In recent years, significant advancements have been made in the field of autonomous driving with ... more In recent years, significant advancements have been made in the field of autonomous driving with the aim of increasing safety and efficiency. However, research that focuses on tractortrailer vehicles is relatively sparse. Due to the physical characteristics and articulated joints, such vehicles require tailored models. While turning, the back wheels of the trailer turn at a tighter radius and the truck often has to deviate from the centre of the lane to accommodate this. Due to the lack of publicly available models, this work develops truck and trailer models using the high-fidelity simulation software Carla, together with several roundabout scenarios, to establish a baseline dataset for benchmarks. Using a twin-q soft actor-critic algorithm, we train a quasi-end-to-end autonomous driving model which is able to achieve a 73% success rate on different roundabouts.
A feasible Mobile Positioning solution is often sought after by network operators and service pro... more A feasible Mobile Positioning solution is often sought after by network operators and service providers alike. Location-dependent applications create a new domain of services which might not only be of interest to the next generation of mobile users but also create new potential revenue streams. Applications vary from emergency services and tracking to location-based information services, location-based billing and location-dependent advertising. Due to the shortcomings of location-related information present in GSM networks, and the lack of positioning functionality in most of the commonly sold mobile devices, a straightforward solution for mobile positioning does not currently exist. This research intends to propose cellular positioning methods which do not require any significant changes to the network or the mobile device itself, which are feasible and cost effective, and which provide sufficient accuracy for certain categories of location-based services. These techniques are ba...
Non-linear continuous change is common in real-world problems, especially those that model physic... more Non-linear continuous change is common in real-world problems, especially those that model physical systems. We present an algorithm which builds upon existent temporal planning techniques based on linear programming to approximate non-linear continuous monotonic functions. These are integrated through a semantic attachment mechanism, allowing external libraries or functions that are difficult to model in native PDDL to be evaluated during the planning process. A new planning system implementing this algorithm was developed and evaluated. Results show that the addition of this algorithm to the planning process can enable it to solve a broader set of planning problems.
Fifteenth ACM Conference on Recommender Systems, 2021
Music recommendation systems typically use collaborative filtering to determine which songs to re... more Music recommendation systems typically use collaborative filtering to determine which songs to recommend to their users. This mechanism matches a user with listeners that have similar tastes, and uses their listening history to find songs that the user will probably like. The fundamental issue with this approach is that artists already need to have a significant user following to get a fair chance of being recommended. This is known as the music cold-start problem. In this work, we investigate the possibility of making music recommendations based on audio content so that new artists still get a good chance of being recommended, even if they do not have a sufficient number of listeners yet. We propose the use of Siamese Neural Networks (SNNs) to determine the similarity between two audio clips. Each clip is first pre-processed into a Mel-Spectrogram, which is then used as input to an SNN consisting of two identical Convolutional Neural Networks (CNNs). The output of each CNN is then compared together to determine whether two songs are similar or not. These were trained using audio from the Free Music Archive, with the genre used as a heuristic to determine the similarity between song pairs. A query-by-multiple-example (QBME) music recommendation system was developed that makes use of the proposed contentbased similarity metric to find songs that match the user's tastes. This was packaged inside an online blind-test survey, which first prompts participants to select a set of preferred songs, and then recommends a number of songs which the subject is expected to listen to and rate on a Likert scale. The recommendations from the proposed algorithm were stochastically interleaved with songs selected randomly from the preferred genres of the user, as a baseline for comparison. The participants were not aware that the recommendations came from two different algorithms. Our findings show that 60.7% of the 150 participants gave higher ratings to the recommendations made by the proposed SNN-based algorithm. Findings also show that 55% of the recommended songs had less than 1,500 listens, demonstrating that the proposed contentbased approach can provide a fairer exposure to all artists based on their music, independent of their fame and popularity.
Load modelling and simulation of household electricity consumption for the evaluation of demand-side management strategies
IEEE PES ISGT Europe 2013, 2013
In order to evaluate the effectiveness of demand-side management techniques, mechanisms to simula... more In order to evaluate the effectiveness of demand-side management techniques, mechanisms to simulate electricity consumption activities at a granular level are required. A bottom-up approach that uses a non-homogeneous Markov chain to model each appliance within each household is proposed. This model is time-aware and captures the variability of the transition probabilities as they change throughout the day. A simulator was developed based on this model and it was configured with data obtained from a household electricity survey conducted in the UK. The resultant load curves from a simulation of a thousand households are compared with the average hourly load reported in the survey, with significant similarities observed between the two. The same model can be parametrised to simulate hypothetical scenarios, such as a future where electric vehicles are more popular. The simulation framework also supports a plug-in mechanism through which demand control policies can be integrated into the system such that the effects and performance of demand-side management strategies can be evaluated.
IEEE Transactions on Artificial Intelligence, 2022
Temporal planning often involves numeric effects that are directly proportional to their action's... more Temporal planning often involves numeric effects that are directly proportional to their action's duration. These include continuous effects, where a numeric variable is subjected to a rate of change while the action is being executed, and discrete duration-dependent effects, where the variable is updated instantaneously but the magnitude of such change is computed from the action's duration. When these effects are linear, state-of-theart temporal planners often make use of Linear Programming to ensure that these numeric updates are consistent with the chosen start times and durations of the plan's actions. This is typically done for each evaluated state as part of the search process. This exhaustive approach is not scalable to solve realworld problems that require long plans, because the linear program's size becomes larger and slower to solve. In this work we propose techniques that minimise this overhead by computing these checks more selectively and formulating linear programs that have a smaller footprint. The effectiveness of these techniques is demonstrated on domains that use a mix of discrete and continuous effects, which is typical of real-world planning problems. The resultant planner also outperforms most state-ofthe-art temporal-numeric and hybrid planners, in terms of both coverage and scalability. Impact Statement-While current temporal planners perform well on benchmark domains that feature continuous or durationdependent numeric effects, real-world problems are often more challenging to solve. This is mostly due to the complex interactions between the temporal and numeric trajectory dynamics, together with the size of the plans that such planning problems typically require. The techniques proposed in this work take advantage of the structure of such domains to find a plan more efficiently. Empirical testing shows a significant improvement on the time spent on solving linear programs during the search for a plan, with an overall improvement on the total planning time for problems that exhibit such characteristics. Furthermore, the proposed planner outperforms all other state-of-the-art temporalnumeric planners in terms of both coverage and scalability. This is a significant achievement that enables A.I. Planning technology to solve a wider range of problems for industry applications.
Real-time EEG-based Emotion Recognition (EEG-ER) with consumer-grade EEG devices involves classif... more Real-time EEG-based Emotion Recognition (EEG-ER) with consumer-grade EEG devices involves classification of emotions using a reduced number of channels. These devices typically provide only four or five channels, unlike the high number of channels (32 or more) typically used in most current state-of-the-art research. In this work we propose to use Discrete Wavelet Transforms (DWT) to extract time-frequency domain features, and we use time-windows of a few seconds to perform EEG-ER classification. This technique can be used in real-time, as opposed to post-hoc on the full session data. We also apply baseline removal preprocessing, developed in prior research, to our proposed DWT Entropy and Energy features, which improves classification accuracy significantly. We consider two different classifier architectures, a 3D Convolutional Neural Network (3D CNN) and a Support Vector Machine (SVM). We evaluate both models on subject-independent and subject dependent setups to classify the Valence and Arousal dimensions of an individual's emotional state. We test them on both the full 32-channel data provided by the DEAP dataset, and also a reduced 5-channel extract of the same dataset. The SVM model performs best on all the presented scenarios, achieving an accuracy of 95.32% on Valence and 95.68% on Arousal for the full 32-channel subject-dependent case, beating prior real-time EEG-ER subject-dependent benchmarks. On the subject-independent case an accuracy of 80.70% on Valence and 81.41% on Arousal was also obtained. Reducing the input data to 5 channels only degrades the accuracy by an average of 3.54% across all scenarios, making this model appropriate for use with more accessible low-end EEG devices.
Real-time EEG-based Emotion Recognition (EEG-ER) with consumer-grade EEG devices involves classif... more Real-time EEG-based Emotion Recognition (EEG-ER) with consumer-grade EEG devices involves classification of emotions using a reduced number of channels. These devices typically provide only four or five channels, unlike the high number of channels (32 or more) typically used in most current state-of-the-art research. In this work we propose to use Discrete Wavelet Transforms (DWT) to extract time-frequency domain features, and we use time-windows of a few seconds to perform EEG-ER classification. This technique can be used in real-time, as opposed to post-hoc on the full session data. We also apply baseline removal preprocessing, developed in prior research, to our proposed DWT Entropy and Energy features, which improves classification accuracy significantly. We consider two different classifier architectures, a 3D Convolutional Neural Network (3D CNN) and a Support Vector Machine (SVM). We evaluate both models on subject-independent and subject dependent setups to classify the Valence and Arousal dimensions of an individual's emotional state. We test them on both the full 32-channel data provided by the DEAP dataset, and also a reduced 5-channel extract of the same dataset. The SVM model performs best on all the presented scenarios, achieving an accuracy of 95.32% on Valence and 95.68% on Arousal for the full 32-channel subject-dependent case, beating prior real-time EEG-ER subject-dependent benchmarks. On the subject-independent case an accuracy of 80.70% on Valence and 81.41% on Arousal was also obtained. Reducing the input data to 5 channels only degrades the accuracy by an average of 3.54% across all scenarios, making this model appropriate for use with more accessible low-end EEG devices.
Real-world planning problems often feature complex temporal and numeric characteristics. These in... more Real-world planning problems often feature complex temporal and numeric characteristics. These include concurrent activities and also effects that involve continuous change. This work presents the formalism behind reasoning with required concurrency that involves continuous change in temporal planning problems, together with a set of techniques to solve a class of tasks that to date are hard to solve with current state-of-the-art temporal planners. The motivation for this work is scenarios where planning actions have rich numeric effects on some physical system. One such example is automated demand dispatch for electricity provision, where actions that fulfil customer requirements have an effect on various metrics, such as wattage or costs, which could be subject to operational or commercial constraints. An algorithm that handles discrete interference of linear continuous effects, referred to as constants in context, is presented. This technique allows discrete actions to update the rate of change of a continuous effect taking place concurrently. This work builds on techniques used in current temporal planners that make use of linear programming, and also extends the heuristic to guide the search to a solution. This algorithm was implemented in a new temporal and numeric planner called DICE and evaluated with some benchmark domains. PDDL, the current de facto standard language for planning domains and corresponding planning tasks, was extended to support interactions with external class modules. The proposed extension, PDDLx, defines a generic planner-solver interface for both discrete and continuous effects. This enables planners that implement this interface to interact with external solvers and incorporate context-specific effects in a black-box fashion, enabling complex numeric behaviour to be encapsulated within such modules. Non-linear monotonic continuous effects, defined in the proposed PDDLx extension, are integrated within the planner using a non-linear iterative convergence algorithm. It searches for a linear approximation within an acceptable configurable error margin, which is then used within the linear program computed for each temporal state. This algorithm proves to be effective in various domains where non-linear continuous behaviour is prevalent. This technique was implemented as an extension to DICE, called uNICOrn, which performs non-linear iterative convergence for continuous effects whose duration needs to be determined by the planner. uNICOrn was also evaluated with some benchmark non-linear domains. A case study on the automated demand dispatch domain is presented to demonstrate the use of the planning algorithms proposed in this thesis. Linear and non-linear planning problems are evaluated and the performance of uNICOrn on these problem instances was analysed. This work builds on current techniques used for temporal planning with continuous numeric behaviour using linear programming, and enhances them to remove some of their intrinsic limitations. The result is a set of algorithms that are more effective in solving real-world applications that involve continuous change and rich numeric behaviour. This work would not have been possible without the academic, moral and financial support of various people and organisations. I am deeply grateful to my supervisors, Professor Maria Fox and Professor Derek Long, for their guidance, inspiration and continuous encouragement. Apart from their in-depth knowledge, they provided me with unique opportunities to get international exposure within the A.I. community, which has enriched my experience even further. This research was funded and supported by the UK Engineering and Physical Sciences Research Council (EPSRC) as part of the project entitled The Autonomic Power System (Grant Ref: EP/I031650/1). I would like to thank all the researchers participating in the project for their insight and interesting discussions during the various meetings and events. I would also like to thank my colleagues and friends at King's College London, whose company and friendship made the whole experience even more pleasant and enjoyable. The regular meetings of the King's A.I. Planning Research Group were also instrumental to generate ideas and gather invaluable feedback throughout my research. Last and foremost, I would like to show my utmost gratitude to my wife Thérèse, my parents, and the rest of our respective families for their immeasurable patience and support throughout my academic and professional endeavours.
With the ever-growing variety of object detection
approaches, this study explores a series of exp... more With the ever-growing variety of object detection approaches, this study explores a series of experiments that combine reinforcement learning (RL)-based visual attention methods with saliency ranking techniques to investigate transparent and sustainable solutions. By integrating saliency ranking for initial bounding box prediction and subsequently applying RL techniques to refine these predictions through a finite set of actions over multiple time steps, this study aims to enhance RL object detection accuracy. Presented as a series of experiments, this research investigates the use of various image feature extraction methods and explores diverse Deep Q-Network (DQN) architectural variations for deep reinforcement learning-based localisation agent training. Additionally, we focus on optimising the detection pipeline at every step by prioritising lightweight and faster models, while also incorporating the capability to classify detected objects, a feature absent in previous RL approaches. We show that by evaluating the performance of these trained agents using the Pascal VOC 2007 dataset, faster and more optimised models were developed. Notably, the best mean Average Precision (mAP) achieved in this study was 51.4, surpassing benchmarks set by RL-based single object detectors in the literature.
Uploads
Papers by Josef Bajada
approaches, this study explores a series of experiments that combine
reinforcement learning (RL)-based visual attention methods
with saliency ranking techniques to investigate transparent
and sustainable solutions. By integrating saliency ranking for
initial bounding box prediction and subsequently applying RL
techniques to refine these predictions through a finite set of
actions over multiple time steps, this study aims to enhance RL
object detection accuracy. Presented as a series of experiments,
this research investigates the use of various image feature
extraction methods and explores diverse Deep Q-Network (DQN)
architectural variations for deep reinforcement learning-based
localisation agent training. Additionally, we focus on optimising
the detection pipeline at every step by prioritising lightweight and
faster models, while also incorporating the capability to classify
detected objects, a feature absent in previous RL approaches. We
show that by evaluating the performance of these trained agents
using the Pascal VOC 2007 dataset, faster and more optimised
models were developed. Notably, the best mean Average Precision
(mAP) achieved in this study was 51.4, surpassing benchmarks
set by RL-based single object detectors in the literature.