教学课程资源库 更多>>

关键词： Electrical engineering   Computer science   Biomedical engineering  

关键词： Computer science   Electrical engineering   Artificial intelligence  

摘要： In modern settings of data analysis, we may be running our algorithms on datasets that are sensitive in nature. However, classical machine learning and statistical algorithms were not designed with these risks in mind, and it has been demonstrated that they may reveal personal information. These concerns disincentivize individuals from providing their data, or even worse, encouraging intentionally providing fake data. To assuage these concerns, we import the constraint of differential privacy to the statistical inference, considered by many to be the gold standard of data privacy. This thesis aims to quantify the cost of ensuring differential privacy, i.e., understanding how much additional data is required to perform data analysis with the constraint of differential privacy. Despite the maturity of the literature on differential privacy, there is still inadequate understanding in some of the most fundamental settings. In particular, we make progress in the following problems: What is the sample complexity of DP hypothesis testing? Can we privately estimate distribution properties with a negligible cost? What is the fundamental limit in private distribution estimation? How can we design algorithms to privately estimate random graphs? What is the trade-off between the sample complexity and the interactivity in private hypothesis selection?

关键词： Engineering   Computer engineering   Educational technology   Computer science  

摘要： Visual behavior provides a dynamic trail of where attention is directed. It is considered the behavioral interface between engagement and gaining information, and researchers have used it for several decades to study user's behavior. This thesis focuses on employing visual attention to understand user's behavior in two contexts: 3D learning and gauging URL safety. Such understanding is valuable for improving interactive tools and interface designs. In the first chapter, we present results from studying learners' visual behavior while engaging with tangible and virtual 3D representations of objects. This is a replication of a recent study, and we extended it using eye tracking. By analyzing the visual behavior, we confirmed the original study results and added more quantitative explanations for the corresponding learning outcomes. Among other things, our results indicated that the users allocate similar visual attention while analyzing virtual and tangible learning material. In the next chapter, we present a user study's outcomes wherein participants are instructed to classify a set of URLs wearing an eye tracker. Much effort is spent on teaching users how to detect malicious URLs. There has been significantly less focus on understanding exactly how and why users routinely fail to vet URLs properly. This user study aims to fill the void by shedding light on the underlying processes that users employ to gauge the UR L's trustworthiness at the time of scanning. Our findings suggest that users have a cap on the amount of cognitive resources they are willing to expend on vetting a URL. Also, they tend to believe that the presence of "www" in the domain name indicates that the URL is safe.

关键词： Color   Stars & galaxies   Software   Supernovae   Star & galaxy formation   Light   Observatories   Morphology   Calibration   Astronomy   Astrophysics   Biology   Computer science   Physics  

摘要： The use of Type Ia Supernovae (SNe Ia) as astronomical distance indicators relies on their intrinsically bright and homogeneous luminosities. By applying empirical relationships to remove any intrinsic, first-order variation in brightness between individual SNe Ia, the apparent brightness of these objects is used to determine a relative measure of distance. Upcoming surveys like the Vera C. Rubin Observatory's Legacy Survey of Space and Time (LSST) will observe on order 100,000 SNe Ia, representing an order of magnitude increase over previous surveys. LSST also promises to provide an impressive sub-percent level of precision between individual measurements. In this work, I present research targeted at two specific challenges faced by SN Ia research in the LSST era. First, I classify SNe Ia that exhibit non-standard photometric behavior, such as lower luminosities and faster evolution of brightness over time. With LSST promising on order a million new SNe over a 10-year survey, spectroscopic classifications will be possible for only a small subset of observed targets. As such, photometric classification has become increasingly important in preparing for the next generation of astronomical surveys. Using observations from the Sloan Digital Sky Survey II (SDSS-II) SN Survey, I apply an empirically based classification technique targeted at identifying SN 1991bg-like SNe in photometric data sets and classify 16 previously unidentified 91bg-like SNe. Furthermore, I show that these SNe are preferentially found at a further physical distance from the center of their host galaxies and in host environments with an older average stellar age. Second, I discuss the impact of atmospheric variability on the calibration of LSST observed SNe Ia. LSST will incorporate multiple calibration systems designed to estimate the atmospheric state and isolate systematic errors, including a GPS to quantify the time-dependent column density of precipitable water vapor (PWV) over the observator

关键词： Computer science   Artificial intelligence   Mathematics   Morphology   Neurosciences  

摘要： Background The study of mindfulness, or intentional awareness and non-judgmental acceptance of moment-to-moment experience, has grown exponentially over the past two decades. Despite the growth and innovation in the science and its multi-sectorial implementation, mindfulness is still trained and cultivated vis-à-vis centuries-old traditional meditation practices. Although proven and effective, nevertheless, these practices require effort and discipline, and are associated with various difficulties and barriers for sustained engagement and thereby sustained salutary gains. Consequently, novel methods have recently been developed to assist practitioners with these barriers and difficulties, including neuromodulation (e.g., Transcranial Direct Current Stimulation (tDCS)). Although Transcranial Alternating Current Stimulation (tACS), may have great promise for augmenting mindfulness meditation, due to its capacity to selectively target specific frequencies of interest as well as entrain endogenous oscillations, critical gaps in our understanding of the electrophysiology of mindfulness must be addressed. Indeed, although there is extensive study todate of electrophysiological correlates of mindfulness meditation, we have limited understanding of the systems-level oscillatory activity that characterizes mindfulness meditation. Such systems-level knowledge is critical for the discovery of electrophysiological patterns that best distinguish mindful states from other mental states and identification of possible candidates for neuromodulation broadly and tACS specifically. The present study is thereby focused on advancing such systems-level insights. Machine Learning (ML) algorithms are designed for pattern recognition in complex and highdimensional data, and thereby constitute a promising method for research of such systems-level patterns in electrophysiological activity. Preliminary efforts have been made to utilize these ML algorithms for classification of mindful states.

关键词： Computer science   Artificial intelligence  

关键词： Computer engineering   Artificial intelligence   Computer science  

摘要： While significant progress has been made to craft Deep Neural Networks (DNNs) with super-human recognition performance, their reliability and robustness in challenging operating conditions is still a major concern. In this work, we study multiple facets of the DNN robustness problem by pursuing two main threads of research. The key methodological linkage throughout our investigations is the consistent design/development/utilization/deployment of Adversarial Machine Learning techniques, which have remarkable abilities to both degrade and enhance model performance. Our ultimate goal is to help construct the more safe and reliable models of the future. In the first thread of research, we take the perspective of an adversary who wishes to find novel and increasingly potent ways to fool current DNN models. Our approach is centered around the development of a feature space attack, and the construction of novel adversarial threat models that work to reduce required knowledge assumptions. Interestingly, we find that a transfer-based blackbox adversary can be significantly more powerful than previously believed, and can reliably cause targeted misclassifications with imperceptible noises. Further, we find that the attacker does not necessarily require access to the target model's training distribution to create transferable attacks, which is a more practically concerning scenario due to the reduction of required attacker knowledge. Along the second thread of research, we take the perspective of a DNN model designer whose job is to create systems capable of robust operation in ``open-world'' environments, where both known and unknown target types may be encountered. Our approach is to establish a classifier + out-of-distribution (OOD) detector system co-design that is centered around an adversarial training procedure and an outlier exposure-based learning objective. Through various experiments, we find that our systems can achieve high accuracy in extended operating condition

关键词： Computer science   Artificial intelligence  

摘要： Automatic identification of speaker state is essential for spoken language understanding, with broad potential in various real-world applications. However, most existing work has focused on recognizing a limited set of emotional states using cues from a single modality. This thesis describes my research that addresses these limitations and challenges associated with speaker state identification by studying a wide range of speaker states, including emotion and sentiment, humor, and charisma, using features from speech, text, and visual modalities. The first part of this thesis focuses on emotion and sentiment recognition in speech. Emotion and sentiment recognition is one of the most studied topics in speaker state identification and has gained increasing attention in speech research recently, with extensive emotional speech models and datasets published every year. However, most work focuses only on recognizing a set of discrete emotions in high-resource languages such as English, while in real-life conversations, emotion is changing continuously and exists in all spoken languages. To address the mismatch, we propose a deep neural network model to recognize continuous emotion by combining inputs from raw waveform signals and spectrograms. Experimental results on two datasets show that the proposed model achieves state-of-the-art results by exploiting both waveforms and spectrograms as *** to the higher number of existing textual sentiment models than speech models in low-resource languages, we also propose a method to bootstrap sentiment labels from text transcripts and use these labels to train a sentiment classifier in speech. Utilizing the speaker state information shared across modalities, we extend speech sentiment recognition from high-resource languages to low-resource languages. Moreover, using the natural verse-level alignment in the audio Bibles across different languages, we also explore cross-lingual and cross-modality sentiment transfer. In the se

关键词： Computer science   Information technology   Artificial intelligence  

摘要： This dissertation has two main objectives -- solving power and latency issues in mobile augmented reality. For power, we showcase the power drain due to the two heaviest components -- simultaneous localization and mapping (SLAM) and deep convolutional neural networks (DNNs) and design solutions to reduce the power consumption on mobile devices. Our single-user solution is to use DNNs as needed, to detect new objects or recapture objects that significantly change in appearance, and otherwise depend on low-power object tracking. For multi-user solutions, we use peer-to-peer communications to exchange key information among devices, and finally assign roles to each of them -- primary or secondary. A primary device continuously tracks target objects and shares their information to slaves. Secondary devices do not need SLAM or DNN but leverage the shared information from the master and other lightweight methods to keep track of the objects with high precision, and thus significantly reduce power consumption. In addition, we can rotate the master functionality across participants in order to distribute energy expenditures among them and increase the longevity of the AR experience. For latency, we perform a first-of-its-kind measurement study on both public LTE and industry LTE testbed for two popular multi-user AR applications, yielding several insights such as: (1) The radio access network (RAN) accounts for a significant fraction of the end-to-end latency (31.2%, or 3.9 s median); (2) AR network traffic is characterized by large intermittent spikes on a single uplink TCP connection, resulting in frequent TCP slow starts that can increase user-perceived latency; (3) Applying a common traffic management mechanism of cellular operators, QoS Class Identifiers (QCI), can help by reducing AR latency by 33% but impacts non-AR users. Based on these insights, we propose AR solutions to intelligently adapt IP packet sizes and periodically provide information on uplink data availab

关键词： Chemistry   Artificial intelligence   Bioinformatics   Computer science   Genetics   Pharmaceutical sciences  

摘要： Fragment based drug design plays an important role in the drug discovery process asa way to reduce the complex small molecule space into a more manageable fragmentspace. This thesis explores mathematical techniques and deep learning methods toexplore computational ways of describing and encoding proteins and drug molecules,with the goal of extracting information to predict chemical binding. The initial chaptersreveal and highlight the challenges of modelling protein-ligand interactions to identifythe best computational tools to use. Firstly, the the viability of experimental bio-assay data and statistical machine learning tools are explored. Secondly, differentclustering algorithms are studied for their ability to retain physicochemical informationof molecule encoding and thirdly, a preliminary off-the-shelf deep learning frameworkis proposed to correlate proteins and inhibitor fragments and the emerging problemsare studied. The main project leverages the availability of a custom built deep learningarchitecture to design ChemPLAN-Net - a model that incorporates both the proteindrug target and inhibitor information and learns from the thousands of protein co-crystal structures in the PDB database. Its purpose is to reliably suggest a number ofinhibitor fragments for a novel query protein structure and offer corresponding bindingmodes for future facilitated drug design. The model is validated thoroughly from astatistical, chemical and experimental literature perspective and its applicability isdemonstrated on the kinase and protease protein families.