Audrey Mbogho - Profile on Academia.edu (original) (raw)

Papers by Audrey Mbogho

Learning activities with tangible user interfaces provide the benefits of active and peer mediate... more Learning activities with tangible user interfaces provide the benefits of active and peer mediated learning, while offering assistance from an autonomous guide on the side. Yet tangible user interfaces must typically be custom developed by computer scientists, and only rarely is the assistance of teachers sought. We present a new tool that gives teachers the power to create their own educational applications with tangible user interfaces. Using actual scientific specimens, the teachers can define object attributes for the students to sort on, and also develop curriculum-appropriate hints. We describe our computer vision-based approach, which enables recognition of tags representing dichotomous keys defined by a teacher. Teachers use our back-end system to define the dichotomous keys and other parameters for the learning activities, while our front-end system uses those parameters to guide the students.

Tackling Classroom Apathy Among Undergraduate Students in a Developing World Context at Pwani University

INTED2017 Proceedings, 2017

Probabilistic Expert Systems for Reasoning in Clinical Depressive Disorders

2016 15th IEEE International Conference on Machine Learning and Applications (ICMLA), 2016

Undereducation, Motivating Intervention in Rural Schools with MAPPS

Proceedings of the First African Conference on Human Computer Interaction - AfriCHI'16, 2016

Selecting Learning Algorithms for Simultaneous Identification of Depression and Comorbid Disorders

Procedia Computer Science, 2016

Predictive Strength of Bayesian Networks for Diagnosis of Depressive Disorders

Smart Innovation, Systems and Technologies, 2016

Due to the portability, mobility as well as the significantly enhanced processing and computing c... more Due to the portability, mobility as well as the significantly enhanced processing and computing capabilities of mobile phones, applications based on the use of mobile phones are increasingly being used in the learning environment. However, there have not been comprehensive studies of how accessible these phone-based educational applications are, or how easy it is for novice users to learn to use them. This paper presents a study that focuses on the evaluation of 2D visual tag (or barcode) systems for accessing digital library content. The study is designed to determine the ease with which novice users can learn to use a tag-based system, and the appropriateness of use of such systems in Africa. The results show that visual tag applications are appropriate for use in Africa, and are also easy to learn to operate. The study incorporates guides on how to further enhance the usability of visual tag-based systems. It is expected that these findings are applicable to developing countries in general.

Proceedings of the Seventh International Conference on Information and Communication Technologies and Development - ICTD '15, 2015

Poor academic performance in final exams at primary school level in Kenya is a strong indicator t... more Poor academic performance in final exams at primary school level in Kenya is a strong indicator that the student will not attain the desired career in future. It is therefore important to be able to predict the students who are likely to achieve below average marks and need high intervention early enough for them to improve their marks. This paper reports on a study to classify primary school students into two categories, those that need high intervention and the rest. The prediction can be initiated as early as two years before the final exam. An important highlight of this study is its focus on rural schools in a developing country. A total of 2426 records of students are used to build intervention prediction models. In the first set of experiments all the features are used. An optimal subset of features is then determined and a second set of experiments carried out. Results demonstrate that it is possible to attain reasonably accurate intervention prediction models even with the reduced dataset. The insights obtained will be used to build a mobile prediction tool that can be utilized by education stakeholders in rural regions where there is lack of electricity.

Improving the Transcription of Academic Lectures for Information Retrieval

2013 12th International Conference on Machine Learning and Applications, 2013

ABSTRACT Recording university lectures through lecture capture systems is increasingly common, ge... more ABSTRACT Recording university lectures through lecture capture systems is increasingly common, generating large amounts of audio and video data. Transcribing recording s greatly enhances their usefulness by making them easy to search. However, the number of recordings accumulates rapidly, rendering manual transcription impractical. Automatic transcription, on the other hand, suffers from low levels of accuracy, partly due to the special language of academic disciplines, which standard language models do not cover. This paper looks into the use of Wikipedia to dynamically adapt language models for scholarly speech. We propose Ranked Word Correct Rate as a new metric better aligned with the goals of improving transcript searchability and specialist word recognition. The study shows that, while overall transcription accuracy may remain low, targeted language modeling can substantially improve searchability, an important goal in its own right.

Proceedings of the 9th annual conference on Genetic and evolutionary computation - GECCO '07, 2007

This paper reports on a case study on segmentation of colored visual tags for object identificati... more This paper reports on a case study on segmentation of colored visual tags for object identification. Lighting variations result in uncertainty in color thresholds leading to unreliable overall system behavior. We describe an experiment with a genetic algorithm (GA) approach for generating reliable thresholds for color identification. We compare it with a maximum distance (MD) approach, and demonstrate that the genetic approach is far more accurate and reliable.

Selecting Relevant Features for Classifier Optimization

Communications in Computer and Information Science, 2014

ABSTRACT Feature selection is an important data pre-processing step that comes before applying a ... more ABSTRACT Feature selection is an important data pre-processing step that comes before applying a machine learning algorithm. It removes irrelevant and redundant attributes from the dataset with an aim of improving the algorithm performance. There exist feature selection methods which focus on discovering features that are most suitable. These methods include wrappers, a subroutine of the learning algorithm itself, and filters, which discover features according to heuristics, based on the data characteristics and not tied to a specific algorithm. This paper improves the filter approach by enabling it to select strongly relevant and weakly relevant features and gives room to the researcher to decide which of the weakly relevant features to include. This new approach brings clarity and understandability to the feature selection preprocessing step.

Web-Based Corpus Acquisition for Swahili Language Modelling

Reliability and testing in vision-based interaction /

"A dissertation submitted to the Graduate Faculty in Philosphy ... " Thesis (Ph. D.) --... more "A dissertation submitted to the Graduate Faculty in Philosphy ... " Thesis (Ph. D.) -- City University of New York, 2006. Includes bibliographical references (leaves 102-108).

A major obstacle in the development of computer vision-based interfaces is the uncertainty in the... more A major obstacle in the development of computer vision-based interfaces is the uncertainty in the image data. Guaranteeing reliable program behavior while the inputs cannot be relied upon to take any specific values is desirable but extremely challenging. We have experimented with various strategies for addressing this problem in a controlled environment and have identified some that we find promising. Using visual tags, we encode object attributes, capture them with a camera, and read them. The test applications are implemented as Macromedia Director movies with Lingo scripting, while the image analysis module is implemented in C++ as a Lingo Xtra. We conclude that evaluation through rigorous experimentation and testing is a suitable approach for discovering techniques that work well and the conditions that maximize the chances for optimal performance.

Computer Communications, 2007

Potential use of wireless sensor networks (WSNs) can be seen in various fields like disaster mana... more Potential use of wireless sensor networks (WSNs) can be seen in various fields like disaster management, battle field surveillance and border security surveillance since last few years. In such applications, a large number of sensor nodes are deployed, which are often unattended and work autonomously. Clustering is a key technique used to extend the lifetime of a sensor network by reducing energy consumption. It can also increase network scalability. Sensor nodes are considered to be homogeneous since the researches in the field of WSNs have been evolved, but some nodes may be of different energy to prolong the lifetime of a WSN and its reliability. In this paper, we study the impact of heterogeneity of nodes to the performance of WSNs. This paper surveys different clustering algorithms for heterogeneous WSNs by classifying algorithms depending upon various clustering attributes.

acm.org

Diabetes Advisor – A Medical Expert System for Diabetes Management

The accent with which words are spoken can have a strong effect on the performance of a speech re... more The accent with which words are spoken can have a strong effect on the performance of a speech recognition system. In a multilingual country such as South Africa where English is not the first language of most citizens, the need to address this issue is critical when building speech-based systems. In this project we trained two sets of hidden Markov Models for isolated word English speech. The first set of models was trained with native English speakers and the second set was trained with non-native speakers from a representative sample of major South African accent groups. We compared the recognition accuracies of the two sets of models and found that the models trained with accented English performed better. This preliminary research indicates that there is merit to committing resources to the task of accented training.

Cell phones now pack as much computing power as the high-end desktop of a decade ago. This makes ... more Cell phones now pack as much computing power as the high-end desktop of a decade ago. This makes nontrivial mobile applications possible. Interacting with them is, however, a significant challenge due to the small size and limitations of the keypad. Because many new phone models are equipped with an integrated camera, there is growing interest in the use of visual tags to more easily provide inputs to mobile applications. A number of tag designs have been proposed, but more needs to be done in this yet nascent technology. We present CipherCode, a visual tagging system designed for speed, accuracy and compactness. CipherCode includes encryption and parameterisation, features not found in most other visual tagging schemes. We have made CipherCode freely available on the Web as an open-source SDK in order to ignite broader interest and participation in creating highly usable visual tags.