of Philosophy - PhD
Graduation Date : 2018-05
Supervisor : Dr. Heidrich

Despite the tremendous progress in computer vision over the last decade, images captured by a digital camera are often regarded as the instantaneous 2D projection of the scene for simplification. This assumption poses a significant challenge to machine perception algorithms due to the existence of many imaging- and scene-induced artifacts in the camera's measurements such as the hand-shake blur and time-of-flight multi-path interference. In this thesis we introduce time-resolved image formation models for color and depth cameras by exploiting the temporal structure within their raw sensor measurements of the scene. Specifically, we present our efforts on leveraging the inter-scanline and cross-frame content correlations for image and video deblurring, as well as utilizing the complementary components of time-of-flight frequency measurements in collaboration for improved depth acquisition. By tackling these limitations we also enable novel imaging applications such as direct material classification from raw time-of-flight data. In addition, we devise post-processing algorithms with temporal structure awareness so that the hidden information can be decoded efficiently with existing off-the-shelf hardware devices. We believe that the proposed time-resolved modeling of the encoding-decoding process of a digital camera opens the door to many exciting directions in computational photography research.

The Role of Information in Economic Games
Tandberg, Gudbrand Andreas Duff Morris
Degree : Master of Science – MSc
Graduation Date : 2018-05
Supervisor : Dr. Fu

[no title]
Thorhallsson, Halldor Bjarni
Degree : Master of Science – MSc
Graduation Date : 2018-05
Supervisor : Dr. Ng

[no title]
Venkataswamy, Meghana
Degree : Master of Science – MSc
Graduation Date : 2018-05
Supervisor : Dr. Maclean

Data driven auto-completion for keyframe animation
Zhang , Xinyi
DOI : 10.14288/1.0371204
URI : http://hdl.handle.net/2429/66905
Degree : Master of Science – MSc
Graduation Date : 2018-11
Supervisor : Dr. van de Panne