<p>Through the computing advance and GPU (Graphics Processing Unit) availability for math calculation, the deep learning field becomes more popular and prevalent. Object detection with deep learning, which is the part of image processing, plays an important role in automatic vehicle drive and computer vision. Object detection includes object localization and object classification. Object localization involves that the computer looks through the image and gives the correct coordinates to localize the object. Object classification is that the computer classification targets into different categories. The traditional image object detection pipeline idea is from Fast/Faster R-CNN [32] [58]. The region proposal network generates the contained objects areas and put them into classifier. The first step is the object localization while the second step is the object classification. The time cost for this pipeline function is not efficient. Aiming to address this problem, You Only Look Once (YOLO) [4] network is born. YOLO is the single neural network end-to-end pipeline with the image processing speed being 45 frames per second in real time for network prediction. In this thesis, the convolution neural networks are introduced, including the state of art convolutional neural networks in recently years. YOLO implementation details are illustrated step by step. We adopt the YOLO network for our applications since the YOLO network has the faster convergence rate in training and provides high accuracy and it is the end to end architecture, which makes networks easy to optimize and train. </p>

Research on organisational learning (OL) was mainly positioned within the psychological and sociological domains. Past and extant research on OL focused on the behavioural, cognitive and intuitive perspectives in addition to a growing track of research grounded on social theory. So far, a countless number of research studies attempted to address inter-organisational learning (IOL) from various perspectives. However, the lack of understanding of how IOL occurs in networks can be observed due to the social tensions that are created at the inter-organisational level such as free-riding and knowledge leakage. This thesis, therefore, aims to draw theoretical explanations of IOL and how it occurs in learning networks, taking into consideration similarities and contradictions amongst a network’s participating organisations. Towards this end, the thesis employs two theoretical lenses, namely structure-agency and social exchange theories to draw conclusions that provide fresh explanations of how networks are helpful in fostering or hindering learning activities in addition to how reciprocity as an efficacy device mediates IOL dynamics. Positioned within a qualitative vein, the thesis employs an interpretive perspective to collect and analyse empirical evidence. The qualitative data were developed through a mixture of participant observations, semi-structured interviews and casual conversations with network administrators and participants. The data were analysed using thematic analysis which generated codes, following which conclusions were drawn. The main contributions of this article are (1) unfolding the network as agency which provides a fresh understanding of how the agential role of networks mediates IOL and (2) drawing a framework of dimensions of reciprocal exchanges that explains how IOL occurs in networks. The first conclusion of this thesis explained how the agential role is socially constructed and how the interpretive device facilitated this construction. The second conclusion of this thesis explained how reciprocal exchanges mediate IOL and provide a framework that suggested IOL can be better understood through temporal, spatial, directional and symmetrical perspectives.

E-learning has grown to such an extent that paper-based testing is being replaced by computer-based testing otherwise known as e-exams. Because these e-exams can be delivered outside of the traditional proctored environment, additional authentication measures must be employed in order to offer similar authentication assurance as found in proctored, paper-based testing. This dissertation addressed the need for valid authentication in e-learning systems, in e-examinations in particular, and especially in professional certification e-examinations. Furthermore, this dissertation proposed a more robust method for learner authentication during e-examination taking. Finally, this dissertation extended e-learning research by comparing e-examination scores and durations of three separate groups of exam takers using different authentication methods: Online Using Username/Password (OLUP), In-Testing Center (ITC), and Online with Multibiometrics (OLMB) to better understand the role as well as the possible effect of continuous and dynamic multibiometric authentication on professional certification e-examination scores and durations. The sample used in this study was based on participants who were all professional members of a technology professional certification organization. The methodology used to collect data was a posttest only, multiple, non-equivalent groups quasi-experiment, where age, gender, and Information Technology Proficiency (ITP) were also recorded. The analyses performed in this study included pre-analysis data screening, reliability analyses for each instrument used, and the main analysis to address each hypothesis. Group affiliation, i.e. type of authentication methods, was found to have no significant effect on differences among exam scores and durations. While there was a clear path of increased mean e-examination score as authentication method was relaxed, it was evident from the analysis that these were not significant differences. Age was found to have a significant effect on exam scores where younger participants were found to have higher exam scores and lower exam durations than older participants. Gender was not found to have a significant effect on exam scores nor durations. ITP was found to have a significant effect on exam scores and durations where greater scores with the ITP instrument indicated greater exam scores and lower exam durations. This study's results can help organizations better understand the role, possible effect, and potential application of continuous and dynamic multibiometric authentication as a justifiable approach when compared with the more common authentication approach of User Identifier (UID) and password, both in professional certification e-examinations as well as in an online environment.

Power disspiation is a growing concern in VLSI circuits. In this work we model the data dependence of power dissipation by learning an input model which we use for estimation of both switching activity and crosstalk for every node in the circuit. We use Bayesian networks to effectively model the spatio-temporal dependence in the inputs and we use the probabilistic graphical model to learn the structure of the dependency in the inputs. The learned structure is representative of the input model. Since we learn a causal model, we can use a larger number of independencies which guarantees a minimal structure. The Bayesian network is converted into a moral graph, which is then triangulated. The junction tree is formed with its nodes representing the cliques. Then we use logic sampling on the junction tree and the sample required is really low. Experimental results with ISCAS '85 benchmark circuits show that we have achieved a very high compaction ratio with average error less than 2%. As HSPICE was used the results are the most accurate in terms of delay consideration. The results can further be used to predict the crosstalk between two neighboring nodes. This prediction helps in designing the circuit to avoid these problems.

The open source structure of Android applications introduces security vulnerabilities that can be readily exploited by third-party applications. We address certain vulnerabilities at both installation and runtime using machine learning. Effective classification techniques with neural networks can be used to verify the application categories on installation. We devise a novel application category verification methodology that involves machine learning the application permissions and estimating the likelihoods of different categories. To detect malicious patterns in runtime, we present a Hidden Markov Model (HMM) method to analyze the activity usage by tracking Intent log information. After applying our technique to nearly 1,700 popular third-party Android applications and malware, we report that a major portion of the category declarations were judged correctly. This demonstrates the effectiveness of neural network decision engines in validating Android application categories. The approach, using HMM to analyze the Intent log for the detection of malicious runtime behavior, is new. The test results show promise with a limited input dataset (69.7% accuracy). To improve the performance, further work will be carried out to: increase the dataset size by adding game applications, to optimize Baum-Welch algorithm parameters, and to balance the size of the Intent sequence. To better emulate the participant's usage, some popular applications can be selected in advance, and the remainder can be randomly chosen. / Master of Science

This thesis examines the collaborative development of corporate responsibility (CR) standards from the perspective of organisational learning theory. The author proposes that standards development projects can be understood as Network Learning episodes where learning is reflected in changes in structures, interpretations and practices accompanied by learning processes. Network Learning alone is seen as insufficient to reflect the diverse contributions and outcomes in the special case of CR standards. Concepts from multi-stakeholder learning like the role of dissensus in learning and the empowerment of weaker stakeholders are therefore used to create a synthesis of the two theories in a single conceptual framework. This framework is then tested against a pilot case and three case studies of corporate social responsibility (CSR) standards including the development of the new ISO international standard on social responsibility (SR). The data validates and extended this framework to yield a Multi-Stakeholder Network Learning theory capable of describing the how participants and non-participant stakeholders learn in this context. New concepts are generated from the data, like dislocated learning, which demonstrate how participants in the process and those they represent can experience quite different learning outcomes. Stakeholders whose learning is aligned with the learning of their participant representatives truly have a stake in these influential standards. However, where representatives fail to learn from those represented, the latter's stake is diminished. By shedding light on the mechanisms of effective collaborative learning this work contributes to learning theory, the practice of standardization and the normative stakeholder empowerment agenda.

This thesis examines the relationship between individual differences in learning, personality, and social success in two groups of brown capuchin monkeys (Sapajus sp.) housed at the “Living Links Centre for Human Evolution” at Edinburgh Zoo, UK. Being able to learn quickly and efficiently likely helps primates achieve social success (defined here in terms of centrality within a social network), such as acquiring knowledge of others or learning social skills. Therefore, individuals that are better at learning were predicted to have greater social success than other group members. This prediction, however, contrasts with hypotheses generated from two other disciplines at the individual level: 1) the study of behavioural innovation, and 2) the study of individual differences, i.e. “personality”. In terms of behavioural innovation, better learners should have less social success than other group members because they are expected to rely more on problem-solving, rather than physical combativeness or status, to gain access to socioecological resources. In terms of personality, learning should have little or no direct relationship with social success because other individual differences, like sociability and fearfulness, should mediate primates’ social decision making. This thesis investigates each of these hypotheses. Personality was assessed in 127 capuchins from 7 international sites using the Hominoid Personality Questionnaire, and then validated at Living Links (LL) using behavioural codings; this was the first-ever description of personality structure in brown capuchins. Brown capuchins have five personality dimensions: Assertiveness, Openness, Sociability, Neuroticism, and Attentiveness. Ratings were consistent across observers, and predicted relevant behaviours among the LL capuchins over a year later (e.g. scores on Sociability predicted time spent in close proximity to others). “Social success” in the LL capuchins was assessed in terms of centrality in spatial proximity networks. Individual scores on social network centrality were significantly correlated with scores derived from a Principal Components Analysis of eight affiliative and agonistic behaviours among the LL capuchins, indicating that spatial proximity is a reliable measure of the quality of subjects’ social embeddedness within their groups. Social rank and two personality traits (Assertiveness and Sociability) were positively related to network centrality, while another personality trait (Neuroticism) was negatively related to centrality. Sociability was a significant predictor of network centrality even after controlling for social rank and the other personality traits, highlighting the importance of this personality trait in shaping the social success of capuchins beyond that of basic social rules (e.g. kinship, sex, and rank). Individual learning was assessed in the LL capuchins by administering two operant tasks to subjects under conditions of free choice participation. In Task 1, thirteen monkeys participated, and eight individuals met learning criteria (i.e. >80% trials correct over 3 consecutive sessions). In Task 2, fifteen monkeys participated, and five individuals met learning criteria; the monkeys that learned this second task were also among those individuals that learned Task 1. For monkeys that regularly participated in both tasks (i.e. >50% of sessions), their average performances (i.e. % trials correct) were significantly correlated with individual scores on Assertiveness, but not the other four personality traits, or individual differences in attention span during testing, the percent of sessions subjects participated during testing, the amount of scrounging events subjects directed towards others within their social group, or the percent of observation time subjects spent feeding within their main indoor/outdoor enclosures. In terms of social success, relatively better learners had lower social rank and network centrality compared to relatively poor learners. Also, compared to poorer learners, better learners were generally less likely to direct affiliative acts (e.g. grooming, food sharing, coalitionary support) to other group members. Controlling for Assertiveness (i.e. the only variable related to individual differences in subjects’ average learning performance), individual differences in learning performance were no longer significantly related to social rank, network centrality, or the amount of affiliative acts subjects initiated with others. Collectively, such findings contrast the hypothesis that better learners should (concurrently) be more socially successful than poorer learners, and instead are more reflective of hypotheses pertaining to behavioural innovation and/or the study of individual differences. Social rank and certain traits of personality (Assertiveness, Openness, Neuroticism, and Sociability) appear to interact with capuchins’ patterns of social interaction, and one personality trait (Assertiveness) may mediate how individual differences in learning are associated with differences in social success.

Person identification has in the recent years gained notoriety as one of the most powerful ways of extracting information from image data. This thesis is focused on the task of human identification from facial photographs. To solve this task, we employ algorithms based on neural networks, which produce more robust results than traditional algorithms. In this thesis, we studied the common approaches for solving this problem and based on the gathered knowledge we created an architecture of a neural network trained to tackle the task of human identification and verification based on facial photographs. We have then further improved the model architecture and the training process by performing various experiments and observing the results. The final model has reached an accuracy comparable to other state-of-the-art models. Furthermore, we created a desktop application to demonstrate the results visually and to enable easier manipulation with the identity database. The knowledge gathered in this thesis can be used for improvements of current identification models or models modified for solving similar tasks.

No description available.

No description available.