Lyudmila Mihaylova | The University of Sheffield (original) (raw)
Papers by Lyudmila Mihaylova
Faraday discussions, 2021
2018 Sensor Data Fusion: Trends, Solutions, Applications (SDF), 2018
This paper proposes a deep learning approach for traffic flow prediction in complex road networks... more This paper proposes a deep learning approach for traffic flow prediction in complex road networks. Traffic flow data from induction loop sensors are essentially a time series, which is also spatially related to traffic in different road segments. The spatio-temporal traffic data can be converted into an image where the traffic data are expressed in a 3D space with respect to space and time axes. Although convolutional neural networks (CNNs) have been showing surprising performance in understanding images, they have a major drawback. In the max pooling operation, CNNs are losing important information by locally taking the highest activation values. The interrelationship in traffic data measured by sparsely located sensors in different time intervals should not be neglected in order to obtain accurate predictions. Thus, we propose a neural network with capsules that replaces max pooling by dynamic routing. This is the first approach that employs the capsule network on a time series forecasting problem, to our best knowledge. Moreover, an experiment on real traffic speed data measured in the Santander city of Spain demonstrates the proposed method outperforms the state-of-the-art method based on a CNN by 13.1% in terms of root mean squared error.
2019 10th International Conference on Dependable Systems, Services and Technologies (DESSERT), 2019
This is a repository copy of Non-random weight initialisation in deep learning networks for repea... more This is a repository copy of Non-random weight initialisation in deep learning networks for repeatable determinism.
Neurocomputing, 2021
This paper proposes a novel framework of flame region-based convolutional neural network for auto... more This paper proposes a novel framework of flame region-based convolutional neural network for autonomous flame detection. The task of flame detection is especially challenging since flames have greater diversity in colour, texture, and shape than regular rigid objects. To cope with these difficulties due to the various appearances and unclear edges of flames, a proposal generation approach is developed to effectively select candidate flame regions based on two crucial properties of flames, i.e., their dynamics and colours. The candidate flame regions together with a convolutional feature map are further processed by additional layers to output detected flames. The diversity in flame colours is well represented by approximating the distribution using a Dirichlet Process Gaussian mixture model with variational inference. The proposed framework is evaluated on publicly available videos and achieves an average frame-wise accuracy higher than 88%, which outperforms the state-of-the-art methods.
2017 20th International Conference on Information Fusion (Fusion), Jul 1, 2017
Vehicle logo recognition is an important part of vehicle identification in intelligent transporta... more Vehicle logo recognition is an important part of vehicle identification in intelligent transportation systems. Stateof-the-art vehicle logo recognition approaches typically consider training models on large datasets. However, there might only be a small training dataset to start with and more images can be obtained during the real-time applications. This paper proposes an online image recognition framework which provides solutions for both small and large datasets. Using this recognition framework, models are built efficiently using a weight updating scheme. Another novelty of this work is that the Cauchy prior logistic regression with conjugate gradient descent is proposed to deal with the multinomial classification tasks. The Cauchy prior results in a quicker convergence speed for the weight updating process which could decrease the computational cost for both online and offline methods. By testing with a publicly available dataset, the Cauchy prior logistic regression deceases the classification time by 59%. An accuracy of up to 98.80% is achieved when the proposed framework is applied.
IEEE Transactions on Aerospace and Electronic Systems, 2019
A novel robust Rauch-Tung-Striebel smoothing framework is proposed based on a generalized Gaussia... more A novel robust Rauch-Tung-Striebel smoothing framework is proposed based on a generalized Gaussian scale mixture (GGScM) distribution for a linear state-space model with heavy-tailed and/or skew noises. The state trajectory, mixing parameters and unknown distribution parameters are jointly inferred using the variational Bayesian approach. As such, a major contribution of this work is unifying results within the GGScM distribution framework. Simulation and experimental results demonstrate that the proposed smoother has better accuracy than existing smoothers.
2019 22th International Conference on Information Fusion (FUSION)
This is a repository copy of Multi-band image fusion using Gaussian process regression with spars... more This is a repository copy of Multi-band image fusion using Gaussian process regression with sparse rational quadratic kernel.
Sensors
Railway networks systems are by design open and accessible to people, but this presents challenge... more Railway networks systems are by design open and accessible to people, but this presents challenges in the prevention of events such as terrorism, trespass, and suicide fatalities. With the rapid advancement of machine learning, numerous computer vision methods have been developed in closed-circuit television (CCTV) surveillance systems for the purposes of managing public spaces. These methods are built based on multiple types of sensors and are designed to automatically detect static objects and unexpected events, monitor people, and prevent potential dangers. This survey focuses on recently developed CCTV surveillance methods for rail networks, discusses the challenges they face, their advantages and disadvantages and a vision for future railway surveillance systems. State-of-the-art methods for object detection and behaviour recognition applied to rail network surveillance systems are introduced, and the ethics of handling personal data and the use of automated systems are also co...
2019 IEEE International Symposium on Signal Processing and Information Technology (ISSPIT), 2019
This paper describes a machine assistance approach to grading decisions for values that might be ... more This paper describes a machine assistance approach to grading decisions for values that might be missing or need validation, using a mathematical algebraic form of an Expert System, instead of the traditional textual or logic forms and builds a neural network computational graph structure. This Experts System approach is also structured into a neural network like format of: input, hidden and output layers that provide a structured approach to the knowledge-base organization, this provides a useful abstraction for reuse for data migration applications in big data, Cyber and relational databases. The approach is further enhanced with a Bayesian probability tree approach to grade the confidences of value probabilities, instead of the traditional grading of the rule probabilities, and estimates the most probable value in light of all evidence presented. This is ground work for a Machine Learning (ML) experts system approach in a form that is closer to a Neural Network node structure.
2019 Sensor Data Fusion: Trends, Solutions, Applications (SDF), 2019
This is a repository copy of Leveraging uncertainty in adversarial learning to improve deep learn... more This is a repository copy of Leveraging uncertainty in adversarial learning to improve deep learning based segmentation.
IEEE Transactions on Aerospace and Electronic Systems, 2021
This paper proposes a recursive Gaussian process regression with a joint optimization-based itera... more This paper proposes a recursive Gaussian process regression with a joint optimization-based iterative learning control algorithm to estimate and predict disturbances and model uncertainties affecting a flight. The algorithm proactively compensates for the predicted disturbances, improving precision in aircraft trajectory tracking. Higher precision in trajectory tracking implies an improvement of the aircraft trajectory predictability and therefore of the air traffic management system efficiency. Airlines can also benefit from this higher predictability by reducing the number of alterations when following their designed trajectories, which entails a reduction of costs and emissions. The iterative learning control algorithm is divided into two steps: first, a recursive Gaussian process regression estimates and predicts perturbations and model errors with no need for prior knowledge about their dynamics and with low computational cost, and second, this information is used to update the control inputs so that the subsequent aircraft intending to fly the same planned trajectory will follow it with greater precision than the previous ones. This method is tested on a simulated commercial aircraft performing a continuous climb operation and compared to an iterative learning algorithm using a Kalman filter estimator in a similar scenario. The results show that the proposed approach provides 62% and 42% precision improvement in tracking the desired trajectory, as compared to the Kalman filter approach, in two experiments where no prior knowledge of the unmodeled dynamics was available, also achieving it in less iterations.
Signal Processing, 2021
Although kernel approximation methods have been widely applied to mitigate the O(n 3) cost of the... more Although kernel approximation methods have been widely applied to mitigate the O(n 3) cost of the n × n kernel matrix inverse in Gaussian process methods, they still face computational challenges. The 'residual' matrix between the covariance and the approximating component is often discarded as it prevents the computational cost reduction. In this paper, we propose a computationally efficient Gaussian process approach that achieves better computational efficiency, O(mn 2), compared with standard Gaussian process methods, when using m ≪ n data. The proposed approach incorporates the 'residual' matrix in its symmetric diagonally dominant form which can be further approximated by the Neumann series. We have validated and compared the approach with full Gaussian process approaches and kernel approximation based Gaussian process
Engineering Applications of Artificial Intelligence, 2021
Vehicle localisation is an important and challenging task in achieving autonomous driving. This w... more Vehicle localisation is an important and challenging task in achieving autonomous driving. This work presents a box particle filter framework for vehicle selflocalisation in the presence of sensor and map uncertainties. The proposed feature-refined box particle filter incorporates line features extracted from a multi-layer Light Detection And Ranging (LiDAR) sensor and information from OpenStreetMap to estimate the vehicle state. A particle weight balance strategy is incorporated to account for the OpenStreetMap inaccuracy, which is assessed by comparing it to a high definition road map. The performance of the proposed framework is evaluated on a LiDAR dataset and compared with box particle filter variants. Experimental results show that the proposed framework achieves respectively 10% and 53% localisation accuracy improvement with reduced box volumes of 25% and 41%, when compared with the state-of-the-art interval analysis based box regularisation particle filter and the box particle filter.
Ad Hoc Networks, 2020
The growth of connected vehicles in smart cities increases the number of information being commun... more The growth of connected vehicles in smart cities increases the number of information being communicated on the Internet of Vehicle networks. This causes wireless channel congestion problems, which degrades the network performance and reliability due to the low throughput, high average delay and the high packets loss. Therefore, this paper proposes a non-cooperative game approach to control congestion in the vehicular ad-hoc network channel where the nodes behave as selfish players requesting high data transmission rates. Moreover, the satisfaction of the Nash equilibrium condition for the optimum data transmission rate for each vehicle, is proven. A utility function is introduced based on data transmission rates, the priority of vehicles and contention delay in order to obtain the optimal rates. The performance of the proposed approach has been evaluated and validated in comparison with three others approaches over two testing scenarios for highway and urban traffic. The results show that the network performance and efficiency have been improved by an overall average of 35%, 30% and 37.17% in terms of packets loss, channel busy time and number of collision messages, respectively, as compared with the state-of-the-art-strategies for the highway testing scenario. Similar performance is achieved for the urban testing scenario.
Future Generation Computer Systems, 2019
This is a repository copy of Information and resource management systems for Internet of Things: ... more This is a repository copy of Information and resource management systems for Internet of Things: Energy management, communication protocols and future applications.
Signal Processing, 2021
The performance of the state estimation for Gaussian state space models can be degraded if the mo... more The performance of the state estimation for Gaussian state space models can be degraded if the models are affected by the non-Gaussian process and measurement noises with uncertain degree of non-Gaussianity. In this paper, we propose a flexible robust Student's t-based multimodel approach. More specifically, the degrees of freedom parameter from the Student's t-distribution is assumed unknown and modelled by a Markov chain of state values. In order to capture more information of the Student's tdistributions propagated through multiple models, we establish a model-based Versoria cost function in the form of a weighted mixture rather than the original form, and maximize the function to interact and fuse the multiple models. Simulated results prove the flexibility of the robustness of the proposed Student's t-basedmultimodel approach when the existence probability of the outliers is uncertain.
This paper presents a novel non-parametric backpropagation Bayesian compressive sensing (BBCS) cl... more This paper presents a novel non-parametric backpropagation Bayesian compressive sensing (BBCS) classification approach. While the state-of-the-art parametric classifiers such as logistic regression require model training and can result in inadequate models, the developed approach does not require model training. It is combined with a column-based subspace sampling process and it can deal efficiently with uncertainties and highly computational tasks. Validation on a publicly available vehicle logo dataset shows that the proposed classifier can achieve up to 98% recognition accuracy as compared with the state-of-the-art non-parametric classifiers. Compared with the generic Bayesian compressive sensing classification, the proposed approach decreases the mean number of misclassifications by 87% and with 68% reduction of the computational time. The robustness of the BBCS approach is demonstrated over scene recognition tasks, and its outperformance over the AlexNet convolutional neural networks algorithm is demonstrated in noisy conditions. The proposed BBCS approach is generic and can be used in different areas, for example, it has shown robustness over the CIFAR-10 dataset.
Faraday discussions, 2021
2018 21st International Conference on Information Fusion (FUSION), 2018
Faraday discussions, 2021
2018 Sensor Data Fusion: Trends, Solutions, Applications (SDF), 2018
This paper proposes a deep learning approach for traffic flow prediction in complex road networks... more This paper proposes a deep learning approach for traffic flow prediction in complex road networks. Traffic flow data from induction loop sensors are essentially a time series, which is also spatially related to traffic in different road segments. The spatio-temporal traffic data can be converted into an image where the traffic data are expressed in a 3D space with respect to space and time axes. Although convolutional neural networks (CNNs) have been showing surprising performance in understanding images, they have a major drawback. In the max pooling operation, CNNs are losing important information by locally taking the highest activation values. The interrelationship in traffic data measured by sparsely located sensors in different time intervals should not be neglected in order to obtain accurate predictions. Thus, we propose a neural network with capsules that replaces max pooling by dynamic routing. This is the first approach that employs the capsule network on a time series forecasting problem, to our best knowledge. Moreover, an experiment on real traffic speed data measured in the Santander city of Spain demonstrates the proposed method outperforms the state-of-the-art method based on a CNN by 13.1% in terms of root mean squared error.
2019 10th International Conference on Dependable Systems, Services and Technologies (DESSERT), 2019
This is a repository copy of Non-random weight initialisation in deep learning networks for repea... more This is a repository copy of Non-random weight initialisation in deep learning networks for repeatable determinism.
Neurocomputing, 2021
This paper proposes a novel framework of flame region-based convolutional neural network for auto... more This paper proposes a novel framework of flame region-based convolutional neural network for autonomous flame detection. The task of flame detection is especially challenging since flames have greater diversity in colour, texture, and shape than regular rigid objects. To cope with these difficulties due to the various appearances and unclear edges of flames, a proposal generation approach is developed to effectively select candidate flame regions based on two crucial properties of flames, i.e., their dynamics and colours. The candidate flame regions together with a convolutional feature map are further processed by additional layers to output detected flames. The diversity in flame colours is well represented by approximating the distribution using a Dirichlet Process Gaussian mixture model with variational inference. The proposed framework is evaluated on publicly available videos and achieves an average frame-wise accuracy higher than 88%, which outperforms the state-of-the-art methods.
2017 20th International Conference on Information Fusion (Fusion), Jul 1, 2017
Vehicle logo recognition is an important part of vehicle identification in intelligent transporta... more Vehicle logo recognition is an important part of vehicle identification in intelligent transportation systems. Stateof-the-art vehicle logo recognition approaches typically consider training models on large datasets. However, there might only be a small training dataset to start with and more images can be obtained during the real-time applications. This paper proposes an online image recognition framework which provides solutions for both small and large datasets. Using this recognition framework, models are built efficiently using a weight updating scheme. Another novelty of this work is that the Cauchy prior logistic regression with conjugate gradient descent is proposed to deal with the multinomial classification tasks. The Cauchy prior results in a quicker convergence speed for the weight updating process which could decrease the computational cost for both online and offline methods. By testing with a publicly available dataset, the Cauchy prior logistic regression deceases the classification time by 59%. An accuracy of up to 98.80% is achieved when the proposed framework is applied.
IEEE Transactions on Aerospace and Electronic Systems, 2019
A novel robust Rauch-Tung-Striebel smoothing framework is proposed based on a generalized Gaussia... more A novel robust Rauch-Tung-Striebel smoothing framework is proposed based on a generalized Gaussian scale mixture (GGScM) distribution for a linear state-space model with heavy-tailed and/or skew noises. The state trajectory, mixing parameters and unknown distribution parameters are jointly inferred using the variational Bayesian approach. As such, a major contribution of this work is unifying results within the GGScM distribution framework. Simulation and experimental results demonstrate that the proposed smoother has better accuracy than existing smoothers.
2019 22th International Conference on Information Fusion (FUSION)
This is a repository copy of Multi-band image fusion using Gaussian process regression with spars... more This is a repository copy of Multi-band image fusion using Gaussian process regression with sparse rational quadratic kernel.
Sensors
Railway networks systems are by design open and accessible to people, but this presents challenge... more Railway networks systems are by design open and accessible to people, but this presents challenges in the prevention of events such as terrorism, trespass, and suicide fatalities. With the rapid advancement of machine learning, numerous computer vision methods have been developed in closed-circuit television (CCTV) surveillance systems for the purposes of managing public spaces. These methods are built based on multiple types of sensors and are designed to automatically detect static objects and unexpected events, monitor people, and prevent potential dangers. This survey focuses on recently developed CCTV surveillance methods for rail networks, discusses the challenges they face, their advantages and disadvantages and a vision for future railway surveillance systems. State-of-the-art methods for object detection and behaviour recognition applied to rail network surveillance systems are introduced, and the ethics of handling personal data and the use of automated systems are also co...
2019 IEEE International Symposium on Signal Processing and Information Technology (ISSPIT), 2019
This paper describes a machine assistance approach to grading decisions for values that might be ... more This paper describes a machine assistance approach to grading decisions for values that might be missing or need validation, using a mathematical algebraic form of an Expert System, instead of the traditional textual or logic forms and builds a neural network computational graph structure. This Experts System approach is also structured into a neural network like format of: input, hidden and output layers that provide a structured approach to the knowledge-base organization, this provides a useful abstraction for reuse for data migration applications in big data, Cyber and relational databases. The approach is further enhanced with a Bayesian probability tree approach to grade the confidences of value probabilities, instead of the traditional grading of the rule probabilities, and estimates the most probable value in light of all evidence presented. This is ground work for a Machine Learning (ML) experts system approach in a form that is closer to a Neural Network node structure.
2019 Sensor Data Fusion: Trends, Solutions, Applications (SDF), 2019
This is a repository copy of Leveraging uncertainty in adversarial learning to improve deep learn... more This is a repository copy of Leveraging uncertainty in adversarial learning to improve deep learning based segmentation.
IEEE Transactions on Aerospace and Electronic Systems, 2021
This paper proposes a recursive Gaussian process regression with a joint optimization-based itera... more This paper proposes a recursive Gaussian process regression with a joint optimization-based iterative learning control algorithm to estimate and predict disturbances and model uncertainties affecting a flight. The algorithm proactively compensates for the predicted disturbances, improving precision in aircraft trajectory tracking. Higher precision in trajectory tracking implies an improvement of the aircraft trajectory predictability and therefore of the air traffic management system efficiency. Airlines can also benefit from this higher predictability by reducing the number of alterations when following their designed trajectories, which entails a reduction of costs and emissions. The iterative learning control algorithm is divided into two steps: first, a recursive Gaussian process regression estimates and predicts perturbations and model errors with no need for prior knowledge about their dynamics and with low computational cost, and second, this information is used to update the control inputs so that the subsequent aircraft intending to fly the same planned trajectory will follow it with greater precision than the previous ones. This method is tested on a simulated commercial aircraft performing a continuous climb operation and compared to an iterative learning algorithm using a Kalman filter estimator in a similar scenario. The results show that the proposed approach provides 62% and 42% precision improvement in tracking the desired trajectory, as compared to the Kalman filter approach, in two experiments where no prior knowledge of the unmodeled dynamics was available, also achieving it in less iterations.
Signal Processing, 2021
Although kernel approximation methods have been widely applied to mitigate the O(n 3) cost of the... more Although kernel approximation methods have been widely applied to mitigate the O(n 3) cost of the n × n kernel matrix inverse in Gaussian process methods, they still face computational challenges. The 'residual' matrix between the covariance and the approximating component is often discarded as it prevents the computational cost reduction. In this paper, we propose a computationally efficient Gaussian process approach that achieves better computational efficiency, O(mn 2), compared with standard Gaussian process methods, when using m ≪ n data. The proposed approach incorporates the 'residual' matrix in its symmetric diagonally dominant form which can be further approximated by the Neumann series. We have validated and compared the approach with full Gaussian process approaches and kernel approximation based Gaussian process
Engineering Applications of Artificial Intelligence, 2021
Vehicle localisation is an important and challenging task in achieving autonomous driving. This w... more Vehicle localisation is an important and challenging task in achieving autonomous driving. This work presents a box particle filter framework for vehicle selflocalisation in the presence of sensor and map uncertainties. The proposed feature-refined box particle filter incorporates line features extracted from a multi-layer Light Detection And Ranging (LiDAR) sensor and information from OpenStreetMap to estimate the vehicle state. A particle weight balance strategy is incorporated to account for the OpenStreetMap inaccuracy, which is assessed by comparing it to a high definition road map. The performance of the proposed framework is evaluated on a LiDAR dataset and compared with box particle filter variants. Experimental results show that the proposed framework achieves respectively 10% and 53% localisation accuracy improvement with reduced box volumes of 25% and 41%, when compared with the state-of-the-art interval analysis based box regularisation particle filter and the box particle filter.
Ad Hoc Networks, 2020
The growth of connected vehicles in smart cities increases the number of information being commun... more The growth of connected vehicles in smart cities increases the number of information being communicated on the Internet of Vehicle networks. This causes wireless channel congestion problems, which degrades the network performance and reliability due to the low throughput, high average delay and the high packets loss. Therefore, this paper proposes a non-cooperative game approach to control congestion in the vehicular ad-hoc network channel where the nodes behave as selfish players requesting high data transmission rates. Moreover, the satisfaction of the Nash equilibrium condition for the optimum data transmission rate for each vehicle, is proven. A utility function is introduced based on data transmission rates, the priority of vehicles and contention delay in order to obtain the optimal rates. The performance of the proposed approach has been evaluated and validated in comparison with three others approaches over two testing scenarios for highway and urban traffic. The results show that the network performance and efficiency have been improved by an overall average of 35%, 30% and 37.17% in terms of packets loss, channel busy time and number of collision messages, respectively, as compared with the state-of-the-art-strategies for the highway testing scenario. Similar performance is achieved for the urban testing scenario.
Future Generation Computer Systems, 2019
This is a repository copy of Information and resource management systems for Internet of Things: ... more This is a repository copy of Information and resource management systems for Internet of Things: Energy management, communication protocols and future applications.
Signal Processing, 2021
The performance of the state estimation for Gaussian state space models can be degraded if the mo... more The performance of the state estimation for Gaussian state space models can be degraded if the models are affected by the non-Gaussian process and measurement noises with uncertain degree of non-Gaussianity. In this paper, we propose a flexible robust Student's t-based multimodel approach. More specifically, the degrees of freedom parameter from the Student's t-distribution is assumed unknown and modelled by a Markov chain of state values. In order to capture more information of the Student's tdistributions propagated through multiple models, we establish a model-based Versoria cost function in the form of a weighted mixture rather than the original form, and maximize the function to interact and fuse the multiple models. Simulated results prove the flexibility of the robustness of the proposed Student's t-basedmultimodel approach when the existence probability of the outliers is uncertain.
This paper presents a novel non-parametric backpropagation Bayesian compressive sensing (BBCS) cl... more This paper presents a novel non-parametric backpropagation Bayesian compressive sensing (BBCS) classification approach. While the state-of-the-art parametric classifiers such as logistic regression require model training and can result in inadequate models, the developed approach does not require model training. It is combined with a column-based subspace sampling process and it can deal efficiently with uncertainties and highly computational tasks. Validation on a publicly available vehicle logo dataset shows that the proposed classifier can achieve up to 98% recognition accuracy as compared with the state-of-the-art non-parametric classifiers. Compared with the generic Bayesian compressive sensing classification, the proposed approach decreases the mean number of misclassifications by 87% and with 68% reduction of the computational time. The robustness of the BBCS approach is demonstrated over scene recognition tasks, and its outperformance over the AlexNet convolutional neural networks algorithm is demonstrated in noisy conditions. The proposed BBCS approach is generic and can be used in different areas, for example, it has shown robustness over the CIFAR-10 dataset.
Faraday discussions, 2021
2018 21st International Conference on Information Fusion (FUSION), 2018