Search results
(1 - 5 of 5)
- Title
- DATA PRIVACY AND DEEP LEARNING IN THE MOBILE ERA: TRACEABILITY AND PROTECTION
- Creator
- Chen, Linlin
- Date
- 2020
- Description
-
Privacy and deep learning have been two of the most exciting research trends in both academia and industry. On the one hand, big data rapidly...
Show morePrivacy and deep learning have been two of the most exciting research trends in both academia and industry. On the one hand, big data rapidly expedite lots of data orientated applications, especially like deep learning services. With the tremendous value exhibited by the data, the privacy of data subjects who generate the data, has also raised much attention. Meanwhile more regulations and legislation have been enacted or enforced, intending to enforce the companies and organizations to strictly comply with the personal privacy protection while collecting or utilizing their data. All these moves will substantially change the ways to train the deep learning models and provide AI services, and in some ways might hinder the development of deep learning if not coming up with some sophisticated mechanisms. On the other hand, deep learning has been showing incredibly promising performance in a variety of areas like face recognition, voice recognition, recommendation & advertising, autonomous driving, medical imaging, etc.. This keeps us thinking will deep learning also in turn influence privacy and be leveraged to compromise privacy. Meanwhile we also observe that mobile devices become so ubiquitous that more shares of data are generated on mobile devices, and mostly those data are both extremely sensitive for data subjects as well as extremely valuable for developing deep learning. We shouldn’t neglect the impact of mobile devices on both privacy and deep learning.In this thesis I explore the research on the interactions between privacy and deep learning, especially with the mobile devices being involved in. Specifically I work on: 1). How does privacy change the way we use the data when building deep learning models, and present the mechanism for privacy protection towards deep learning. 2). How does deep learning in turn make privacy more vulnerable to be compromised, and demonstrate the privacy compromise by facilitating deep learning to trace the source mobile devices and link the personal identities.
Show less
- Title
- Image Synthesis with Generative Adversarial Networks
- Creator
- Ouyang, Xu
- Date
- 2023
- Description
-
Image synthesis refers to the process of generating new images from an existing dataset, with the objective of creating images that closely...
Show moreImage synthesis refers to the process of generating new images from an existing dataset, with the objective of creating images that closely resemble the target images, learned from the source data distribution. This technique has a wide range of applications, including transforming captions into images, deblurring blurred images, and enhancing low-resolution images. In recent years, deep learning techniques, particularly Generative Adversarial Network (GAN), has achieved significant success in this field. GAN consists of a generator (G) and a discriminator (D) and employ adversarial learning to synthesize images. Researchers have developed various strategies to improve GAN performance, such as controlling learning rates for different models and modifying the loss functions. This thesis focuses on image synthesis from captions using GANs and aims to improve the quality of generated images. The study is divided into four main parts:In the first part, we investigate the LSTM conditional GAN which is to generate images from captions. We use the word2vec as the caption features and combine these features’ information by LSTM and generate images via conditional GAN. In the second part, to improve the quality of generated images, we address the issue of convergence speed and enhance GAN performance using an adaptive WGAN update strategy. We demonstrate that this update strategy is applicable to Wasserstein GAN(WGAN) and other GANs that utilize WGAN-related loss functions. The proposed update strategy is based on a loss change ratio comparison between G and D. In the third part, to further enhance the quality of synthesized images, we investigate a transformer-based Uformer GAN for image restoration and propose a two-step refinement strategy. Initially, we train a Uformer model until convergence, followed by training a Uformer GAN using the restoration results obtained from the first step.In the fourth part, to generate fine-grained image from captions, we delve into the Recurrent Affine Transformation (RAT) GAN for fine-grained text-to-image synthesis. By incorporating an auxiliary classifier in the discriminator and employing a contrastive learning method, we improve the accuracy and fine-grained details of the synthesized images.Throughout this thesis, we strive to enhance the capabilities of GANs in various image synthesis applications and contribute valuable insights to the field of deep learning and image processing.
Show less
- Title
- Deep Learning and Model Predictive Methods for the Control of Fuel-Flexible Compression Ignition Engines
- Creator
- Peng, Qian
- Date
- 2022
- Description
-
Compression ignited diesel engines are widely used for transportation and power generation because of their high fuel efficiency. However,...
Show moreCompression ignited diesel engines are widely used for transportation and power generation because of their high fuel efficiency. However, diesel engines can cause concerning environmental pollution because of their high nitrogen oxide (NOx) and soot emissions. In addition to meeting the stringent emission regulations, the demand to reduce greenhouse gas emissions has become urgent due to the more frequent destructive catastrophes caused by global warming in recent decades. In an effort to reduce emissions and improve fuel economy, many techniques have been developed and investigated by researchers. Air handling systems like exhaust gas recirculation and variable geometry turbochargers are the most widely used techniques on the market for modern diesel engines. Meanwhile, the concept of low temperature combustion is widely investigated by researchers. Low temperature combustion can increase the portion of pre-mixed fuel-air combustion to reduce the peak in-cylinder temperature so that the formation of NOx can be suppressed. Furthermore, the combustion characteristics and performance of bio-derived fuel blends are also studied to reduce overall greenhouse gas emissions through the reduced usage of fossil fuels. All the above mentioned systems are complicated because they involve not only chemical reactions but also complex fluid motion and mixing processes. As such, the control of these systems is always challenging and limits their commercial application. Currentlymost control methods are feed-forward control based on load condition and engine speed due to the simplicity in real-time application. With the development of faster control unit and deep learning techniques, the application of more complex control algorithms is possible to further improve the emissions and fuel economy. This work focuses on improvements to the control of engine air handling systems and combustion processes that leverage alternative fuels.Complex air handling systems, featuring technologies such as exhaust gas recirculation (EGR) and variable geometry turbochargers (VGTs), are commonly used in modern diesel engines to meet stringent emissions and fuel economy requirements. The control of diesel air handling systems with EGR and VGTs is challenging because of their nonlinearity and coupled dynamics. In this thesis, artificial neural networks (ANNs) and recurrent neural networks (RNNs) are applied to control the low pressure (LP) EGR valve position and VGT vane position simultaneously on a light-duty multi-cylinder diesel engine. In addition, experimental examination of a low temperature combustion based on gasoline compression ignition as well as its control has also been studied in this work. This type of combustion has been explored on traditional diesel engines in order to meet increasingly stringent emission regulations without sacrificing efficiency. In this study, a six-cylinder heavy-duty diesel engine was operated in a mixing controlled gasoline compression ignition mode to investigatethe influence of fuels and injection strategies on the combustion characteristics, emissions, and thermal efficiencies. Fuels, including ethanol (E), isobutanol (IB), and diisobutylene (DIB), were blended with a gasoline fuel to form E10, E30, IB30, and DIB30 based on volumetric fraction. These four blends along with gasoline formed the five test fuels. With these fuels, three injections strategies were investigated, including late pilot injection, early pilot injection, and port fuel injection/direct injection. The impact of moderate exhaust gas recirculation on nitrogen oxides and soot emissions was examined to determine the most promising fuel/injection strategy for emissions reduction. In addition, first and second law analyses were performed to provide insights into the efficiency, loss, and exergy destruction of the various gasoline fuel blends at low and medium load conditions. Overall, the emission output, thermal efficiency, and combustion performances of the five fuels were found to be similar and their differences are modest under most test conditions.While experimental work showed that low temperature combustion with alternative fuels could be effective, control is still challenging due to not only the properties of different gasoline-type fuels but also the impacts of injection strategies on the in-cylinder reactivity. As such, a computationally efficient zero-dimension combustion model can significantly reduce the cost of control development. In this study, a previously developed zero-dimension combustion model for gasoline compression ignition was extended to multiple gasoline-type fuel blends and a port fuel injection/direct fuel injection strategy. Tests were conducted on a 12.4-liter heavy-duty engine with five fuel blends. A modification was made to the functional ignition delay model to cover the significantly different ignition delay behavior between conventional and oxygenated fuel blends. The parameters in the model were calibrated with only gasoline data at a load of 14 bar brake mean effective pressure. The results showed that this physics-based model can be applied to the other four fuel blends at three differentpilot injection strategies without recalibration. In order to also facilitate the control of emissions, machine learning models were investigated to capture NOx emissions. A kernel-based extreme learning machine (K-ELM) performed best and had a coefficient of correlation (R-squared) of 0.998. The combustion and NOx emission models are valid for not only conventional gasoline fuel but also oxygenated alternative fuel blends at three different pilot injection strategies. In order to track key combustion metrics while keeping noise and emissions within constraints, a model predictive control(MPC) was applied for a compression ignition engine operating with a range of potential fuels and fuel injection strategies. The MPC is validated under different scenarios, including a load step change, fuel type change, and injection strategy change, with proportional-integral (PI) control as the baseline. The simulation results show that MPC can optimize the overall performance through modifying the main injection timing, pilot fuel mass, and exhaust gas recirculation (EGR) fraction.
Show less
- Title
- Machine learning applications to video surveillance camera placement and medical imaging quality assessment
- Creator
- Lorente Gomez, Iris
- Date
- 2022
- Description
-
In this work, we used machine learning techniques and data analysis to approach two applications. The first one, in collaboration with the...
Show moreIn this work, we used machine learning techniques and data analysis to approach two applications. The first one, in collaboration with the Chicago Police Department (CPD), involves analyzing and quantifying the effect that the installation of cameras had on crime, and developing a predictive model with the goal of optimizing video surveillance camera location in the streets. While video surveillance has become increasingly prevalent in policing, its intended effect on crime prevention has not been comprehensively studied in major cities in the US. In this study, we retrospectively analyzed the crime activities in the vicinity of 2,021 surveillance cameras installed between 2005 and 2016 in the city of Chicago. Using Difference-in-Differences (DiD) analysis, we examined the daily crime counts that occurred within the fields-of-view of these cameras over a 12-month period, both before and after the cameras were installed. We also investigated their potential effect on crime displacement and diffusion by examining the crime activities in a buffer zone (up to 900 ft) extended from the cameras. The results show that, collectively, there was an 18.6% reduction in crime counts within the direct viewsheds of all of the study cameras (excluding District 01 where the Loop -Chicago's business center- is located). In addition, we adapted the methodology to quantify the effect of individual cameras. The quantified effect on crime is the prediction target of our 2-stage machine learning algorithm that aims to estimate the effect that installing a videocamera in a given location will have on crime. In the first stage, we trained a classifier to predict if installing a videocamera in a given location will result in a statistically significant decrease in crime. If so, the data goes through a regression model trained to estimate the quantified effect on crime that the camera installation will have. Finally, we propose two strategies, using our 2-stage predictive model, to find the optimal locations for camera installations given a budget. Our proposed strategies result in a larger decrease in crime than a baseline strategy based on choosing the locations with higher crime density.The second application that forms this thesis belongs to the field of model observers for medical imaging quality assessment. With the advance of medical imaging devices and technology, there is a need to evaluate and validate new image reconstruction algorithms. Image quality is traditionally evaluated by using numerical figures of merit that indicate similarity between the reconstruction and the original. In medical imaging, a good reconstruction strategy should be one that helps the radiologist perform a correct diagnosis. For this reason, medical imaging reconstruction strategies should be evaluated on a task-based approach by measuring human diagnosis accuracy. Model observers (MO) are algorithms capable of acting as human surrogates to evaluate reconstruction strategies, reducing significantly the time and cost of organizing sessions with expert radiologists. In this work, we develop a methodology to estimate a deep learning based model observer for a defect localization task using a synthetic dataset that simulates images with statistical properties similar to trans-axial sections of X-ray computed tomography (CT). In addition, we explore how the models access diagnostic information from the images using psychophysical methods that have been previously employed to analyze how the humans extract the information. Our models are independently trained for five different humans and are able to generalize to images with noise statistic backgrounds that were not seen during the model training stage. In addition, our results indicate that the diagnostic information extracted by the models matches the one extracted by the humans.
Show less
- Title
- ENLARGED PERIVASCULAR SPACES IN COMMUNITY-BASED OLDER ADULTS
- Creator
- Javierre Petit, Carles
- Date
- 2020
- Description
-
Enlarged perivascular spaces (EPVS) have been associated with aging, increased stroke risk, decreased cognitive function and vascular dementia...
Show moreEnlarged perivascular spaces (EPVS) have been associated with aging, increased stroke risk, decreased cognitive function and vascular dementia. Furthermore, recent studies have investigated the links of EPVS with the glymphatic system (GS), since perivascular spaces are thought to play a major role as the main channels for clearance of interstitial solutes from the brain. However, the relationship of EPVS with age-related neuropathologies is not well understood. Therefore, more conclusive studies are needed to elucidate specific relationships between EPVS and neuropathologies. After demonstration of their neuropathologic correlates, detailed assessment of EPVS severity could provide as a potential biomarker for specific neuropathologies.In this dissertation, our focus was twofold: to develop a fully automatic EPVS segmentation model via deep learning with a set of guidelines for model optimization, and to evaluate both manual and automatic assessment of EPVS severity to investigate the neuropathologic correlates of EPVS, and their contribution to cognitive decline, by combining ex-vivo brain magnetic resonance imaging (MRI) and pathology (from autopsy) in a large community-based cohort of older adults. This project was structured as follows. First, a manual approach was used to assess neuropathologic and cognitive correlates of EPVS burden in a large dataset of community-dwelling older adults. MR images from each participant were rated using a semiquantitative 4-level rating scale, and a group of identified EPVS was histologically evaluated. Two groups of participants in descending order of average cognitive impairment were defined based and studied. Elasticnet regularized ordinal logistic regression was used to assess the neuropathologic correlates of EPVS burden in each group, and linear mixed effects models were used to investigate the associations of EPVS burden with cognitive decline. Second, a fully automatic EPVS segmentation model was implemented via deep learning (DL) using a small dataset of 10 manually segmented brain MR images. Multiple techniques were evaluated to optimize performance, mainly by implementing strategies to reduce model overfitting. The final segmentation model was evaluated in an independent test set and the performance was validated with an expert radiologist. Third, the DL segmentation model was used to segment and quantify EPVS. Quantified EPVS (qEPVS) were evaluated by combining ex-vivo MRI, pathology, and longitudinal cognitive evaluation. EPVS quantification allowed to study qEPVS both in the whole brain and regionally. Two different qEPVS metrics were studied. Elastic-net regularized linear regression was used to assess the neuropathologic correlates of qEPVS within each region of interest (ROI) under study, and linear mixed effects models were used to investigate the associations of qEPVS with cognitive decline. Finally, a preliminary study investigated the longitudinal associations of qEPVS with time. The DL segmentation model was re-trained using 4 in-vivo MR images. EPVS were segmented and quantified in a large longitudinal cohort where each participant was imaged at multiple timepoints. Factors that influenced segmentation performance across timepoints were evaluated, and linear mixed effects models controlling for these factors were used to investigate the associations of qEPVS with time.
Show less