教学课程资源库 更多>>

关键词： Computer science  

摘要： Peer loss [1] is a new family of loss functions proposed to deal with the problem of learning with noisy labels. It claims to handle a wide range of label noise in binary classification tasks without explicitly estimating the noise rates. Numerical experiments demonstrate the effectiveness of peer loss. However, its extension to the multi-class classification remains unclear, and its working mechanism is not fully understood. In this thesis, we study the theory of peer loss from three distinct perspectives. Follow the original method in [1], we first consider the multi-class extension of peer loss and investigate its noise tolerance properties. From this perspective, we see peer loss as a class of loss functions inspired by the truthful and proper scoring rules in the peer prediction literature. It turns out that this perspective is a static one and cannot provide a satisfactory explanation of how peer loss works in practical training. To gain an intuitive picture of the working mechanism, we further develop a divergence perspective towards peer loss, expressing it as the difference between two KL divergences. Thus, we recognize that peer loss has a built-in regularization effect, encouraging the model to make confident predictions. This regularization effect partially explains why peer loss works well under the label noise, as the existence of noise often blurs the data distribution and makes the resulting model prediction uncertain. Finally, we show that peer loss potentially suggests a new type of risk in decision theory, i.e., the correlation risk. This new perspective helps us to understand better what the model learns when trained with peer loss. To complete the discussion of the correlation risk perspective, we develop a novel method to investigate the training dynamics of peer loss. This dynamical analysis justifies that with peer loss, the resulting model tends to grasp the positive correlations in the training datasets. In addition to the theoretical analy

关键词： Task analysis   Pediatrics   Machine learning   Medical services   Virtual environments   Computer science   Medical diagnostic imaging   Attention deficit and hyperactivity disorder   virtual reality   neuro-behavior   machine learning   assessment  

摘要： Attention-deficit/Hyperactivity disorder(ADHD) is a common neurodevelopmental disorder among children. Traditional assessment methods generally rely on behavioral rating scales (BRS) performed by clinicians, and sometimes parents or teachers. However, BRS assessment is time consuming, and the subjective ratings may lead to bias for the evaluation. Therefore, the major purpose of this study was to develop a Virtual Reality (VR) classroom associated with an intelligent assessment model to assist clinicians for the diagnosis of ADHD. In this study, an immersive VR classroom embedded with sustained and selective attention tasks was developed in which visual, audio, and visual-audio hybrid distractions, were triggered while attention tasks were conducted. A clinical experiment with 37 ADHD and 31 healthy subjects was performed. Data from BRS was compared with VR task performance and analyzed by rank-sum tests and Pearson Correlation. Results showed that 23 features out of total 28 were related to distinguish the ADHD and non-ADHD children. Several features of task performance and neuro-behavioral measurements were also correlated with features of the BRSs. Additionally, the machine learning models incorporating task performance and neuro-behavior were used to classify ADHD and non-ADHD children. The mean accuracy for the repeated cross-validation reached to 83.2%, which demonstrated a great potential for our system to provide more help for clinicians on assessment of ADHD.

ISBN: (数字)9780271088167 ISBN: (纸本)9780271088167

摘要： Winner of the 2006 Andrés Bello Award for Memory and Ibero-American Thought In this evocatively written book, Germán Patiño Ossa presents the cultural universe and national identities of Colombia through the lens of traditional cuisine. Focusing on the Cauca Valley, a fertile area in southwestern Colombia where Spanish, Native American, and African communities converged over the centuries, Patiño Ossa studies the food of these communities and its place in the region’s culture. Using Jorge Isaacs’s nineteenth-century Romantic novel María as a realistic source for cultural practices among Colombia’s slaveholding elite, Patiño Ossa examines cooking, kitchens, and the division of labor; flora and fauna; agriculture, hunting, and fishing; hospitality; slavery; and literature. Through the community of Afro-descendants who appear in Isaacs’s novel, Patiño Ossa shows how this culinary culture, originating in the cookstoves used by female black slaves, resulted in the Creole fusions that characterize this geographical region of Latin America. Cooking and food, as Patiño Ossa eloquently demonstrates, are essential for us to understand the process of the formation of culture and the origins, evolution, and effects of transculturation. Innovative, engaging, and accompanied by an introductory preface by the author, this English-language edition of Patiño Ossa’s prizewinning book is a model for food and cultural studies that will appeal to scholars, students, and the intellectually curious.

关键词： Computer science  

摘要： Low-rank matrix approximation is extremely useful in the analysis of data that arises in scientific computing, engineering applications, and data science. However, as data sizes grow, traditional low-rank matrix approximation methods, such as singular value decomposition (SVD) and column pivoting QR decomposition (CPQR), are either prohibitively expensive or cannot provide sufficiently accurate results. A solution is to use randomized low-rank matrix approximation methods such as randomized SVD and randomized LU decomposition on extremely large data sets. In this dissertation, we focus on the randomized LU decomposition method. First, we employ a reorthogonalization procedure to perform the power iteration of the existing randomized LU algorithm to compensate for the rounding errors caused by the power method. Then to solve the fixed precision low rank approximation problem, we block the existing randomized LU algorithm. Our proposed randomized blocked LU algorithm is accurate and has comparable speed with randomized blocked QB algorithm by Martinsson and Voronin. Then we propose a novel randomized LU algorithm, called PowerLU, for the fixed low-rank approximation problem. PowerLU allows for an arbitrary number of passes of the input matrix, v ≥ 2. Recall that the existing randomized LU decomposition only allows an even number of passes. We prove the theoretical relationship between PowerLU and the existing randomized LU. Numerical experiments show that our proposed PowerLU is generally faster than the existing randomized LU decomposition, while remaining accurate. We also propose a version of PowerLU, called PowerLU_FP, for the fixed precision low-rank matrix approximation problem. PowerLU_FP is based on an efficient blocked adaptive rank determination Algorithm 18 proposed in this dissertation. We present numerical experiments that show that PowerLU_FP can achieve almost the same accuracy and is faster than the randomized blocked QB algorithm. We finally propose a

关键词： Computer Science   HPC   Reinforcement learning   High performance computing   scheduling   reinforcement learning asynchronous actor critic   A3C   Cost-aware  

摘要： In this thesis we introduce a new scheduling algorithm MARS based on a cost-aware multi-scalable reinforcement learning approach, which serves as an intermediate layer between HPC resource manager and user application workflow. MARS ensembles the pre-generated models from users’ workflows and decides on the most suitable strategy for optimization. A whole workflow application would be split into several optimized sub-tasks. Then, based on a pre-defined resource management plan, a reward will be generated after executing a scheduled task. Lastly, MARS updates the Deep Neural Network (DNN) model for future use. MARS is designed to be able to optimize the existing models through reinforcement mechanism. MARS can adapt to the shortage of training samples by optimizing the performance through combining small tasks together or switching between pre-built scheduling strategies such as Backfilling, SJF, etc., and choosing the most suitable approach. After testing MARS using different real-world workflow traces, results shows that MARS can achieve between 5%-60% better performance against the other approaches.

关键词： Psychology   Computer science  

摘要： Categorizing and stereotyping others are unavoidable features of human life. However, despite decades of research, there is still no complete consensus on the dimensions that perceivers use to make sense of others. Various models have proposed dimensions such as warmth, competence, socioeconomic status, and progressive-conservative beliefs, but how to integrate these models and whether other dimensions should also be modeled remains controversial. A limitation of current models is their reliance on predetermined numerical ratings on dimensions that are explicitly queried. Here I develop and introduce free response measures of stereotype content in order to study more spontaneous impressions. This approach has several advantages over traditional metrics: (a) it circumvents researcher biases in the selection of evaluative dimensions, (b) it provides information about the salience of evaluative dimensions, and (c) it allows for an examination of additional relevant cognitive processes (e.g., reaction times). Across three chapters, I describe the process of developing text analysis instruments, their application to studying information-gathering processes (in an “adversarial” collaboration), and a spontaneous stereotype content model, a taxonomy of free-response stereotypes. These chapters provide evidence for the role of spontaneous stereotypes in an integrative and generative framework, uncovering moderators of stereotype dimension priority, dimensional usage rates and intercorrelations, as well as stereotype processes and properties that improve our understanding of person perception. This research has implications for the measurement of psychological dimensions in text (e.g., hateful stereotypes in social media), the integration of adversary models of social cognition, and the discovery of novel constructs that may better model the complexity of an increasingly diverse social world.

关键词： Management   Information technology   Computer science  

摘要： Historically, managers of small- and medium-sized enterprises (SMEs) have had concerns regarding cloud computing and cybersecurity. Their resistance to using cloud computing has influenced their ability to do business effectively and to compete with businesses that use cloud computing. The purposes of this descriptive phenomenological study were to explore the lived experiences and perceptions of SME managers that might influence their decisions to adopt cloud computing. Watson’s concept of resistance to change and Davis, Bagozzi, and Warhaw’s technology acceptance model were the conceptual frameworks that guided this qualitative study. Data collection consisted of conducting 16 semi-structured interviews with open-ended questions with SME managers. Data were coded and compared to identify emerging themes among responses. The findings showed positive cloud-based experiences, such as availability of training, flexibility, efficiency, cost-effectiveness, ease of use, and assurance data security. The findings indicated some negative experiences with cloud-based applications, such as fear of cybercrime, expensive licenses, software complexity, and concern for data security. The results of the study may lead to positive social change by providing a better understanding of the perceptions and experiences that influence SME managers’ decisions regarding the adoption of cloud-based computing technology. Such understanding could be used to provide resources to allay the fears of SMEs and encourage them to be more willing to consider cloud computing.

关键词： Statistical physics   Artificial intelligence   Computer science  

摘要： Due to their exceptional properties, rare earth elements (REEs) are critical to technological innovation in renewable energy production, electronics, health care, and national defense. They make up key components for many applications in the above areas. Many countries rely upon rare earth element imports. The high demand for rare earth elements has led to the development of alternative methods for exploration and capture. Coal has been labeled a viable potential source of rare earth elements and yttrium (REY). Statistical evaluation of REY concentrations and the properties of various coal samples is critical for successful characterization. The USGS COALQUAL database Version 3.0 is an industry standard database for coal research that contains 7658 non-weathered, full-bed coal samples from the United States. 5485 of these samples contain a full spectrum of REY concentrations. The data quality in the COALQUAL database will be analyzed to ensure that the data is reliable, and characteristics will be analyzed using conventional statistical methodology. This methodology includes accounting for samples with REY concentrations below the lowest limits of detection. Mean concentrations for each REY will be adjusted to fit a distribution of mean REY concentrations from the National Coal Resources Data System (NCRDS) normalized by the Upper Continental Crust standard dataset of REY mean concentrations. All samples are classified as unpromising or promising using total rare earth oxide concentration and the ratio of critical REYs to excess REYs called the outlook coefficient. Machine learning is a powerful tool that can utilize data to classify new data points added to a database based on data attributes. A machine learning model was developed to use existing data from the COALQUAL database to train and test algorithms to classify coal samples as unpromising or promising based on the samples ASTM ash percentage. The 5485 adjusted coal samples from the COALQUAL database were us

关键词： Computer science  

摘要： Twitter, one of the most popular micro-blogging services, allows users to publish short messages on a wide variety of subjects such as news, events, stories, ideas, and opinions, called tweets. The popularity of Twitter, to some extent, arises from its capability of letting users promptly and conveniently contribute tweets to convey diverse information. Specifically, with people discussing what is happening outside in the real world by posting tweets, Twitter captures invaluable information about real-world news and events, spanning a wide scale from large national or international stories like a presidential election to small local stories such as a local farmers market. Detecting and extracting small news and events for a local place is a challenging problem and is the focus of this thesis. In particular, we explore several directions to extract and detect local news and events using tweets in Twitter: a) how to identify local influential people on Twitter for potential news seeders; b) how to recognize unusualness in tweet volume as signals of potential local events; c) how to overcome the data sparsity of local tweets to detect more and smaller undergoing local news and events. Additionally, we also try to uncover implicit correlations between location, time, and text in tweets by learning embeddings for them using a universal representation under the same semantic space. In the first part, we investigate how to measure the spatial influence of Twitter users by their interactions and thereby identify the locally influential users, which we found are usually good news and event seeders in practice. In order to do this, we built a large-scale directed interaction graph of Twitter users. Such a graph allows us to exploit PageRank based ranking procedures to select top local influential people after innovatively incorporating in geographical distance to the transition matrix used for the random walking. In the second part, we study how to recognize the unusualness i

关键词： Computer science  

摘要： The ability to recognize human actions in video has many potential applications. Human action recognition, however, is tremendously challenging for computers due to the complexity of video data and the subtlety of human actions. Most current recognition systems still flounder on the inability to separate human actions from contextual factors that usually dominate subtle human actions in realistic video. This thesis investigates several approaches to learn factorized representations that can focus on the actual human action elements instead of the contextual factors. I will start by describing an unsupervised and a supervised method for learning the factorized representation. I will then describe a weakly-supervised method that does not require detailed annotation. This method exploits the benefits of conjugate samples, which are video clips that are contextually similar to human action samples, but do not contain the action. This method can: (1) explicitly factorize human actions from the co-occurring context; (2) deliberately build a model for human actions and a separate model for all correlated contextual elements; and (3) effectively combine the models for human action recognition. After that, I will present a method for grounding the action and context factors, localizing the spatiotemporal regions that `define' the actions. This method is developed based on a novel attentional mechanism that utilizes conjugate samples to spatially and temporally separate human actions from the co-occurring contextual factors. Finally, I will present a better way of modeling long-range action-context relations of video through relational inference over a graph-based model. These proposed methods can be used to build human action classifiers with higher accuracy and better interpretability.