Modern automobiles have been proven vulnerable to hacking by security researchers. By exploiting vulnerabilities in the car's external interfaces, attackers can access a car's controller area network (CAN) bus and cause malicious effects. We seek to detect these attacks on the bus as a last line of defence against automotive cyber attacks. The CAN bus standard defines a low-level message structure, upon which manufacturers layer their own proprietary command protocols; attacks must similarly be tailored for their target. This variability makes intrusion detection methods difficult to apply to the automotive CAN bus. Nevertheless, the bus traffic is generated by machines; thus we hypothesize that it can be characterized with machine learning, and that attacks produce anomalous traffic. Our goals are to show that anomaly detection trained without understanding of the message contents can detect attacks, and to create a framework for understanding how the characteristics of a novel attack can be used to predict its detectability. We developed a model that describes attacks based on their effect on bus traffic, informed by a review of published material on car hacking in combination with analysis of CAN traffic from a 2012 Subaru Impreza. The model specifies three high-level categories of effects: attacks that insert foreign packets, attacks that affect packet timing, and attacks that only modify data within packets. Foreign packet attacks are trivially detectable. For timing-based anomalies, we developed features suitable for one-class classification methods. For packet stream data word anomalies, we adapted recurrent neural networks and multivariate Markov model methods to sequence anomaly detection and compared their performance. We conducted experiments to evaluate our detection methods with special attention to the trade-off between precision and recall, given that a practical system requires a very low false alarm rate. The methods were evaluated by synthesizing anomalies within each attack category, parameterized to adjust their covertness. We generalize from the results to enable prediction of detection rates for new attacks using these methods.

This diploma thesis is about design and realization of neural network MIA GMDH for ekonomical modelling by inductive method. Models are compared with statistical methods by quallity and usebility degree. An application was developed for verification of functionality on experiments. The same experiments were run in econometrical software. The results were compared. The MIA GMDH is suitable for economic modelling.

Falls are a leading cause of injury and accidental injury death worldwide. Fall-risk prevention techniques exist but fall-risk identification can be difficult. While clinical assessment tools are the standard for identifying fall risk, wearable-sensors and machine learning could improve outcomes with automated and efficient techniques. Machine learning research has focused on older adults. Since people with lower limb amputations have greater falling and injury risk than the elderly, research is needed to evaluate these approaches with the amputee population. In this thesis, random forest and fully connected feedforward artificial neural network (ANN) machine learning models were developed and optimized for fall-risk identification in amputee populations, using smartphone sensor data (phone at posterior pelvis) from 89 people with various levels of lower-limb amputation who completed a 6-minute walk test (6MWT). The best model was a random forest with 500 trees, using turn data and a feature set selected using correlation-based feature selection (81.3% accuracy, 57.2% sensitivity, 94.9% specificity, 0.59 Matthews correlation coefficient, 0.83 F1 score). After extensive ANN optimization with the best ranked 50 features from an Extra Trees Classifier, the best ANN model achieved 69.7% accuracy, 53.1% sensitivity, 78.9% specificity, 0.33 Matthews correlation coefficient, and 0.62 F1 score. Features from a single smartphone during a 6MWT can be used with random forest machine learning for fall-risk classification in lower limb amputees. Model performance was similarly effective or better than the Timed Up and Go and Four Square Step Test. This model could be used clinically to identify fall-risk individuals during a 6MWT, thereby finding people who were not intended for fall screening. Since model specificity was very high, the risk of accidentally misclassifying people who are a no fall-risk individual is quite low, and few people would incorrectly be entered into fall mitigation programs based on the test outcomes.

My dissertation, titled “From the machine mind to the human mind: using machine learning to understand (ir)rationality, bias and polarization in human beings,” investigates ways in which human minds operate and seeks to uncover the causes of biasedness, limited rationality, and polarization of human minds, to eventually devise tools to compensate for such human limitations. Chapter 2 of the thesis focuses on the evaluation of information and decision making under enormous information asymmetry, in the setting of patients evaluating doctors’ medical advice. Patients were found to be poor evaluators who were unable to distinguish good from bad due to their lack of medical expertise, and unable to overcome their own irrationality and bias. I emphasize the ramification of such limited rationality, which might lead to the adoption of suboptimal or bad medical opinions, and propose ways to improve this situation by redesigning some features of the platform, and/or implementing new policies to help good doctors on the platform. Chapter 3 focuses on developing a new metric that reliably measures the ideology of the US elites. This metric was developed based on congressional reports which made it unique and relatively independent from established metrics based on roll call votes, such as DW-NOMINATE. First, I leveraged a neural network-based approach to decompose the speech documents into frames and topics components, with all ideological information funneled into the frames component. Eventually, two different ideology metrics were obtained and validated: an embedding vector and an ideological slant score. Later I showed that our new metrics can predict party switchers and trespassers with high recall. In chapter 4, I applied the newly obtained metric (mainly slant scores) to investigate various aspects of the congress, such as the heterogeneity of ideology among the members, the temporal evolution of partisan division, the bill passing, and the re-election strategy of the senators.

<p>Meningioma represent more than one-third of all primary central nervous system (CNS) tumors,and it can be classified into three grades according to WHO (World Health Organization) in terms of clinical aggressiveness and risk of recurrence. A key component of meningioma grades is the mitotic count, which is defined as quantifying the number of cells in the process of dividing (i.e., undergoing mitosis) at a specific point in time. Currently, mitosis counting is done manually by a pathologist looking at 10 consecutive high-power fields (HPF) on a glass slide under a microscope, which is an extremely laborious and time-consuming process. The goal of this thesis is to investigate the use of computerized methods to automate the detection of mitotic nuclei with limited labeled data.We built computational methods to detect and quantify the histological features of mitotic cells on a whole slides image which mimic the exact process of pathologist workflow. Since we do not have enough training data from meningioma slide, we learned the mitotic cell features through public available breast cancer datasets, and predicted on meingioma slide for accuracy. We use either handcrafted features that capture certain morphological, statistical, or textural attributes of mitoses or features learned with convolutional neural networks (CNN). Hand crafted features are inspired by the domain knowledge, while the data-driven VGG16models tend to be domain agnostic and attempt to learn additional feature bases that cannot be represented through any of the handcrafted features. Our work on detection of mitotic cells shows 100% recall, 9% precision and 0.17 F1 score. The detection using VGG16performs with 71% recall, 73% precision, and 0.77 F1 score.Finally, this research of automated image analysis could drastically increase diagnostic efficiency and reduce inter-observer variability and errors in pathology diagnosis, which would allow fewer pathologists to serve more patients while maintaining diagnostic accuracy and precision. And all these methodologies will increasingly transform practice of pathology, allowing it to mature toward a quantitative science.</p>

Multiphase flow provides a solution to the high heat flux and precision required by modern-day gadgets and heat transfer devices as phase change processes make high heat transfer rates achievable at moderate temperature differences. An application of multiphase flow commonly used in industry is the condensation of refrigerants in inclined tubes. The identification of two-phase flow patterns, or flow regimes, is fundamental to the successful design and subsequent optimisation given that the heat transfer efficiency and pressure gradient are dependent on the flow structure of the working fluid. This study showed that with visualisation data and artificial neural networks (ANN), a machine could learn, and subsequently classify the separate flow patterns of condensation of R-134a refrigerant in inclined smooth tubes with more than 98% accuracy. The study considered 10 classes of flow pattern images acquired from previous experimental works that cover a wide range of flow conditions and the full range of tube inclination angles. Two types of classifiers were considered, namely multilayer perceptron (MLP) and convolutional neural networks (CNN). Although not the focus of this study, the use of a principal component analysis (PCA) allowed feature dimensionality reduction, dataset visualisation, and decreased associated computational cost when used together with multilayer perceptron neural networks. The superior two-dimensional spatial learning capability of convolutional neural networks allowed improved image classification and generalisation performance across all 10 flow pattern classes. In both cases, the classification was done sufficiently fast to enable real-time implementation in two-phase flow systems. The analysis sequence led to the development of a predictive tool for the classification of multiphase flow patterns in inclined tubes, with the goal that the features learnt through visualisation would apply to a broad range of flow conditions, fluids, tube geometries and orientations, and would even generalise well to identify adiabatic and boiling two-phase flow patterns. The method was validated by the prediction of flow pattern images found in the existing literature. / Dissertation (MEng)--University of Pretoria, 2021. / NRF / Mechanical and Aeronautical Engineering / MEng / Restricted

Robots are no longer constrained to cages in factories and are increasingly taking on roles alongside humans. Before robots can accomplish their tasks in these dynamic environments, they must be able to navigate while avoiding collisions with pedestrians or other robots. Humans are able to move through crowds by anticipating the movements of other pedestrians and how their actions will influence others; developing a method for predicting pedestrian trajectories is a critical component of a robust robot navigation system. A current state-of-the-art approach for predicting pedestrian trajectories is Social-LSTM, which is a recurrent neural network that incorporates information about neighboring pedestrians to learn how people move cooperatively around each other. This thesis extends and modifies that model to output parameters for a multimodal distribution, which better captures the uncertainty inherent in pedestrian movements. Additionally, four novel architectures for representing neighboring pedestrians are proposed; these models are more general than current trajectory prediction systems and have fewer hyper-parameters. In both simulations and real-world datasets, the multimodal extension significantly increases the accuracy of trajectory prediction. One of the new neighbor representation architectures achieves state-of-the-art results while reducing the number of both parameters and hyper-parameters compared to existing solutions. Two techniques for incorporating the trajectory predictions into a planning system are also developed and evaluated on a real-world dataset. Both techniques plan routes that include fewer near-collisions than algorithms that do not use trajectory predictions. Finally, a Python library for Agent-Based-Modeling and crowd simulation is presented to aid in future research.

The aim of this thesis is to study problems of deep convolutional neural networks and the connected classification of images and to experiment with the architecture of particular network with the aim to get the most accurate results on the selected dataset. The thesis is divided into two parts, the first part theoretically outlines the properties and structure of neural networks and briefly introduces selected networks. The second part deals with experiments with this network, such as the impact of data augmentation, batch size and the impact of dropout layers on the accuracy of the network. Subsequently, all results are compared and discussed with the best result achieved an accuracy of 86, 44% on test data.

This thesis deals with designing an automatic generator of deep neural networks for image classification. Theoretical part clarifies what a neural network and formal neuron are. Furthermore, the types of neural network architectures are presented. The focus of this thesis is convolutional neural networks, several pieces of research from this field are mentioned. The practical part of this thesis describes information with regards to the implementation of neural network generator, possible frameworks and programming languages for such implementation. Brief description of the implementation itself is presented as well as implemented layers. Generated neural networks are tested on Google-Landmarks dataset and results are commented upon.

This paper deals with processing and recognition of events in audio signal. The work explores the possibility of using audio signal visualization and subsequent use of convolutional neural networks as a classifier for recognition in real use. Recognized audio events are gunshots placed in a sound background such as street noise, human voice, animal sounds, and other forms of random noise. Before the implementation, a large database with various parameters, especially reverberation and time positioning within the processed section, is created. In this work are used freely available platforms Keras and TensorFlow for work with neural networks.