Faculty of Computer Science (Fredericton)
Pages
-
-
An empirical study on comparison between transfer learning and semi-supervised learning
-
by Ao Cheng, Transfer learning and semi-supervised learning attract considerable attention since the traditional machine learning methods yield insufficient performance in many practical applications with scarce labeled data. In such cases, knowledge transfer from a related domain or information extraction of unlabeled data, if done properly, would significantly upgrade the classifier by avoiding costly labeling expense. These two branches of machine learning try to use auxiliary data to make up for the shortage of labeled instances. In this study, a set of experiments are conducted on several typical algorithms for both transfer learning and semi-supervised learning to test whether these auxiliary data should be beneficial. The empirical study shows that these auxiliary instances may not be permanently helpful comparing to the traditional learning methods. However, when a special situation with an extremely small number of labeled instances arises, the auxiliary data would improve the performance significantly. The internal characteristics influencing the performance in each branch is also explored in this study., Electronic Only.
(UNB thesis number) Thesis 9246.
(OCoLC) 960967924, M.C.S., University of New Brunswick, Faculty of Computer Science, 2013.
-
-
An empirical study on machine learning for tweet sentiment analysis
-
by Hao Tao, Tweet sentiment analysis has been an effective and valuable technique in the sentiment analysis domain. As the most widely used approach for tweet sentiment analysis, machine learning algorithms work well on the sentiment classification, just as they have been successfully applied for many other purposes. In this thesis, we conduct a systematic and thorough empirical study on the machine learning algorithms for tweet sentiment analysis, and expect to provide a guideline for applying machine learning algorithms for tweet sentiment analysis. Based on our experiments, we found that the Support Vector Machine (SVM) and the Random Forest (RF) work better than Maximum Entropy (MaxEnt), Adaptive Boosting (AdaBoost) and Naive Bayes on tweet sentiment analysis. For the pre-processing methods, stop words removal can improve the performance of classifiers obviously, and the combination of bi-grams + SentiWordNet + Stop words removal is the most effective pre-processing method combination in our experiments.
-
-
An intelligent malware classification framework
-
by Elaheh Biglar Beigi Samani, Malicious software or malware has risen to become a primary source of most of the attacks taking place across the Internet over the last decades. This prevalence of new malware, for which signatures are not available, along with the challenge of anti-malware software to keep up with the continuous stream of new malware, has made the adoption of classification/-clustering approaches necessary. Machine-learning methods have been excessively applied to classify or cluster malware into families, based on different features derived from static or dynamic review of the malware. While these approaches demonstrate promise, they are themselves subject to a growing array of countermeasures. In this work, we propose a framework to enhance the traditional machine learning-based classification by utilizing high-level domain knowledge. We outline major behaviours of Windows malware from an analyst's point of view and provide possible methods (rules) to extract them from the output of static and dynamic analysis tools. We also take advantage of memory forensics to extract other stealthy aspects of an executable, which otherwise remain undetected. Our comparative experimental results with the state-of-the-art malware classification approaches, confirm the effectiveness of our framework by an average classification accuracy of 81%, while leaving only 0.5% of samples unlabeled.
-
-
Analyzing mobile games using a social network analysis approach
-
by Nimat Onize Umar, In recent times, analysis of user-generated data acquired from social media has proven to be beneficial in helping organizations make decisions about their businesses. This forms a basis for exploring other areas where social network analysis might be useful. In this report, I decided to look at the mobile games industry and see how accurate social network analysis can be, in making predictions of possible real world outcomes. Three approaches considered were how often a game is mentioned in social media (Frequency Count), sentiments attached to each game (Sentiment Analysis), and a game’s position with other games in the network (Centrality Measures). Furthermore,using multiple linear regression analysis, five different predictive models were created by combining the three approaches. Finally, an evaluation of these approaches was done by performing correlation analysis between the rankings produced by each approach with the rankings in the Google Playstore. The best approach had a correlation coefficient of 0.58, which meant that the predictive ability of social network analysis for this industry is moderate., Electronic Only
(UNB thesis number) Thesis 9501
(OCoLC)958877849, M.C.S University of New Brunswick, Faculty of Computer Science, 2014.
-
-
Application of reinforcement learning to autonomous aircraft in partially observable environments
-
by Dmitry Shcherbakov, This thesis provides a brief survey of the mathematical background of the reinforcement learning (RL) method and sketches the current state of arguably the most developed area of RL application, to the problem of controlling autonomous vehicles (self-driven car-like vehicles). This is then compared to RL solutions in autonomous piloting tasks. Contrasting the two shows that the latter may benefit from a common framework for RL applications. We propose a framework for autonomous piloting tasks, provide a detailed description of the toolkit available for the framework, and perform an experiment with described instruments. The experiment is designed to determine whether a small fixed memory window can mitigate the adverse influence of such unobserved factors as wind bursts and turbulence. Tests show that the memory mechanism that encapsulates control feedback is an informative input for the learning agent, as long as the unobserved factors affect control behavior significantly.
-
-
Augmented biofeedback for partial weight-bearing learning
-
by Ian Smith, Assistive devices, including canes and crutches, are used in partial weight-bearing (PWB)–offloading weight from limbs weakened by disease or injury to promote recovery and prevent reinjury. While it is important to accurately offload weight to target loads prescribed by healthcare providers, current training methods result in poor compliance. It is currently unknown how to most effectively provide feedback during training to allow users accurate execution of the skill later on. In this work, three studies were conducted to investigate the effects of feedback modality, delay, and resolution on both regulation and learning of PWB while stationary and during gait. Results indicate that concurrent feedback is best suited for continuous skill regulation whereas retrospective feedback is preferable for training PWB, and that task-specific training is critical for compliance. This work presents design guidelines for improved clinical PWB training methods and highlights the importance of researching retrospective motor learning methods.
-
-
Authentication in a body area network (BAN) using OpenSSL
-
by Josiah Mololuwa Oyinola, Internet of Things (IoT), which enable the connection and communication of objects (Things) over the internet, have received considerable attention in recent years. The internetisthe main medium (backbone) of communication, while the things are smart devices, machineries, industry level equipment, etc., which generate data to share and process for some intelligent decision making. Ever since the term IoT was coined out and described by Kevin Ashton in 1999, industries have adopted the idea of IoT and began integrating it into their product development process widely.
As a popular application scenario of IoT, body area network (BAN) is one IoT network formed by body sensors, which can sense the health related data, deliver the data to the remote eHealthcare center for a better health monitoring through some gateways over the Internet. However, due to the congestion of the internet and the fast rise in cyberattacks, it is very important to secure packet exchange via advanced level encryption methods to prevent possible session hijacking, Man-In- The-Middle (MITM) attacks, cross site scripting, etc. Fortunately, there are numerous encryption standards available. In this study, in order to address the cyber security issues in body area network, we will be considering the OpenSSL, one popular software library which is open source for user revision.
According to openssl.org, OpenSSL is a robust, commercial-grade, and full-featured toolkit for the Transport Layer Security (TLS) and Secure Sockets Layer (SSL) protocols. It is also a general- purpose cryptography library, and it islicensed under the Apache-style license which makes it free to use by everyone.
Concretely, the report examines the OpenSSL cryptographic architecture, propose and implement a “layer-in-layer” level cryptographic model in a bid to secure our communication while interfacing with generated data between the BAN and IoT. The study involves dissecting the selected authentication process, seeing how it ticks, and why, look into its applications and its setbacks. The ultimate goal of this study is to create an extra layer of security on the same technology and put it to use. It is expected that this strategy will make the authentication algorithm more secure., Electronic Only.
"A Report Submitted In Partial Fulfillment Of The Requirement For The Degree Of Master of Computer Science In the Graduate Academic Unit of Computer Science".
Title page has family name before given names.
-
-
Authorship attribution in the dark web
-
by Britta Sennewald, This thesis is about authorship attribution (AA) within multiple Dark Web forums and the question of whether AA is possible beyond the boundaries of a single forum. AA can become a curse for users that try to protect their anonymity and simultaneously become a blessing for law enforcement groups that try to track users. To determine to what extent AA threatens the anonymity of Dark Web users, a dataset of four Dark Web forums was created. Within the analysis, two different approaches are considered: feeding classifiers with posts from two forums, and training classifiers with posts from another forum than what is used for testing. Even for the largest dataset, the author of a post is at least 94% within the top three most likely candidates. This shows that AA can be a danger to the anonymity of Dark Web users across the boundaries of different forums.
-
-
Automatic application performance improvements through VM parameter modification after runtime behavior analysis
-
by Nicolas Markus Neu, This thesis presents an approach to automatically adjust the parameters of a Java application run on the IBM J9 Virtual Machine in order to improve its performance. It works by analyzing the logfile the VM generates and searching for specific behavioral patterns. These patterns are matched against a list of known patterns for which rules exist that specify how to adapt the VM to the given application. Adapting the application is done by adding parameters and changing existing ones, for example to achieve a better heap usage. The process is fully automated and carried out by a toolkit developed for this thesis. The toolkit iteratively cycles through multiple possible parameter sets, benchmarks them and proposes the best alternative to the user. The user can, without any prior knowledge about the Java application or the VM improve the performance of the deployed application., Electronic Only.
(UNB thesis number) Thesis 9426.
(OCoLC) 962306539., M.C.S.,University of New Brunswick, Faculty of Computer Science, 2014.
-
-
Automating post-mortem debugging analysis in Node.js
-
by Anil Hitang, Post-mortem debugging involves analyzing a raw memory dump of either a portion of memory or the whole memory of the system at an instance in time. The process of post-mortem debugging is complex and time consuming, as it requires many operations in order to be executed. Automating post-mortem debugging can potentially improve its effectiveness in localizing software faults or bugs. The aim of this study was to implement an automated post-mortem debugging solution for Node.js from the information available in the bug reports. The automated solution was aimed to automatically process the bug reports and generate more concrete results in order to be used in troubleshooting software faults such as memory-leak problems and optimize memory usage in Node.js applications. Overall, the implementation of an automated solution was only partially achieved in this study., A MASTER'S REPORT SUBMITTED IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF Masters of Computer Science In the Graduate Academic Unit of Computer Science
-
-
Basis enumeration of hyperplane arrangements up to symmetries
-
by Aaron Moss, This thesis details a method of enumerating bases of hyperplane arrangements
up to symmetries. I consider here automorphisms, geometric symmetries
which leave the set of all points contained in the arrangement setwise
invariant. The algorithm for basis enumeration described in this thesis is
a backtracking search over the adjacency graph implied on the bases by
minimum-ratio simplex pivots, pruning at bases symmetric to those already
seen. This work extends Bremner, Sikiric, and Schiirmann's method for basis
enumeration of polyhedra up to symmetries, including a new pivoting
rule for finding adjacent bases in arrangements, a method of computing automorphisms
of arrangements which extends the method of Bremner et al.
for computing automorphisms of polyhedra, and some associated changes to
optimizations used in the previous work. I include results of tests on ACEnet
clusters showing an order of magnitude speedup from the use of C++ in my
implementation, an up to 3x speedup with a 6-core parallel variant of the
algorithm, and positive results from other optimizations.
Pages
Zircon - This is a contributing Drupal Theme
Design by
WeebPal.
