教学课程资源库 更多>>

关键词： Penaeus monodon   Feeding frequency   Indoor cement ponds   Growth rate  

摘要： This study examined the growth response of black tiger shrimp, Penaeus monodon, reared in indoor cement ponds fed at variable daily feeding frequencies. Five hundred healthy shrimp (1.26 +/- 0.37 g initial weight) were stocked in 12 cement ponds (5.5x2.5x1.5m). Three ponds in each group were assigned to four different feeding frequency i.e., two, three, four and six times/day designated as FF1, FF2, FF3 and FF4, respectively. Artificial diet (40% fishmeal protein) was used to feed at eight percent of total shrimp body weight per day for a total period of 56 days. The shrimp length and weight were measured fortnightly and the feeding rate was adjusted according to the new weight. At the end of the trial, all shrimp remained healthy and active with the survival rate of 70-85%. The specific growth rate (SGR) were the highest in groups FF3 (4.21 +/- 0.53) and FF4 (4.23 +/- 1.04), which were significantly higher than those in FF1 (3.82 +/- 0.49) and FF2 (4.02 +/- 1.03) groups. Whole body proximate composition (protein, ash, moisture and lipid) remained similar in all the treatment groups. It could be concluded that the black tiger shrimp, P. monodon, would be grown in indoor cement ponds effectively by feeding four times day(-1) to achieve significantly better growth.

关键词： Electrical engineering   Computer science   Engineering  

摘要： Machine learning (ML) has attracted a significant amount of attention from the artificial intelligence community. ML has shown state-of-art performance in various fields, such as signal processing, healthcare system, and natural language processing (NLP). However, most conventional ML algorithms suffer from three significant difficulties: 1) insufficient high-quality training data, 2) costly training process, and 3) domain discrepancy. Therefore, it is important to develop solutions for these problems, so the future of ML will be more sustainable. Recently, a new concept, data-efficient ma- chine learning (DEML), has been proposed to deal with the current bottlenecks of ML. Moreover, transfer learning (TL) has been considered as an effective solution to address the three shortcomings of conventional ML. Furthermore, TL is one of the most active areas in the DEML. Over the past ten years, significant progress has been made in TL. In this dissertation, I propose to address the three problems by developing a software- oriented framework and TL algorithms. Firstly, I introduce a DEML framework and a evaluation system. Moreover, I present two novel TL algorithms and applications on real-world problems. Furthermore, I will first present the first well-defined DEML framework and introduce how it can address the challenges in ML. After that, I will give an updated overview of the state-of-the-art and open challenges in the TL. I will then introduce two novel algorithms for two of the most challenging TL topics: distant domain TL and cross-modality TL (image-text). A detailed algorithm introduction and preliminary results on real-world applications (Covid-19 diagnosis and image classification) will be presented. Then, I will discuss the current trends in TL algorithms and real-world applications. Lastly, I will present the conclusion and future research directions.

关键词： Computer science   Computer Engineering   Artificial intelligence  

摘要： Assisted living environments must be able to efficiently and unobtrusively gather information on a person's well-being. Human gaze direction provides some of the strongest indicators of how a person behaves and interacts with their environment. To that end, this thesis proposes a gaze tracking method that uses a neural network regressor to estimate gaze direction from facial keypoints and integrates them over time using various temporal methods, specifically through moving averages and a Kalman filter. Our gaze regression model uses confidence gated units to handle cases of keypoint occlusion and is able to estimate its own prediction uncertainty. This approach makes it possible to understand gaze direction patterns over time, which then can be used in the assessment of the well-being of individuals in assisted living environments. Experimental results on a dataset collected in an assisted living facility demonstrate that our gaze regression network performs on par with a complex, dataset-specific baseline, while its uncertainty predictions are highly correlated with the actual angular error of corresponding estimations. Furthermore, evaluations of our temporal integration methods on the assisted living facility dataset and on a publicly available gaze estimation dataset show promising results for more accurate and stable gaze predictions.

关键词： Algorithms   Remote searching   Portable computers   Data replication   User behavior   Servers   Traffic congestion   Computer science   Transportation  

摘要： Data intensive applications are usually designed using a multi-tier architecture that comprises a web tier, an application tier and a backend database tier. To process requests, application servers fetch data from the backend database. For faster processing, the accessed data or objects are cached at the application servers which is known as application caching. To fully utilize the aggregated cache space of all application server caches, they can connect together to build a cooperative cache. In such a case, an application server can read an object from its local cache or from a remote cache. While a cooperative cache can reduce frequent access to the backend database, local cache access is still preferable over remote access because it is much faster. However, achieving a high rate of local cache access is a challenging task. Current enterprise workloads are complex and dynamic; user requests typically access several objects, different requests access overlapping object sets, object access is highly skewed and object popularity changes over time. In this thesis, we propose mechanisms to increase the efficiency of cooperative cache solutions. In a first step, we analyze a real workload to understand the major characteristics of its type of requests and how they access objects. Furthermore, we develop a suite of request and object partitioning approaches that partition requests across servers and objects across caches to achieve a high local cache hit as well as balance application loads. We further augment the data partition techniques with a novel data replication technique that replicates objects into caches of the application servers that need them the most. While doing so, we carefully consider different kinds of overhead such as the need to update all replicas, the overhead of different consistency protocols and the fact that cache space is limited. Both are purely distributed as well as our replication solutions support dynamic workloads where access pattern

关键词： Software   Energy efficiency   Cooling   Green buildings   Energy modeling   Computer science   Energy  

摘要： Building energy performance is a function of numerous building parameters. In this study, sensitivity analysis on twenty parameters is performed to determine the top three parameters which have the most significant impact on the energy performance of buildings. Actual data from two fully operational commercial buildings were collected and used to develop a building energy model in eQUEST. The model is calibrated using Normalized Mean Bias Error (NMBE) and Coefficient of Variation of Root Mean Square Error (CV(RMSE)) method. The model satisfies the NMBE and CV(RMSE) criteria set by the American Society of Heating, Refrigeration, and Air-Conditioning (ASHRAE) Guideline 14, Federal Energy Management Program (FEMP), and International Performance Measurement and Verification Protocol (IPMVP) for building energy model calibration. The values of the parameters are varied in two levels, and then the percentage change in output is calculated. Fractional factorial analysis on eight parameters with the highest percentage change in energy performance is performed at two levels in a statistical software JMP. For Building A, top 3 parameters from percentage change method are: Heating setpoint, cooling setpoint and server room. From fractional factorial design, top 3 parameters are: heating setpoint (p-value= 0.00129), cooling setpoint (p-value= 0.00133), and setback control (p-value= 0.00317). For Building B, top 3 parameters from both methods are: Server room (p-value= 0.0000), heating setpoint (p-value= 0.00014), and cooling setpoint (p-value= 0.00035). If the best values for all top three parameters are taken simultaneously, energy efficiency improves by 29% for Building A and 35 % for Building B.

关键词： Computer science   Artificial intelligence   Information technology   COVID-19  

摘要： Distributed denial of service (DDoS) is a network-based attack where the aim of the attacker is to overwhelm the victim server. The attacker floods the server by sending enormous amount of network packets in a distributed manner beyond the server’s capacity and thus causing the disruption of its normal service. In this dissertation, we focus to build intelligent detectors that can learn by themselves with less human interactions and detect DDoS attacks accurately. Machine learning (ML) has promising outcomes throughout the technologies including cybersecurity and provides us with intelligence when applied on Intrusion Detection Systems (IDSs). In addition, from the state-of-the-art ML-based IDSs, the Ensemble classifier (combination of classifiers) outperforms single classifier. Therefore, we have implemented both supervised and unsupervised ensemble frameworks to build IDSs for better DDoS detection accuracy with lower false alarms compared to the existing ones. Our experimentation, done with the most popular and benchmark datasets such as NSL-KDD, UNSW-NB15, and CICIDS2017, have achieved at most detection accuracy of 99.1% with the lowest false positive rate of 0.01%. As feature selection is one of the mandatory preprocessing phases in ML classification, we have designed several feature selection techniques for better performances in terms of DDoS detection accuracy, false positive alarms, and training times. Initially, we have implemented an ensemble framework for feature selection (FS) methods which combines almost all well-known FS methods and yields better outcomes compared to any single FS method. The goal of my dissertation is not only to detect DDoS attacks precisely but also to demonstrate explanations for these detections. Interpretable machine learning (IML) technique is used to explain a detected DDoS attack with the help of the effectiveness of the corresponding features. We also have implemented a novel feature selection approach based on IML which he

摘要： It has been projected that the U.S. economy will have a surplus of computer science jobs in the future to keep up with innovation and the evolution of the digital world, and more computer science professionals will be needed to fill these jobs. In the past 15–20 years, efforts have been underway that focus on the recruitment and retention of students in computer science within elementary, secondary, and postsecondary education. However, only in the past 10 years have these efforts been focused on underrepresented students— those who are not White or Asian—who are underrepresented in the field of computer science. In education, more programs should focus on these groups to fill future needs with respect to growth in industry, representation in innovative solutions, and social mobility. Studies have shown that a student with a positive computer science identity is more likely to persist in the field and graduate from higher education with a major in computer science. Therefore, this particular study explored factors before and during college that underrepresented students have experienced that they perceive to have contributed to the development of a positive computer science identity. Five overarching themes emerged during data analysis: a) coping with diversity in the classroom; b) engaging in high-quality co-curricular and extra-curricular activities; c) teachers and faculty who care; d) family support and role models; e) individual student attributes; and f) computer science programs designed with intentionality. Based on these findings, recommendations of practice are provided for K–12 school districts as well as higher education institutions to systemically improve the experience and inclusivity of computer science in education. Finally, a capstone project was completed after this study, which involved creating a Strategic Vision for computer science at the Boston Public School district, with the goal of encouraging more Boston Public School students to see them

关键词： Computer science   Bioinformatics   Artificial intelligence  

摘要： The preponderance of Electronic Health Records (EHRs) motivates and enables the exploitation of predictive measures, e.g., Artificial Intelligence (AI), in support of personalized medicine. These measures aid physicians in their daily clinical practice by predicting possible disease diagnoses and outcomes of drug prescriptions. With well studied data mining techniques and machine learning models, e.g., deep neural networks, building such systems seems relatively straightforward. However, in practice vast difficulties exists. Real-world EHRs are noisy and are abundant with temporal relationships, hindering deep neural networks performance, preventing clinical applicability. Equally important for clinical acceptance is interpretability; clinicians must understand the reasoning behind a predicted outcome. Unfortunately, existing approaches fail to achieve both high accuracy and model interpretability. I develop graphical representations of EHRs for better model interpretability. Additionally, the developed graph kernels achieve noise resistance and obtain superior accuracy. Distance-based kernel learning with rigorous empirical evaluation concludes a guideline of distance selection for variety of data distribution. Thus, using the proposed approach, I address both aforementioned issues.

关键词： Computer science