Algorithmic trading already dominates modern financial markets, yet most live systems still rely on fixed heuristics that falter when conditions change. Deep reinforcement learning agents promise adaptive decision making, but their behaviour is driven entirely by the reward function - a design choice that remains poorly researched. Given the critical importance of reward function design in RL problems, this paper investigates the impact of different choices of reward functions while trading in the Forex market and focusing on the EUR/USD pair. We explore different rewarding methods such as profit‑only, risk‑adjusted, multi‑objective, imitation‑learning based, and self‑rewarding mechanisms. Overall, we demonstrate the impact of careful reward engineering and whether it can boost performance and efficiency, highlighting reward design as a critical and previously under‑examined part of RL for deploying reliable trading models.

This paper examines how different exploration strategies affect the learning behavior and trading performance of reinforcement learning (RL) agents in a custom foreign exchange (forex) environment. By holding all other components constant—including model architecture, features, and reward function—the study isolates the role of exploration in deep Q-learning. Three strategies were compared: Epsilon-Greedy, Boltzmann, and Max-Boltzmann. The hybrid Max-Boltzmann approach delivered the most stable and profitable outcomes, suggesting that weighted, value-aware exploration can be beneficial in highrisk domains. The results also highlight the impact of non-Markovian structure in financial environments and limitations of equity-based rewards. Beyond empirical results, this work contributes a modular, reproducible RL framework for trading, and opens new questions about the suitability of exploration techniques in environments with asymmetric risk and irreversible actions.

This study explores how different features impact a Reinforcement Learning agent's performance in forex trading. Using a Deep Q-Network (DQN) agent and EUR/USD data from 2022-2024, we found that performance is highly sensitive to the information provided. Key findings show that for feature types like momentum and volatility, a single indicator outperformed a combination of them, as the latter tended to introduce noise. Including information about the agent's own status, such as its current trade duration, was beneficial. Counter-intuitively, providing more historical data consistently worsened performance, leading to overfitting where the agent memorized training data rather than learning general strategies. The main conclusion is that creating an effective state representation is a trade-off; the complexity of the input data must match the learning algorithm's ability to process it without overfitting.

This paper investigates the effectiveness of transfer learning techniques for accelerating the training of deep reinforcement learning (RL) agents in the foreign exchange (Forex) market. Specifically, the transfer of policies learned on the EUR/USD currency pair to the GBP/USD pair is the focus. Four transfer learning approaches are systematically compared: zero-shot transfer, full fine-tuning, partial fine-tuning, and reward-function transfer. A modular pipeline was developed, incorporating sinusoidal and trend/momentum-based market features, stepwise agent-specific metrics, and deep Q-network (DQN) architectures from the Stable Baselines 3 framework. Agent performance is evaluated through cumulative reward and Sharpe ratio metrics.
Experimental results demonstrate that partial fine-tuning accelerates initial learning by preserving generic market features acquired from the EUR/USD pair. However, our results indicate that directly training on the target currency pair yields superior ultimate performance, highlighting the nuanced limitations and potential benefits of cross-currency transfer learning in algorithmic trading.

Cyber attacks have become increasingly more prominent and the associated cost to society is by several estimates reaching trillions of US dollars. A typical cyber attack goes
through the several consecutive phases of the cyber kill chain. As a precursor for any attack, the malicious actor performs network reconnaissance in order to identify potential
entry points through exploitable services connected to the internet. Therefore, early detection of reconnaissance by scanners can mitigate or entirely prevent future attacks. Modern
intrusion detection systems are capable of blocking some scan attempts. However, more sophisticated and resourceful attackers are suspected to distribute their efforts over a large number of sources. This allows them to lower the individual scanrate while achieving the same throughput. Slow scanners are harder to detect, because they differentiate little from baseline noise levels. Additionally, a threat is severely underestimated if a large number collaborating scanners is not treated as a single entity. Therefore, the aim of this thesis is to infer such a coordinated relationship between scanners controlled by a single initiator.

We analyse the data from a network telescope with an observation of one year, much longer than previous work. Initial analysis led to the discovery of similar long-term activity patterns present in distributed scanners. These patterns can be used to uniquely identify a group which formed the basis for the correlation algorithm. We were successfully able to detect a large number of clusters employing various strategies in terms of size, scanrate and targeted services. Due to the absence of ground truth additional effort has been spent to validate potential clusters through other characteristics. We also demonstrate the utility of transforming the raw telescope data for cluster analysis through a case study of very slow scanners.

This paper looks at the different parts of the Critical Path and Resource Utilization (CPRU) heuristic for use in the Resource Constraint Project Scheduling Problem, with variable resources (RCPSP-t programming problem). RCPSP-t has many real-world instances such as in hospitals or manufacturing. Optimizing the solution generation for these instances may improve the efficiency of these industries. CPRU was split into parts and different versions were compared against VSIDS a more modern heuristic to determine if problem-specific heuristics perform better than generally good ones. A new adaptation of CPRU that adapts the research utilisation score by calculating the fraction used based on the amount of resources available given scheduled activities instead of looking at the resources of the problem instance. From the results, we concluded that CPRU may perform better for large instances of RCPSP- t compared to VSIDS in terms of generating good schedules within a few iterations. We also found that CPRU generates better schedules given a small time limit. We did not find evidence that the CPRU adaptation improved performance compared to CPRU in terms of schedule and time limits.

This paper introduces a new approach to the Preemptive Resource Constrained Project Scheduling Problem with setup times. The method makes use of a Constraint Optimization Problem solver, which has been modified to use priority-rule-based heuristics in its variable and value selection procedures. An alternative implementation which uses a combination of a priority-rule heuristic and a domain-independent solver heuristic has also been investigated. Both methods were tested on four well-known problem sets against a problem-independent solver configuration. Experimental results show a significant reduction in the time needed to find optimal solutions for the new methods.

Open Source developers typically use Git repositories to transparently store the source code of projects and contribute to the code of others. There are millions of repositories actively hosted on platforms such as GitHub. This presents an opportunity for sharing knowledge between related projects – the so-called digital siblings. Finding repositories similar to one's own can allow for better developer collaboration and knowledge transfer. However, due to the large volume of projects, manually locating digital siblings of a project can be difficult. Hence, this paper proposes a novel approach, based on the dependency structures of GitHub repositories, that allows for calculating inter-repository similarity and subsequently querying for similar projects. We aim to answer the research question: How can the dependency structures of GitHub repositories be leveraged to find their digital siblings? This research includes an empirical evaluation of various similarity metrics and clustering techniques for GitHub repositories. Our results show that dependency structures are a reliable characteristic for measuring similarity between projects. We also identify the specific metrics and clustering techniques as particularly efficient. Lastly, we propose and evaluate a composable similarity metric to allow our findings to be combined with the research of the other Research Project group members.

This paper aims to study the importance of considering the documentation side of GitHub repositories when assessing the similarity between two or more applications. Readme and Wiki files, along with Comments from the source files, are the dimensions proposed to be analyzed through our methodology and experiments. We propose a pipeline that first extracts text fragments from these dimensions and then applies Natural Language Processing techniques to further prepare our data for evaluation. To gather a similarity score, we first vectorize our processed data with TF-IDF and then use cosine distance to obtain the score. Combinations of the three dimensions, ranging from using only one dimension to using all of them, are considered throughout our study. Moreover, additional information has been extracted from the plain text, such as referenced URLs and License usage, the similarity of which was calculated using Jaccard distance. Two experiments were performed. The first one aims to observe the behavioral tendencies of our methodology applied to a small dataset, while the second one aims to validate our results. By evaluating them, we found sufficient data that supported our presented conclusion: documentation represents a valuable asset in gathering a pool of similar applications.

This study explores the feasibility of categorizing GitHub projects based on their interactions and activities, aiming to assist both researchers and practitioners in navigating the vast landscape of open-source software. Through experiments and analysis, key attributes contributing to project categorization are identified, paving the way for effective grouping of projects in terms of interactions and activities. Findings indicate distinct clusters among GitHub projects, highlighting the influence of interactions and activities on project categorization. The study underscores the importance of refining grouping algorithms and improving project categorization methods for future research. Future work could involve developing user-friendly tools to facilitate project discovery and exploring correlations between interaction related metrics and project development dynamics. Overall, this study contributes to advancing our understanding of project categorization on GitHub, facilitating more efficient knowledge sharing and collaboration within professional fields.

GitHub is an online platform that hosts millions of projects. Many of these projects have the same topic or share the same goal. Finding similar projects which can be used as role models, inspiration or examples can help developers meet their requirements faster and more efficiently. Previous studies have been successful in finding similar GitHub projects, but they do not share how well their proposed metrics indicate similarity.

Our research and analysis seek to find the contribution of source code identifiers to overall project similarity. We define project similarity and define each type of identifier we evaluate. After these steps, we extract the defined types of identifiers from a list of projects. From this list of projects, we use twenty projects as queries for our analysis. We then analyze all identifiers using techniques such as TF-IDF and LSA. Our findings are that combining all types of identifiers results in the highest chance of having the same topic when looking at the most similar project. We also find that splitting each identifier on its casing and combining all split identifiers results in the highest chance that the most similar project found is similar. We therefore see that source code identifiers are reasonably contributing to overall project similarities.

GitHub is the home of hundreds of millions of Open Source Software(OSS) repositories where users collaborate on projects and find inspiration for new ideas. Some of these projects have certain build configurations set up to make building, testing, and deploying the software more time-efficient and less error-prone. However, setting up the correct configurations usually requires a lot of time and a high level of knowledge. This paper aims to analyze the current practices for setting up build configurations like the Maven files and GitHub actions while clustering some of these practices based on the scope of the project. Thus, we provide useful information in terms of discovering similar projects based on the build configurations and discuss the feasibility of build configuration analysis. In summary, we provide a comprehensive analysis of project similarity based on Maven build configurations and workflow files, shedding light on the importance of build configurations for identifying similar projects, and laying the groundwork for future exploration in the realm of build configuration analysis.