3. Electronic Theses and Dissertations (ETDs) - All submissions
Permanent URI for this communityhttps://wiredspace.wits.ac.za/handle/10539/45
Browse
11 results
Search Results
Item Facial action unit classification using weakly supervised learning(2024) Enabor, Oseluole TobiDeep learning has gained popularity because of its supremacy in terms of performance when trained on large datasets. However, collecting and annotating large datasets is laborious, expensive, and time-consuming. Weak supervision learning (WSL) has been at the forefront in exploring solutions to the above limitations. WSL techniques can create accurate classifiers under different scenarios, such as limited sample datasets, inaccurate datasets with noisy labels, and datasets that do not have the desired labels. This work applies WSL to facial Action Unit (AU) recognition, a problem space that relies on subject-matter experts (i.e., certified Facial Action Unit Coders (FACS)) to annotate samples. Two WSL techniques, namely incomplete supervision using a pseudolabelling mechanism, where one has access to vast amounts of unlabelled data and a limited amount of labelled data, and inaccurate supervision using Large-Loss Rejection (LLR) mechanism, where one has access to only noisy labels, were explored. The pseudo-labelling mechanism involves feeding samples with generated pseudo-labels during the training process. Alternatively, the LLR mechanism prevents model learning noisy labels by rejecting samples that reported large-loss during training. To better evaluate the limitations posed by accurate data and label availability and its impact on training models, the authors trained a baseline emotion recognition model and finetuned for AU recognition using transfer learning. This process also helped access the ability to estimate fine-grain labels (AUs) using only coarse-grain labels (facial emotions). The experimental setup included training and validating a VGG16 Convolutional neural network (CNN) using the Extended Denver Intensity of Spontaneous Facial Action Database (DISFA+) and the use of the Karolinska Directed Emotional Faces (KDEF) dataset as cross-dataset evaluation. Pseudo-labelling approach for AU recognition had three models, the first, PL-1, reported subset accuracy of 68% and 0.56 weighted F1- score, PL-2a reported a subset accuracy 89% and 0.9 weighted F1-score, PL-2b reported a subset accuracy of 66% and a weighted F1-score of 0.44. The LLR approach for AU recognition reported a subset accuracy of 69% and a weighted average F1-score of 0.66. The baseline AU model reported accuracy of 97% and an F1-score of 0.98 for AU recognition, signifying the need for large data sets and transfer learning. However, with an average reported accuracy of 68.5%, WSL mechanisms provide a solution in the right direction and can assist researchers in addressing data annotation challengesItem Supervised learning algorithm’s role in flagging programming concepts that call for Information Technology teachers’ attention(2024) Mashite TshidiProgramming syntax and concepts' complexity demands learners possess logical thinking and problem-solving skills to effectively write and complete code. In the context of Information Technology education in South Africa, teachers require a tool that can identify learners' performance in programming concepts and help them prevent misunderstandings. This research proposes a transformational approach that uses a machine learning algorithm to alert teachers of programming concepts that learners may struggle with. The study also investigates how Information Technology teachers shape learning experiences when teaching programming concepts, using a qualitative methodology involving semi-structured interviews with Information Technology teachers. The study employs Educational Data Mining and Learning Analytics as theoretical and conceptual frameworks to showcase the potential of supervised learning algorithms in using prior Information Technology results for significant improvements in learning and performance. The findings indicate that problem-based learning is a commonly used methodology among Information Technology teachers. The algorithm results reveal a high-performance forecasting model based on acceptable accuracy, actual positive rate, and false positive rate. The identified programming concepts that require focus include conditional statements, conceptualizing problems and designing solutions, debugging and exception handling, abstraction/pattern recognition, and differentiating between classes and objects. Overall, this research presents a valuable approach for leveraging a supervised learning algorithm to enhance Information Technology education by identifying and addressing programming concepts that learners struggle with.Item Prediction of Blast Vibrations from Quarries using Machine Learning Algorithms and Empirical Formulae(2019) Morena, Badisheng IsaacThe aim of this study was to, firstly, use machine learning algorithms to predict Peak Particle Velocity (PPV) in order to optimise blasting layouts and reduce the risk of damaging surface structures. Empirical models developed by the United States Bureau of Mines (USBM) (1963) and Ambraseys and Hendron (1968) were compared to the machine learning algorithms. The tests conducted were interpolation and extrapolation. Most of the data used in this report was obtained from the USBM’s Bulletin 656. The data was analysed using a qualitative and quantitative research methods. The Cubist machine learning model (Kuhn, 2018) performed the best in the interpolation test with a coefficient of determination (R2) of 83.39 % and a root mean squared error (RMSE) and mean absolute error (MAE) of 10.64 and 7.30 respectively. The empirical models performed the best with the extrapolation test with an average R2 of 88 % and RMSE and MAE of 9.17 and 6.59 respectively. This research has shown the effectiveness of machine algorithms in predicting PPV and empirical formulae using historical data from different sites. However, explosive and geotechnical information was not available in the dataset and it is therefore recommended that further research be conducted with this data.Item Improving reinforcement learning with ensembles of different learners(2021) Crafford, GerrieDifferent reinforcement learning methods exist to address the problem of combining multiple dif ferent learners to generate a superior learner, from ensemble methods to policy reuse methods. These methods usually assume that each learner uses the same algorithm and/or state represen tation and often require learners to be pre-trained. This assumption prevents very different types of learners, that can potentially complement each other well, from being used together. We propose a novel algorithm, Adaptive Probabilistic Ensemble Learning (APEL), which is an ensemble learner that combines a set of base reinforcement learners and leverages the strengths of the different base learners online, while remaining agnostic to the inner workings of the base learners, thereby allowing it to combine very different types of learners. The ensemble learner selects the base learners that perform best on average by keeping track of the performance of the base learners and then probabilistically selecting a base learner for each episode according the historical performance of the base learners. Along with a description of the proposed algorithm, we present a theoretical analysis of its behaviour and performance. We demonstrate the proposed ensemble learner’s ability to select the best base learner on av erage, combine the strengths of multiple base learners, including Q-learning, deep Q-network (DQN), Actor-Critic with Experience Replay (ACER), and learners with different state repre sentations, as well as its ability to adapt to changes in base learner performance on grid world navigation tasks, the Cartpole domain, and the Atari Breakout domain. The effect that the en semble learner’s hyperparameter has on its behaviour and performance is also quantified through different experiments.Item A comparative analysis of dynamic averaging techniques in federated learning(2020) Reddy, SashlinDue to the advancements in mobile technology and user privacy concerns, federated learning has emerged as a popular machine learning (ML) method to push training of statistical models to the edge. Federated learning involves training a shared model under the coordination of a centralized server from a federation of participating clients. In practice federated learning methods have to overcome large network delays and bandwidth limits. To overcome the communication bottlenecks, recent works propose methods to reduce the communication frequency that have negligible impact on model accuracy also defined as model performance. Naive methods reduce the number of communication rounds in order to reduce the communication frequency. However, it is possible to invest communication more efficiently through dynamic communication protocols. This is deemed as dynamic averaging. Few have addressed such protocols. More so, few works base this dynamic averaging protocol on the diversity of the data and the loss. In this work, we introduce dynamic averaging frameworks based on the diversity of the data as well as the loss encountered by each client. This overcomes the assumption that each client participates equally and addresses the properties of federated learning. Results show that the overall communication overhead is reduced with negligible decrease in accuracyItem Multi-pass deep Q-networks for reinforcement learning with parameterised action spaces(2019) Bester, Craig JamesParameterised actions in reinforcement learning are composed of discrete actions with continuous actionparameters. This provides a framework capable of solving complex domains that require learning highlevel action policies with flexible control. Recently, deep Q-networks have been extended to learn over such action spaces with the P-DQN algorithm. However, the method treats all action-parameters as a single joint input to the Q-network, invalidating its theoretical foundations. We demonstrate the disadvantages of this approach and propose two solutions: using split Q-networks, and a novel multi-pass technique. We also propose a weighted-indexed action-parameter loss function to address issues related to the imbalance of sampling and exploration between different parameterised actions. We empirically demonstrate that both our multi-pass algorithm and weighted-indexed loss significantly outperform P-DQN and other previous algorithms in terms of data efficiency and converged policy performance on the Platform, Robot Soccer Goal, and Half Field Offense domains.Item Using ensemble learning for the network intrusion detection problem(2019-08-01) Kalonji, Roland MpoyiNowadays, most organizations and platforms employ an intrusion detection system (IDS) to enhance their network security and protocol systems. The IDS has therefore become an essential component of any network system; it is a tool with several applications that can be tuned to specific content in a network by identifying various accesses (normal or attack). However, network intrusion detection system (NIDS) that focuses on revealing suspicious activities, is not effective in solving various problems such as identifying false IP packets and encrypted traffic. Hence, this work investigates the use of ensemble learning to solve these types of network intrusion detection problems (NIDPs). Random forest (RF), Decision Tree (DT) and Support Vector Machine (SVM) are introduced as classifiers based on Boruta and Principal Component Analysis (PCA) algorithms. In general, the main difficulties in using ensemble for the intrusion problem are to minimize false alarms and to maximize detection accuracy (Anuar et al., 2008). Additionally, the NIDP is divided into five categories, namely the detection of probe attacks, denial of service, remote to local, user to root and normal instances. Each problem is examined by one of the three aforementioned classifiers. In tackling these problems, the three classifiers achieved competitive results comparing to the works conducted by Balon-Perin (2012), Zainal et al. (2009) and Kevric et al. (2017). The results revealed that ensemble learning achieved more than 99% accuracy in demarcating attacks from normal connections. Particularly, RF, DT and SVM allowed to safeguard the NIDS from known and unknown attacks by developing reliable techniques. The KDD99 and NSL KDD datasets have been used to implement and measure the system performance (Fan et al., 2000; Dhanabal and Shantharajah, 2015).Item Skill discovery from multiple related demonstrators(2018) Ranchod, PraveshAn important ability humans have is that we can recognise that some collec tions of actions are useful in multiple tasks, allowing us to exploit these skills. A human who can run while playing basketball does not need to relearn this ability when he is playing soccer as he can employ his previously learned run ning skill. WeextendthisideatothetaskofLearningfromDemonstration(LfD),wherein an agent must learn a task by observing the actions of a demonstrator. Tradi tional LfD algorithms learn a single task from a set of demonstrations, which limits the ability to reuse the learned behaviours. We instead recover all the latentskillsemployedinasetofdemonstrations. Thedifficultyinvolvedliesin determiningwhichcollectionsofactionsinthedemonstrationscanbegrouped together and termed “skills”? We use a number of characteristics observed in studies of skill discovery in children to guide this segmentation process – use fulness (they lead to some reward), chaining (we tend to employ certain skills in common combinations), and reusability (the same skill will be employed in many different contexts). Weusereinforcementlearningtomodelgoaldirectedbehaviour,hiddenMarkov models to model the links between skills, and nonparametric Bayesian cluster ing to model reusability in a potentially infinite set of skills. We introduce nonparametric Bayesian reward segmentation (NPBRS), an algorithm that is abletosegmentdemonstrationtrajectoriesintocomponentskills,usinginverse reinforcement learning to recover reward functions representing the skill ob i jectives. We then extend the algorithm to operate in domains with continuous state spaces for which the transition model is not specified, with the algorithm suc cessfully recovering component skills in a number of simulated domains. Fi nally, we perform an experiment on CHAMP, a physical robot tasked with mak ingvariousdrinks,anddemonstratethatthealgorithmisabletorecoveruseful skills in a robot domain.Item Using neural networks and support vector machines for default prediction in South Africa(2017) Meltzer, FrancesThis is a thesis on credit risk and in particular bankruptcy prediction. It investigates the application of machine learning techniques such as support vector machines and neural networks for this purpose. This is not a thesis on support vector machines and neural networks, it simply looks at using these functions as tools to preform the analysis. Neural networks are a type of machine learning algorithm. They are nonlinear mod- els inspired from biological network of neurons found in the human central nervous system. They involve a cascade of simple nonlinear computations that when aggre- gated can implement robust and complex nonlinear functions. Neural networks can approximate most nonlinear functions, making them a quite powerful class of models. Support vector machines (SVM) are the most recent development from the machine learning community. In machine learning, support vector machines (SVMs) are su- pervised learning algorithms that analyze data and recognize patterns, used for clas- si cation and regression analysis. SVM takes a set of input data and predicts, for each given input, which of two possible classes comprises the input, making the SVM a non-probabilistic binary linear classi er. A support vector machine constructs a hyperplane or set of hyperplanes in a high or in nite dimensional space, which can be used for classi cation into the two di erent data classes. Traditional bankruptcy prediction medelling has been criticised as it makes certain underlying assumptions on the underlying data. For instance, a frequent requirement for multivarate analysis is a joint normal distribution and independence of variables. Support vector machines (and neural networks) are a useful tool for default analysis because they make far fewer assumptions on the underlying data. In this framework support vector machines are used as a classi er to discriminate defaulting and non defaulting companies in a South African context. The input data required is a set of nancial ratios constructed from the company's historic nancial statements. The data is then Divided into the two groups: a company that has defaulted and a company that is healthy (non default). The nal data sample used for this thesis consists of 23 nancial ratios from 67 companies listed on the jse. Furthermore for each company the company's probability of default is predicted. The results are benchmarked against more classical methods that are commonly used for bankruptcy prediction such as linear discriminate analysis and logistic regression. Then the results of the support vector machines, neural networks, linear discriminate analysis and logistic regression are assessed via their receiver operator curves and pro tability ratios to gure out which model is more successful at predicting default.Item Automatic speech feature extraction using a convolutional restricted boltzmann machine(2017) Anderson, David JohnRestricted Boltzmann Machines (RBMs) are a statistical learning concept that can be interpreted as Arti cial Neural Networks. They are capable of learning, in an unsupervised fashion, a set of features with which to describe a data set. Connected in series RBMs form a model called a Deep Belief Network (DBN), learning abstract feature combinations from lower layers. Convolutional RBMs (CRBMs) are a variation on the RBM architecture in which the learned features are kernels that are convolved across spatial portions of the input data to generate feature maps identifying if a feature is detected in a portion of the input data. Features extracted from speech audio data by a trained CRBM have recently been shown to compete with the state of the art for a number of speaker identi cation tasks. This project implements a similar CRBM architecture in order to verify previous work, as well as gain insight into Digital Signal Processing (DSP), Generative Graphical Models, unsupervised pre-training of Arti cial Neural Networks, and Machine Learning classi cation tasks. The CRBM architecture is trained on the TIMIT speech corpus and the learned features veri ed by using them to train a linear classi er on tasks such as speaker genetic sex classi cation and speaker identi cation. The implementation is quantitatively proven to successfully learn and extract a useful feature representation for the given classi cation tasks