This thesis investigates cost-effective and reliable resource reservation strategies for time-sensitive and safety-critical vehicular (TSSCV) applications, such as autonomous and remote driving. These applications require deterministic and guaranteed access to mobile edge computing (MEC) resources, which is typically achieved through individual reservations. Vehicles submit reservation requests to mobile network operator (MNOs), which allocate computation and communication resources based on these requests. Therefore, optimizing the timing and method for vehicles to place reservation requests in both single and multiple MNO scenarios is crucial, especially in dynamic vehicular environments with fluctuating network conditions and limited resources. Efficient reservation requests are essential as vehicles lack complete information about future resource availability and costs. Additionally, real-world uncertainties, such as unpredictable mobility, which influences future reservation times and costs, complicate the design of an optimal reservation strategy. Furthermore, dynamic pricing models employed by MNOs introduce another layer of complexity to the decision-making process for reservation requests, updates, and exchanges due to their impact on market conditions like resource supply and demand. To address the challenges of resource reservation in single MNO scenarios, this thesis proposes an advanced reservation strategy leveraging a batched long short-term memory (LSTM) model. This approach optimizes the timing of reservations, leading to significant cost savings for vehicles. To further minimize update costs, one-shot and multi-shot reservation update strategies are introduced, complemented by the heuristic greedy reservation updates (HGRU) algorithm. For multiple MNO environments, the thesis addresses cost-effective resource selection by comparing prices and network conditions across MNOs. An adaptive Markov decision process (MDP) framework is proposed, incorporating a deep reinforcement learning (DRL) algorithm, specifically dueling deep Q-learning. To enhance learning efficiency, a novel area-wise approach and adaptive MDP closely resembling real-world conditions are introduced. Furthermore, the temporal fusion transformer (TFT) is employed to effectively handle time-dependent data during model training. The multi-phase training approach, involving both synthetic and real-world data, enables the DRL agent to learn from historical data and adapt to real-time observations. Additionally, a multi-objective approach using a double deep Q-learning algorithm is proposed to minimize the cost of reservation updates while ensuring an optimal strategy and reliable provisioning. Finally, this thesis explores the use of blockchain smart contracts to establish a secure, efficient, and transparent resource trading system for vehicular networks. This blockchain-based architecture optimizes reservation costs and addresses trust issues by enabling decentralized, secure, and cost-effective resource trading among vehicles. By leveraging smart contracts, the system ensures transparency and immutability in transactions, fostering trust among participating vehicles. Simulation results demonstrate that the proposed resource reservation algorithms in single MNO environments outperform benchmark reservation schemes, including immediate reservation schemes, in terms of cost minimization and resource utilization efficiency, which also pose challenges for resource guarantee. In multiple MNO environments, the algorithms effectively manage uncertainties and promote competition between MNOs than single MNO scenario potentially impacting guaranteed resource provisioning. Additionally, while the exchange reservation strategies enhance security, the absence of robust security mechanisms could result in unreliable resource requests from providers, posing challenges to guaranteed resource provisioning.

Emerging machine learning (ML) techniques are showing great potential in prediction performance. However, research and development is often conducted in an environment with extensive computational resources and blinded by prediction performance. In reality, computational resources might be contained on constrained hardware where energy and memory consumption must be restrained. Furthermore, shortages of sufficiently large datasets for ML is a frequent problem, combined with the cost of data retention. This generates a significant demand for sustainable ML. With sustainable ML, practitioners can train ML models on less data, which reduces memory and energy consumption during the training process. To explore solutions to these problems, this thesis dives into several techniques that have been introduced in the literature to achieve energy-savings when training machine learning models. These techniques include Quantization-Aware Training, Model Distillation, Quantized Distillation, Continual Learning and a deeper dive into Siamese Neural Networks (SNNs), one of the most promising techniques for sustainability. Empirical evaluations are conducted using several datasets to illustrate the potential of these techniques and their contribution to sustainable ML. The findings of this thesis show that the energy-saving techniques could be leveraged in some cases to make machine learning models more manageable and sustainable whilst not compromising significant model prediction performance. In addition, the deeper dive into SNNs shows that SNNs can outperform standard classification networks, under both the standard multi-class classification case and the Continual Learning case, whilst being trained on significantly less data. / Maskininlärning har i den senaste tidens forskning visat stor potential och hög precision inom klassificering. Forskning, som ofta bedrivs i en miljö med omfattande beräkningsresurser, kan lätt bli förblindad av precision. I verkligheten är ofta beräkningsresurser lokaliserade på hårdvara där energi- och minneskapacitet är begränsad. Ytterligare ett vanligt problem är att uppnå en tillräckligt stor datamängd för att uppnå önskvärd precision vid träning av maskininlärningsmodeller. Dessa problem skapar en betydande efterfrågan av hållbar maskininlärning. Hållbar maskininlärning har kapaciteten att träna modeller på en mindre datamängd, vilket minskar minne- och energiförbrukning under träningsprocessen. För att utforska hållbar maskininlärning analyserar denna avhandling Quantization-Aware Training, Model Distillation, Quantized Distillation, Continual Learning och en djupare evaluering av Siamesiska Neurala Nätverk (SNN), en av de mest lovande teknikerna inom hållbar maskininlärning. Empiriska utvärderingar utfördes med hjälp av flera olika datamängder för att illustrera potentialen hos dessa tekniker. Resultaten visar att energibesparingsteknikerna kan utnyttjas för att göra maskininlärningsmodeller mer hållbara utan att kompromissa för precision. Dessutom visar undersökningen av SNNs att de kan överträffa vanliga neurala nätverk, med och utan Continual Learning, även om de tränas på betydligt mindre data.