Building a national workforce development model for artificial intelligence.

Generative Adversarial Networks (GANs) are a type of generative modeling that utilizes deep learning techniques to generate highly realistic synthetic data. Since their introduction in 2014, GANs have gained significant attention across various fields for their ability to generate synthetic data, which can be utilized for training machine learning models. Although GANs have demonstrated extensive efficacy, they still face several constraints, one of which is the issue of mode collapse. Mode collapse occurs in GANs when the generator model fails to produce a diverse range of outputs that accurately capture every aspect of diversity present in the real data distribution, which reduces the efficacy of the synthetic data. This research aims to provide an overview of the mode collapse issue in GANs. First, we investigate the causes of mode collapse. Second, we provide an overview of the progression and status of the mode collapse issue across various GAN variants over the years. Third, we identify the gaps and shortcomings in existing mode collapse mitigation approaches. Finally, we present some potential research directions to effectively handle the issue.

Real-time anomaly detection in large, dynamic graph networks is crucial for real-world applications such as network intrusion prevention, fraud transaction identification, fake news detection in social networks, and uncovering abnormal communication patterns. However, existing graph-based methods often focus on static graph structures, which struggle to adapt to the evolving nature of these graphs. In this paper, we propose Adaptive-DecayRank, a real-time and adaptive anomaly detection model for dynamic graph streams. Our method extends the dynamic PageRank algorithm by incorporating an adaptive Bayesian updating mechanism, allowing nodes to dynamically adjust their decay factors based on observed graph changes. This enables real-time detection of sudden structural shifts, improving anomaly identification in streaming graphs. We evaluate Adaptive-DecayRank on multiple real-world security datasets, including DARPA and CTU-13, as well as synthetic dense graphs generated using RTM. Our experiments demonstrate that Adaptive-DecayRank outperforms state-of-the-art methods, such as AnomRank, Sedanspot, and DynAnom, achieving up to 13.94% higher precision, 8.43% higher AUC, and more robust detection in highly dynamic environments.

The ability to detect fraudulent activities such as denial-ofservice attacks (DoS), social media impersonations, or fake recommendations in e-commerce networks is a difficult task. One approach to aiding in detecting these types of anomalous activities is to represent the data as a graph. In general, given a stream of graph edges, where each edge indicates a communication/link in a specific time, the objective is to detect abnormal edges that represent actions such as DoS attacks, port scans, etc., and do so in an online manner while consuming constant time and memory. In this paper, we propose Term Frequency-Inverse Graph Frequency (TF-IGF) as an online anomaly detection approach, which assigns anomaly scores to edge streams. We further analyze the performance of the TF-IGF technique in detecting various types of cyber attacks by using real-world streaming graph datasets and benchmarking against the existing edge streaming models. Performance is compared based on 1. the accuracy in classifying anomalies, 2. the experiment run-time, and 3. the type of anomalies detected. The results show the strengths of TF-IGF in detecting various cyber attacks in streaming graph networks. The paper concludes with recommendations for future work and provides valuable insights for designing and deploying efficient and effective anomaly detection systems in streaming graph networks.

Anomaly detection in dynamic graphs is essential for identifying malicious activities, fraud, and unexpected behaviors in real-world systems such as cybersecurity and power grids. However, existing approaches struggle with scalability, probabilistic interpretability, and adaptability to evolving traffic patterns. In this paper, we propose ADAPTIVE-GRAPHSKETCH, a lightweight and scalable framework for real-time anomaly detection in streaming edge data. Our method integrates temporal multi-tensor sketching with Count-Min Sketch using Conservative Update (CMS-CU) to compactly track edge frequency patterns with bounded memory, while mitigating hash collision issues. We incorporate Bayesian inference for probabilistic anomaly scoring and apply Exponentially Weighted Moving Average (EWMA) for adaptive thresholding tuned to burst intensity. Extensive experiments on four real-world intrusion detection datasets demonstrate that Adaptive-GraphSketch outperforms state-of-the-art baselines such as AnoEdge-G/L, MIDAS-R, and F-FADE, achieving up to 6.5% AUC gain on CIC-IDS2018 and up to 15.6% on CIC-DDoS2019, while processing 20 million edges in under 3.4 seconds using only 10 hash functions. Our results show that Adaptive-GraphSketch is practical and effective for fast, accurate anomaly detection in large-scale streaming graphs.

This special issue encapsulates the multifaceted landscape of contemporary challenges and innovations in Artificial Intelligence (AI) and Machine Learning (ML), with a particular focus on issues related to explainability, fairness, and trustworthiness. The exploration begins with the computational intricacies of understanding and explaining the behavior of binary neurons within neural networks. Simultaneously, ethical dimensions in AI are scrutinized, emphasizing the nuanced considerations required in defining autonomous ethical agents. The pursuit of fairness is exemplified through frameworks and methodologies in machine learning, addressing biases and promoting trust, particularly in predictive policing systems. Human-agent interaction dynamics are elucidated, revealing the nuanced relationship between task allocation, performance, and user satisfaction. The imperative of interpretability in complex predictive models is highlighted, emphasizing a query-driven methodology. Lastly, in the context of trauma triage, the study underscores the delicate trade-off between model accuracy and practitioner-friendly interpretability, introducing innovative strategies to address biases and trust-related metrics.

The integration of renewable energy sources, such as rooftop solar panels, into smart grids poses significant challenges for managing customer-side battery storage. In response, this paper introduces a novel reinforcement learning (RL) approach aimed at optimizing the coordination of these batteries. Our approach utilizes a single-agent, multi-environment RL system designed to balance power saving, customer satisfaction, and fairness in power distribution. The RL agent dynamically allocates charging power while accounting for individual battery levels and grid constraints, employing an actor–critic algorithm. The actor determines the optimal charging power based on real-time conditions, while the critic iteratively refines the policy to enhance overall performance. The key advantages of our approach include: (1) Adaptive Power Allocation: The RL agent effectively reduces overall power consumption by optimizing grid power allocation, leading to more efficient energy use. (2) Enhanced Customer Satisfaction: By increasing the total available power from the grid, our approach significantly reduces instances of battery levels falling below the critical state of charge (SoC), thereby improving customer satisfaction. (3) Fair Power Distribution: Fairness improvements are notable, with the highest fair reward rising by 173.7% across different scenarios, demonstrating the effectiveness of our method in minimizing discrepancies in power distribution. (4) Improved Total Reward: The total reward also shows a significant increase, up by 94.1%, highlighting the efficiency of our RL-based approach. Experimental results using a real-world dataset confirm that our RL approach markedly improves fairness, power efficiency, and customer satisfaction, underscoring its potential for optimizing smart grid operations and energy management systems.

This innovative practice full paper describes a two-semester capstone experience that trains students in agile software engineering principles and incorporates the material into building an actual product for an industry partner. Recently, teaching agile software engineering has garnered considerable attention, and research has focused on effective pedagogical approaches, challenges, and outcomes. However, while computer science students are exposed to agile methodologies in their curriculum, and students even use the approach in a project, the experience tends to be brief and non-real-world. In this work, we believe our approach provides a more cohesive learning experience, better prepares students for jobs in industry, and introduces them to incorporating an overall agile mind set. We outline specific activities, timelines, and best practices for managing team projects and providing a better experience for the students. The results of our efforts are reported through retrospectives and reflections with the students over five years.

Given a large graph stream with dynamically changing structures over a given timestep, it is important to detect the sudden appearance of anomalous patterns, such as sudden spikes in IP-network attacks or unexpected surges in social media followers. In addition, it is important that we promptly identify these abrupt changes in the network by considering swift and short-term responses within the network structure. To design a model capable of adapting to dynamic changes, we introduce an approach that utilizes a modified dynamic "PageRank-with-Decay" as a node scoring function. This method enables the detection of sudden dynamic graph changes based on node importance scores, leveraging the temporal evolution of graph structures at each timestep. This approach provides a refined anomaly detection mechanism for tracking rapid structural changes in the network. Through experiments conducted on a real-world dataset, our model demonstrates faster and more accurate results (in terms of precision and recall) compared to state-of-the-art methods.

Reinforcement learning (RL) is proven effective in optimizing home battery charging coordination within smart grids. However, its vulnerability to adversarial behavior poses a significant challenge to the security and fairness of the charging process. In this study, we, first, craft five stealthy false data injection (FDI) attacks that under-report the state-of-charge (SoC) values to deceive the RL agent into prioritizing their charging requests, and then, we investigate the impact of these attacks on the charging coordination system. Our evaluations demonstrate that attackers can increase their chances of charging compared to honest consumers. As a result, honest consumers experience reduced charging levels for their batteries, leading to a degradation in the system’s performance in terms of fairness, consumer satisfaction, and overall reward. These negative effects become more severe as the amount of power allocated for charging decreases and as the number of attackers in the system increases. Since the total available power for charging is limited, some honest consumers with genuinely low SoC values are not selected, creating a significant disparity in battery charging levels between honest and malicious consumers. To counter this serious threat, we develop a deep learning-based FDI attack detector and evaluated it using a real-world dataset. Our experiments show that our detector can identify malicious consumers with high accuracy and low false alarm rates, effectively protecting the RL-based charging coordination system from FDI attacks and mitigating the negative impacts of these attacks.

This survey paper presents a comprehensive and conceptual overview of anomaly detection using dynamic graphs. We focus on existing graph-based anomaly detection (AD) techniques and their applications to dynamic networks. The contributions of this survey paper include the following: i) a comparative study of existing surveys on anomaly detection; ii) a Dynamic Graph-based Anomaly Detection (DGAD) review framework in which approaches for detecting anomalies in dynamic graphs are grouped based on traditional machine-learning models, matrix transformations, probabilistic approaches, and deep-learning approaches; iii) a discussion of graphically representing both discrete and dynamic networks; and iv) a discussion of the advantages of graph-based techniques for capturing the relational structure and complex interactions in dynamic graph data. Finally, this work identifies the potential challenges and future directions for detecting anomalies in dynamic networks. This DGAD survey approach aims to provide a valuable resource for researchers and practitioners by summarizing the strengths and limitations of each approach, highlighting current research trends, and identifying open challenges. In doing so, it can guide future research efforts and promote advancements in anomaly detection in dynamic graphs. Keywords: Graphs, Anomaly Detection, dynamic networks,Graph Neural Networks (GNN), Node anomaly, Graph mining.

Real-world graphs are complex to process for performing effective analysis, such as anomaly detection. However, recently, there have been several research efforts addressing the issues surrounding graph-based anomaly detection. In this paper, we discuss a comprehensive overview of anomaly detection techniques on graph data. We also discuss the various application domains which use those anomaly detection techniques. We present a new taxonomy that categorizes the different state-of-the-art anomaly detection methods based on assumptions and techniques. Within each category, we discuss the fundamental research ideas that have been done to improve anomaly detection. We further discuss the advantages and disadvantages of current anomaly detection techniques. Finally, we present potential future research directions in anomaly detection on graph-structured data.

Due to the rise in the Internet of Health Things (IoHT), cyber-attacks, particularly data intrusions, have become an issue for security experts. In this work, we analyze the performance of traditional statistical, machine learning, and graph-based anomaly detection approaches in response to this problem. We believe that understanding intrusion patterns can aid in the prevention of future attacks. In this work, we use the ARMA model for statistical analysis. We also use several machine learning approaches such as multinomial naive bayes, ran- dom forest, neural networks, XGBClassifier, and support vector machines (SVM). However, while our experiments show that machine learning (ML) techniques have higher precision, accuracy, and F1 score than graph-based techniques, there are aspects to a graph-based approach that could aid security experts in the discovery of certain data breaches by combining the graph-based with the statistical and ML methods. Experiments also show combining different anomaly detection techniques allows for a diverse set of intrusion patterns to be discovered. By recognizing the power of both machine learning and graph-based approaches, we analyze their precision and accuracy while explaining how existing state-of-the-art methods can detect breach patterns. Finally, by identifying the characteristics of breach patterns, we present information that security experts can use to prevent future data intrusions.

Automated decision-making that impacts human interests, rights, and lives, in particular different data mining and artificial intelligence-based techniques, have become an integral part of many high-stakes applications such as sentencing and bail decisions, credit approvals, hiring, and predictive policing. However, fairness concerns, such as discrimination based on race, age, sex, etc., primarily stemming from data and algorithmic bias, is one of the major and contemporary problems associated with automated decision-making. In a traditional Data Mining, Artificial Intelligence, or Machine Learning course, educators usually teach different automated decision-making techniques but largely with limited coverage on their ethical concerns such as fairness, transparency, and privacy. In this paper, we share our experience in building and incorporating a fairness module 1 in an existing undergraduate Data Mining course, within traditional content, and evaluate the outcome of the initiative. The module includes lectures and hands-on exercises, using state-of-the-art and open-source bias detection and mitigation software, on real-world datasets. The goal is to help instill the consciousness of fairness and bias at the very early stage of a potential future developer of automated decision-making software. The module is easily adaptable, and can be integrated into other relevant courses including introductory Artificial Intelligence and Machine Learning courses.

Many real-world networks, such as computer networks, social networks, and the Internet of Things (loT), can be represented by streaming (or dynamic) graphs. Analysis of these streaming graphs serves as the basis for classification, anomaly detection, community detection, clustering, and visual-ization tasks. This paper uses a Self-Organizing Map (SOM), an unsupervised learning model, to cluster and visualize streaming graphs. As a result, a SOM is used to visualize and interpret the anomaly detection technique on high-dimensional graph-structured data. For this, the SOM-based graph clustering and visualization technique is divided into two phases. In the first phase, we use various existing graph sketching techniques like StreamS pot, SpotLight, and SnapSketch to embed streaming graphs into sketched vectors. Later, in the second phase, we pass the sketched vector inputs into a SOM to cluster and visualize the normal and anomalous graph streams to interpret the anomaly detection technique. In addition, the SOM-based visualization also helps to estimate the quality of embedding (or sketching) techniques.

Data bias and algorithmic bias are the primary contributing factors for fairness-related risks in AI-based decision-making. The concept of fairness is comparatively new and is sometimes only discussed in detail in graduate-level courses such as Ethics of Artificial Intelligence, and Ethics and Governance of Artificial Intelligence. In addition, at the undergraduate level, a standalone course on fairness is not feasible due to the level of difficulty and the many other essential courses that need to be broached in a university's computer science curriculum. Therefore, instead of a standalone course, we have created a concise, high-level concept module on bias and fairness in automated decisions, with (1) lecture material, (2) demonstrations, and (3) assignments (i.e., exercises) on real-world datasets.

Reinforcement Learning (RL) has been around for some time now, and various issues have been associated with the provision.In the last two years, RL has realized a boost in popularity through deep learning.For instance, RL played a critical role in the DeepMind AlphaGo program that was crucial to beating a top-level Go player in 2016 [1].However, there is still work to be done despite RL's advancements before the DeepMind AlphaGo program becomes mainstream.For instance, one of the challenges of the program is its ability to multitask.[1] indicate that agents should perform various functions to achieve general AI.However, in the contemporary understanding, multitasking is one of the challenges of AI and RL scalability; for instance, it should not take over 1000 hrs. of different tasks to learn 1000 various tasks.Instead, AI agents must build up a library of general knowledge and learn general skills that are common and applicable across a variety of tasks.However, this ability is currently non-existent in such programs as the Deep Q-Network (DQN).While the DQN has previously been shown to have the ability to play various games, including Atari games, there is often no learning across the tasks.Each of the games is learned from scratch, which is often not scalable.

In other to designing an efficient and capable machine learning algorithm that will automate penetration testing, one might want to look at realtime strategy computer games, as both involve an entity that attacks another one, be it the penetration tester targeting the network or the computer targeting the player's population within the video game.Historically, artificial intelligence and machine learning in computer games were implemented through reinforcement learning protocols, and such protocols are also well suited for autonomous penetration learning algorithms.An added similarity between real-time strategy games and penetration testing are that the games operate with large state spaces and complex mixtures of options.Over time this has become even more intricate, as commonly available computing power has increased, and human players have also become more proficient.The combination of choices in strategy games is infinite, and this is one more factor that they have in common with computer networks; trying to enter a cybersecurity network allows for many different options that are not always easy to predict -an excellent base for the application of reinforcement learning.Another commonality between real-time strategy games and penetration testing is that the layout of both the network and the game are not entirely known beforehand.Penetration testers must employ reconnaissance to learn about network topology; the 'fog of war' in strategy games prevents the player from knowing the lay of the land before he has explored it.Thus, many arguments favor using computer strategy games -and simulations -as a framework for developing autonomous cybersecurity products, such as penetration tests.Moreover, despite their large number of options, real-time strategy games can be played using a handful of larger strategy types; this can inform the type of attack tree a penetration test chooses to utilize, given the configuration of vulnerabilities found during reconnaissance.

The ability to detect fraudulent activities such as denial-of-service attacks (DoS), social media impersonations, or fake recommendations in e-commerce networks is a difficult task. One approach to aiding with detection of these types of anomalous activities is through representing the data as a graph. However, the scalability of graph-based approaches is problematic. In general, given a stream of graph edges, where each edge indicates a communication/link in a certain time, the objective is to detect anomalous edges that represent actions such as DoS attacks, port scans, etc., and do so in an online manner while consuming constant time and memory. In this paper, we propose Term Frequency-Inverse Graph Frequency (TF-IGF) as an online anomaly detection approach, which assigns anomaly scores to edge streams. TF-IGF offers the following contributions: (1) streaming; processes one edge at a time, (2) online; detects suddenly appearing anomalous edges in near real-time, (3) accurate; has better accuracy than the state-of-the-art approaches, and (4) efficient; consumes constant time and memory to process edges. We evaluate this approach on both synthetic and real-world data sets.