https://www.um.edu.mt/library/oar/handle/123456789/8421 Wed, 22 Apr 2026 03:03:49 GMT 2026-04-22T03:03:49Z https://www.um.edu.mt/library/oar/handle/123456789/145603 Title: AI for partially observable stochastic games Abstract: Reinforcement Learning (RL) has shown significant success in a variety of game environments. Despite this, some game characteristics continue to challenge RL. Jaipur, a competitive two‐player turn‐based board game, contains a number of such challenging features. It is characterised by partial observability, stochasticity and a large discrete action space of 25,499 possible actions. Moreover, it also contains elements of randomness, immediate and long‐term rewards with different consequences, and multiple different strategies that can be adopted. This study aims to analyse the performance of several state‐of‐the‐art techniques and algorithms that can be implemented, to not only mitigate the complexities of applying RL on Jaipur, but also improve the performance and training efficiency. We propose the implementation of the action masking, action embedding, hierarchical RL with centralised critic and policy cloning techniques. Moreover, hyperparameter tuning was applied using the PBT algorithm, and to evaluate the effects of partial observability on the RL process, different levels of observability were provided in separate experiments. For each implementation, the PPO, A2C, DQN and DDQN algorithms were trained with two separate policies, to reflect Jaipur’s two competitive players. The scores obtained during training, as well as when the policies of each model were played against each other on 1000 unique games, were evaluated quantitatively against scores obtained by human players and by the other models. The results demonstrate that all algorithms and techniques achieved good scores, comparable to those achieved by humans. Action masking delivered the best overall performance, with high scores achieved at high computational efficiency across most algorithms. Action embedding obtained better scores for PPO but required the longest training times, whilst the training times for DQN and DDQN were the shortest. Meanwhile hierarchical RL with centralised critic provided greater training stability, however, the scores achieved were lower across most models and the training times were significantly prolonged. Both the hyper‐parameter tuning and policy cloning technique proved to be beneficial, as the performance of the algorithms increased in less training steps. Meanwhile, the varying levels of observability had minimal impact on the policies’ performance, suggesting that the algorithms managed to discover strong strategies that achieved high scores even with partial observability. Furthermore, an action selection analysis of the policies’ decisions during simulated games was carried out, from which it was concluded that all the policies adopted intelligent and interesting strategies, similar to those of human players. Description: M.Sc.(Melit.) Thu, 01 Jan 2026 00:00:00 GMT https://www.um.edu.mt/library/oar/handle/123456789/145603 2026-01-01T00:00:00Z https://www.um.edu.mt/library/oar/handle/123456789/145526 Title: Unified load balancing strategies for enhanced cloud computing solutions Abstract: Cloud computing offers scalable, on-demand resources that enable a variety of services and applications. Effective load balancing in cloud environments is essential for maintaining performance and Quality of Service (QoS). These environments present complex, dynamic conditions that make efficient load balancing challenging. Many existing algorithms focus on single-objective optimisation, such as minimising response time, which often results in trade-offs and inefficiencies when dealing with unpredictable workloads. This dissertation tackles these inefficiencies by introducing a unified, multi-objective load balancing strategy that combines Ant Colony Optimisation (ACO) and Genetic Algorithm (GA) techniques. The hybrid ACO-GA algorithm is implemented within the CloudAnalyst simulation environment, leveraging ACO’s rapid local search and GA’s global exploration capabilities to dynamically balance workloads across cloud resources. Extensive simulation experiments demonstrate that the proposed hybrid approach significantly improves key QoS metrics compared to both conventional and state-of-the-art load balancers. The ACO-GA consistently achieved substantially lower average response times and improved load distribution relative to traditional algorithms. For example, under light workloads it reduced mean response time by roughly 50% versus Round Robin and 40% under heavy loads. The hybrid method also outperformed modern heuristics, sustaining about 8–10% faster response than advanced metaheuristic policies while shortening data centre processing delays. These gains were accompanied by more efficient resource utilisation, as the algorithm prevented server overloading and underutilisation through balanced task allocation. Notably, performance improvements persisted across both low and high demand scenarios, highlighting the algorithm’s robust adaptability to dynamic cloud conditions. Overall, the results affirm that this unified ACO-GA strategy effectively addresses the limitations of single-objective approaches, offering a significant enhancement in cloud service performance, resource utilisation and QoS. Description: M.Sc. ICT(Melit.) Wed, 01 Jan 2025 00:00:00 GMT https://www.um.edu.mt/library/oar/handle/123456789/145526 2025-01-01T00:00:00Z https://www.um.edu.mt/library/oar/handle/123456789/145400 Title: Just-in-time framework for 8-bit systems Abstract: This project presents an emulator-integrated instrumentation framework designed to make runtime behaviour easier to observe during development and validation. The primary aim is to capture useful execution information with minimal disruption to normal emulation, allowing developers to inspect what the emulator is doing rather than relying on guesswork when behaviour differs from expectations. The framework is built around lightweight hooks that record instruction flow, memory access activity, and key events such as interrupts, and it exposes this information through a set of interactive tools including tracing, breakpoints, and memory activity visualisation. The framework is implemented as part of a complete emulator, using the Nintendo Entertainment System (NES) as a case study. The NES provides a well-documented and timing-sensitive platform where small behavioural differences are easy to trigger and useful to analyse. To evaluate the effectiveness of the design, the emulator was tested using established NES accuracy test ROM suites and results were compared against a reference emulator (Mesen). The evaluation focuses on whether the emulator reproduces expected behaviour on representative correctness and timing tests, and whether the integrated tooling provides practical value for fault isolation and behavioural inspection. While the project does not aim to replace mature emulators, it prioritises a clear architecture, extensible instrumentation, and practical usability, providing a solid foundation for further experimentation with correctness analysis and future optimisation work, such as performance profiling or dynamic binary translation. Description: B.Sc. (Hons)(Melit.) Thu, 01 Jan 2026 00:00:00 GMT https://www.um.edu.mt/library/oar/handle/123456789/145400 2026-01-01T00:00:00Z https://www.um.edu.mt/library/oar/handle/123456789/145399 Title: Accelerating diffuse indirect lighting with irradiance caching Abstract: Diffuse indirect lighting is a subset of global illumination that can be quite computationally expensive to compute. While Path Tracing is the standard for computing it, several techniques have been proposed to accelerate its computation, one of which being Irradiance Caching. In order to investigate whether the gain in performance by the irradiance cache is feasible in terms of change in quality, a multithreaded C++ implementation was developed and integrated into a custom hybrid Path-Tracing and Distributed Ray-tracing solution. The solution was evaluated using a set of test scenes and a number of sample counts, of which the results were compared to that of a reference path tracer. The two solutions are compared based on performance metrics such as render time, and quality metrics such as SSIM and PSNR. Based on the results obtained, it was concluded that the solution offers a significant gain in performance, and is deemed appropriate to replace path tracing in cases where accuracy is not the top priority. However, when accuracy and unbiased rendering are the top priority, the more traditional method of path tracing would be more suitable. Description: B.Sc. (Hons)(Melit.) Thu, 01 Jan 2026 00:00:00 GMT https://www.um.edu.mt/library/oar/handle/123456789/145399 2026-01-01T00:00:00Z