Abstract
The effectiveness of optimizing complex problems is closely linked to the configuration of parameters in search algorithms, especially when considering metaheuristic optimization models. Although various automated methods for algorithm configuration have been proposed to alleviate users from manually tuning parameters, there is still unexplored potential in dynamically adjusting certain algorithm parameters during execution, which can lead to enhanced performance. The main objective is to comparatively analyze the effectiveness of manual parameter tuning compared to a dynamic online configuration approach based on reinforcement learning. To this end, the State-Action-Reward-State-Action (SARSA) algorithm is adapted to adjust the parameters of a genetic algorithm, namely population size, crossover rate, mutation rate, and number of generations. Tests are conducted with these two methods on benchmark functions commonly used in the literature. Additionally, the proposed model has been evaluated in a practical problem of optimizing energy trading portfolios in the electricity market. Results indicate that the reinforcement learning-based algorithm tends to achieve seemingly better results than manual configuration, while maintaining very similar execution times. This result suggests that online parameter tuning approaches may be more effective and offer a viable alternative for optimization in metaheuristic algorithms.

Abstract
[No abstract available]

Abstract

Abstract

Abstract

Abstract
Deep Learning (DL) has emerged as a powerful tool for vulnerability detection, often outperforming traditional solutions. However, developing effective DL models requires large amounts of real-world data, which can be difficult to obtain in sufficient quantities. To address this challenge, DiverseVul dataset has been curated as one of the largest datasets of vulnerable and non-vulnerable C/C++ functions extracted exclusively from real-world projects. Its goal is to provide high-quality, large-scale samples for training DL models. Nevertheless, during our study several inconsistencies were identified in the raw dataset while applying pre-processing techniques, highlighting the need for a refined version. In this work, we present a refined version of DiverseVul dataset, which is used to fine-tune a large language model, LLaMA 3.2, for vulnerability detection. Experimental results show that the use of pre-processing techniques led to an improvement in performance, with the model achieving an F1-Score of 66%, a competitive result when compared to our baseline, which achieved a 47% F1-Score in software vulnerability detection.