HARL enables agents to execute cooperative tasks by adopting agent-specific policies. Most of existing HARL methods use individual policy neural networks to ensure monotonic improvement, which leads to substantial computational overhead. The proposed SDE-HARL overcomes this limitation by decomposing each agent's policy neural network into a lightweight local neural network and a global neural network executed at an edge server. Each local neural network generates and sends a compressed latent representation to the edge server, which aggregates the representations and produces agent-specific inferences. As such, SDE-HARL allows to significantly save computing and networking resources while preserving agent-specific behavior. A key feature of SDE-HARL is grouping agents with similar roles via a role-aware mechanism and share partial parameters in their global networks, while an identity-aware mechanism is introduced to promote behavioral diversity among agents within the same group. We prototyped SDE-HARL on an experimental testbed composed of a Jetson Nano and Raspberry PI to measure latency and network resource consumption. We evaluated SDE-HARL's performance on several benchmark datasets, including Google Research Football and StarCraft II. Experimental results show that SDE-HARL reaches up to 90% win rate while reducing latency, energy consumption, and networking overhead respectively by 2x, 2.5x, and 5x compared to existing work.

The deployment of pre-trained perception models in novel environments often leads to performance degradation due to distributional shifts. Although recent artificial intelligence approaches for metacognition use logical rules to characterize and filter model errors, improving precision often comes at the cost of reduced recall. This paper addresses the hypothesis that leveraging multiple pre-trained models can mitigate this recall reduction. We formulate the challenge of identifying and managing conflicting predictions from various models as a consistency-based abduction problem, building on the idea of abductive learning (ABL) but applying it to test-time instead of training. The input predictions and the learned error detection rules derived from each model are encoded in a logic program. We then seek an abductive explanation—a subset of model predictions—that maximizes prediction coverage while ensuring the rate of logical inconsistencies (derived from domain constraints) remains below a specified threshold. We propose two algorithms for this knowledge representation task: an exact method based on Integer Programming (IP) and an efficient Heuristic Search (HS). Through extensive experiments on a simulated aerial imagery dataset featuring controlled, complex distributional shifts, we demonstrate that our abduction-based framework outperforms individual models and standard ensemble baselines, achieving, for instance, average relative improvements of approximately 13.6% in F1-score and 16.6% in accuracy across 15 diverse test datasets when compared to the best individual model. Our results validate the use of consistency-based abduction as an effective mechanism to robustly integrate knowledge from multiple imperfect models in challenging, novel scenarios.

Hyperdimensional computing (HDC) suits memory, energy, and reliability-constrained systems, yet the standard "one prototype per class" design requires $O(CD)$ memory (with $C$ classes and dimensionality $D$). Prior compaction reduces $D$ (feature axis), improving storage/compute but weakening robustness. We introduce LogHD, a logarithmic class-axis reduction that replaces the $C$ per-class prototypes with $n\!\approx\!\lceil\log_k C\rceil$ bundle hypervectors (alphabet size $k$) and decodes in an $n$-dimensional activation space, cutting memory to $O(D\log_k C)$ while preserving $D$. LogHD uses a capacity-aware codebook and profile-based decoding, and composes with feature-axis sparsification. Across datasets and injected bit flips, LogHD attains competitive accuracy with smaller models and higher resilience at matched memory. Under equal memory, it sustains target accuracy at roughly $2.5$-$3.0\times$ higher bit-flip rates than feature-axis compression; an ASIC instantiation delivers $498\times$ energy efficiency and $62.6\times$ speedup over an AMD Ryzen 9 9950X and $24.3\times$/$6.58\times$ over an NVIDIA RTX 4090, and is $4.06\times$ more energy-efficient and $2.19\times$ faster than a feature-axis HDC ASIC baseline.

Spectrum sharing is a key enabler for 5G/6G. We model decentralized spectrum access as a noncooperative, stochastic, imperfect-information extensive-form game and show that running standard Counterfactual Regret Minimization (CFR) independently at each user can exhibitpersistent cyclingrather than converging to a stable equilibrium. We provide a constructive multi-player example and a mapping analysis explaining why the induced regret-matching update need not be contractive in general multi-player, non-zero-sum settings. To address this, we propose CFR-M2, a lightweightregret-mixingscheme that periodically aggregates a small amount ofcumulative regret(not policy parameters) across users at a configurable communication interval. Our analysis shows that inserting an infrequent regret-consensus step over any connected communication graph preserves no-(counterfactual)-regret whilecontracting inter-agent disagreement in regrets. Consequently, the C´esaro-averaged play converges to an ε–extensive-form coarse correlated equilibrium (EFCCE) with ε=Õ(1/√T)+O(RMAXTCOMM/(1-ρ)T) where ρ is the second-largest eigenvalue modulus of the mixing matrix. Under additional structure (e.g., a strongly monotone pseudo-gradient or a strongly convex potential), the EFCCE collapses to a unique Nash equilibrium. Experiments on synthetic networks and a 5G Dynamic Spectrum Sharing (DSS) scenario (WINNER II channel model and practical parameter settings) show that CFR-M2improves convergence stability and system reward over CFR and several learning/game-theoretic baselines, while requiring onlyinfrequentcommunications.

The increasing demand for robust security solutions across various industries has made Video Anomaly Detection (VAD) a critical task in applications such as intelligent surveillance, evidence investigation, and violence detection. Traditional approaches to VAD often rely on finetuning large pre-trained models, which can be computationally expensive and impractical for real-time or resource-constrained environments. To address this, MissionGNN introduced a more efficient method by training a graph neural network (GNN) using a fixed knowledge graph (KG) derived from large language models (LLMs) like GPT-4. While this approach demonstrated significant efficiency in computational power and memory, it faces limitations in dynamic environments where frequent updates to the KG are necessary due to evolving behavior trends and shifting data patterns. These updates typically require cloud-based computation, posing challenges for edge computing applications. In this paper, we propose a novel framework that facilitates continuous KG adaptation directly on edge devices, overcoming the limitations of cloud dependency. Our method dynamically modifies the KG through a three-phase process: pruning, alternating, and creating nodes, enabling real-time adaptation to changing data trends. This continuous learning approach enhances the robustness of anomaly detection models, making them more suitable for deployment in dynamic and resource-constrained environments.

We address the problem of video question answering (video QA) with temporal grounding in a weakly supervised setup, without any temporal annotations. Given a video and a question, we generate an open-ended answer grounded with the start and end time. For this task, we propose TOGA: a vision-language model for Temporally Grounded Open-Ended Video QA with Weak Supervision. We instruct-tune TOGA to jointly generate the answer and the temporal grounding. We operate in a weakly supervised setup where the temporal grounding annotations are not available. We generate pseudo labels for temporal grounding and ensure the validity of these labels by imposing a consistency constraint between the question of a grounding response and the response generated by a question referring to the same temporal segment. We notice that jointly generating the answers with the grounding improves performance on question answering as well as grounding. We evaluate TOGA on grounded QA and open-ended QA tasks. For grounded QA, we consider the NExT-GQA benchmark which is designed to evaluate weakly supervised grounded question answering. For open-ended QA, we consider the MSVD-QA and ActivityNet-QA benchmarks. We achieve state-of-the-art performance for both tasks on these benchmarks.

Vision Transformers (ViTs) have emerged as the backbone of many segmentation models, consistently achieving state-of-the-art (SOTA) performance. However, their success comes at a significant computational cost. Image token pruning is one of the most effective strategies to address this complexity. However, previous approaches fall short when applied to more complex task-oriented segmentation (TOS), where the class of each image patch is not predefined but dependent on the specific input task. This work introduces the Vision Language Guided Token Pruning (VLTP), a novel token pruning mechanism that can accelerate ViT-based segmentation models, particularly for TOS guided by multi-modal large language model (MLLM). We argue that ViT does not need to process every image token through all of its layers—only the tokens related to reasoning tasks are necessary. We design a new pruning decoder to take both image tokens and vision-language guidance as input to predict the relevance of each image token to the task. Only image tokens with high relevance are passed to deeper layers of the ViT. Experiments show that the VLTP framework reduces the computational costs of ViT by approximately 25% without performance degradation and by around 40% with only a 1% performance drop. The code associated with this study can be found at this URL.

We address the problem of detecting out-of-context (OOC) objects in a scene. Given an image, we aim to detect whether the image has objects that are not present in their usual context and localize such OOC objects. Existing approaches for OOC detection rely on defining the common context in terms of the manually constructed features, such as the co-occurrence of objects, spatial relations between objects, and shape and size of the objects, and then learning such context for a given dataset. But context is often nu-anced ranging from very common to very surprising. Further, learned context from specific datasets may not be generalized as datasets may not truly represent the human notion of what is in context. Motivated by the success of large language models and more generally, foundation models (FMs) in common sense reasoning, we investigate the FM's ability to capture a more generalized notion of context. We find that a pre-trained FM, such as GPT-4, provides a more nuanced notion of OOC and enables zero-shot OOC detection when coupled with other pre-trained FMs for caption generation such as BLIP-2, and image in-painting with Sta-ble Diffusion 2.0. Our approach does not need any dataset-specific training. We demonstrate the efficacy of our approach on two OOC object detection datasets, achieving 90.8% zero-shot accuracy on the MIT-OOC dataset and 87.26% on the IJCAI22-COCO-OOC dataset.

Current data-driven AI approaches in Network Intrusion Detection System (NIDS) face challenges related to high resource consumption, high computational demands, and limited interpretability. Moreover, they often struggle to detect unknown and rapidly evolving cyber threats. This survey explores the integration of Neurosymbolic AI (NeSy AI) into NIDS, combining the data-driven capabilities of Deep Learning (DL) with the structured reasoning of symbolic AI to address emerging cybersecurity threats. The integration of NeSy AI into NIDS demonstrates significant improvements in both the detection and interpretation of complex network threats by exploiting the advanced pattern recognition typical of neural processing and the interpretive capabilities of symbolic reasoning. In this survey, we categorise the analysed NeSy AI approaches applied to NIDS into logic-based and graph-based representations. Logic-based approaches emphasise symbolic reasoning and rule-based inference. On the other hand, graph-based representations capture the relational and structural aspects of network traffic. We examine various NeSy systems applied to NIDS, highlighting their potential and main challenges. Furthermore, we discuss the most relevant issues in the field of NIDS and the contribution NeSy can offer. We present a comparison between the main XAI techniques applied to NIDS in the literature and the increased explainability offered by NeSy systems.

Machine learning (ML) and deep learning (DL) advancements have greatly enhanced anomaly detection of network intrusion detection systems (NIDS) by empowering them to analyze big data and extract patterns. ML/DL-based NIDS are trained using either flow-based or packet-based features. Flow-based NIDS are suitable for offline traffic analysis, while packet-based NIDS can analyze traffic and detect attacks in real-time. Current packet-based approaches analyze packets independently, overlooking the sequential nature of network communication. This results in biased models that exhibit increased false negatives and positives. Additionally, most literature-proposed packet-based NIDS capture only payload data, neglecting crucial information from packet headers. This oversight can impair the ability to identify header-level attacks, such as denial-of-service attacks. To address these limitations, we propose a novel artificial intelligence-enabled methodological framework for packet-based NIDS that effectively analyzes header and payload data and considers temporal connections among packets. Our framework transforms sequential packets into two-dimensional images. It then develops a convolutional neural network-based intrusion detection model to process these images and detect malicious activities. Through experiments using publicly available big datasets, we demonstrate that our framework is able to achieve high detection rates of 97.7% to 99% across different attack types and displays promising resilience against adversarial examples.

Network intrusion detection systems (NIDS) utilize signature and anomaly-based methods to detect malicious activities within networks. Advances in machine learning (ML) and deep learning (DL) algorithms have enabled NIDS to analyze large volumes of data and identify complex patterns. However, traditional ML/DL approaches in NIDS have primarily relied on flow-based features and utilized flat data formats, such as vectors or grids, which limit their ability to recognize the structural and contextual nuances of network attacks, particularly in real-time. Additionally, most NIDS depend on supervised or semi-supervised learning, requiring extensive labeled data that is timeconsuming to generate and not always feasible. This reliance restricts their ability to detect novel attacks, as they typically only recognize threats similar to those encountered during training. Hence, there is a significant need to develop NIDS that can operate in near real-time, eliminate the need for labeled data, and effectively identify novel attack patterns. We propose GTAE-IDS, a novel unsupervised packet-based graph neural network framework aimed at early and precise anomaly detection in network traffic. GTAE-IDS employs graph embeddings to capture and process network traffic data swiftly, creating sequential packet-based graphs that reflect network communications. Our approach employs graph autoencoders to identify structural and global patterns in benign data without needing labeled graph data, enhancing detection capabilities against novel attacks. Incorporating transformers in the encoder segment, GTAE-IDS effectively discerns contextual patterns in network traffic, achieving over 98% accuracy in identifying malicious activities on benchmark network intrusion data sets.

Recent advancements in artificial intelligence (AI) and machine learning (ML) algorithms, coupled with the availability of faster computing infrastructure, have enhanced the security posture of cybersecurity operations centers (defenders) through the development of ML-aided network intrusion detection systems (NIDS). Concurrently, the abilities of adversaries to evade security have also increased with the support of AI/ML models. Therefore, defenders need to proactively prepare for evasion attacks that exploit the detection mechanisms of NIDS. Recent studies have found that the perturbation of flow-based and packet-based features can deceive ML models, but these approaches have limitations. Perturbations made to the flow-based features are difficult to reverse-engineer, while samples generated with perturbations to the packet-based features are not playable. Our methodological framework, Deep PackGen, employs deep reinforcement learning to generate adversarial packets and aims to overcome the limitations of approaches in the literature. By taking raw malicious network packets as inputs and systematically making perturbations on them, Deep PackGen camouflages them as benign packets while still maintaining their functionality. In our experiments, using publicly available data, Deep PackGen achieved an average adversarial success rate of 66.4% against various ML models and across different attack types. Our investigation also revealed that more than 45% of the successful adversarial samples were out-of-distribution packets that evaded the decision boundaries of the classifiers. The knowledge gained from our study on the adversary’s ability to make specific evasive perturbations to different types of malicious packets can help defenders enhance the robustness of their NIDS against evolving adversarial attacks.

Resilience engineering is the ability to design, build, and sustain systems that can deal effectively with disruptive events. Previous research focused on resilience models that were not designed to predict multiple disruptions and recoveries, and resilience metrics, which are typically calculated after disruptions. Therefore, this article introduces a new approach combining regression and time series methods to track and predict system performance under multiple shocks, offering a framework for planning resilience tests and guiding data collection applicable to various systems and processes. To illustrate, subsets ranging from 50% to 80% of a historical job loss dataset from the 1980 U.S. recession were used for model fitting to assess generalization and stability. Goodness-of-fit measures, confidence intervals, and resilience metrics validated this approach against established statistical methods and a neural network model. The results indicate that traditional statistical models fail to capture minor changes when fitted with small datasets, and neural networks are overly sensitive to the size of the training data. In contrast, the novel mixture approach considering immediate and delayed disruptions exhibits superior long-term predictive performance and greater accuracy in forecasting resilience metrics, even when only 50% of the data is used for model fitting.

Ever-changing cyber threats require strong and flexible network security solutions. This paper suggests a new method to improve the performance of detecting both known and unknown attacks using a neuro-symbolic artificial intelligence (NSAI) network intrusion detection system (NIDS). Deep neural networks (DNN) learn complex network data patterns, which create a detailed overview of cyber-attack characteristics. Symbolic logic integration into the DNN allows for model training guidance by applying penalties when the DNN fails to differentiate between malicious and benign network traffic. This improves our model’s adaptability to new attacks and overcomes traditional signature-based NIDS limitations. By testing our NSAI NIDS on a large cyber dataset that includes novel attack scenarios, we show that it delivers an improvement in how accurately it detects attacks compared to traditional DNN methods. While our system maintains its high accuracy in recognizing known attacks, it outperforms conventional NIDS in discovering unknown attacks. This work improves cybersecurity by introducing a new way to detect both known and unknown network intrusions by combining DNNs with symbolic logic.

With the evolution of the Internet of Things, network intrusion detection systems (NIDS) are vital for protecting networks by monitoring and analyzing network traffic to detect potential cyber threats. Deep neural networks (DNNs) are widely used in NIDS for their accurate classification and response capabilities against threats. However, there is a lack of detailed discussion in the literature about evaluating DNN performance in real-time scenarios for monitoring and assurance of NIDS. This paper fills this gap by applying multiple linear regression models to predict the F1-score of a DNN attack classifier. The predictive models are evaluated using a pre-trained DNN on a NIDS benchmark dataset, where three different distance metrics computed between real-time instances and known attack patterns stored in historical data are considered as model covariates. Our findings show that the multiple linear regression model with interaction between covariates confidently forecasts the F1-score for future periods, demonstrating its ability to anticipate future observations with an empirical coverage of 94.4%.

Deep Reinforcement Learning (DRL) algorithms have achieved great success in solving many challenging tasks while their black-box nature hinders interpretability and real-world applicability, making it difficult for human experts to interpret and understand DRL policies. Existing works on interpretable reinforcement learning have shown promise in extracting decision tree (DT) based policies from DRL policies with most focus on the single-agent settings while prior attempts to introduce DT policies in multi-agent scenarios mainly focus on heuristic designs which do not provide any quantitative guarantees on the expected return. In this paper, we establish an upper bound on the return gap between the oracle expert policy and an optimal decision tree policy. This enables us to recast the DT extraction problem into a novel non-euclidean clustering problem over the local observation and action values space of each agent, with action values as cluster labels and the upper bound on the return gap as clustering loss. Both the algorithm and the upper bound are extended to multi-agent decentralized DT extractions by an iteratively-grow-DT procedure guided by an action-value function conditioned on the current DTs of other agents. Further, we propose the Return-Gap-Minimization Decision Tree (RGMDT) algorithm, which is a surprisingly simple design and is integrated with reinforcement learning through the utilization of a novel Regularized Information Maximization loss. Evaluations on tasks like D4RL show that RGMDT significantly outperforms heuristic DT-based baselines and can achieve nearly optimal returns under given DT complexity constraints (e.g., maximum number of DT nodes).

While there has been exciting recent progress in developing curricula for data education, more work is needed to establish connection points between data science, computer science, and other disciplines. This position paper argues for a broader, more all-encompassing perspective on data education to ensure opportunities are not missed. Our primary contribution is a comprehensive framework to visualize the data education landscape with the goal of improving understanding of how the various data education disciplines, work roles, core competencies, and skills fit together. Students and educators could benefit from such a framework, and all constituents of data education might better communicate requirements and more effectively make use of data and the data workforce.

Ensuring the security and integrity of computer and network systems is of utmost importance in today’s digital landscape. Network intrusion detection systems (NIDS) play a critical role in continuously monitoring network traffic and identifying unauthorized or potentially malicious activities that could compromise the confidentiality, availability, and integrity of these systems. However, traditional NIDS face a daunting challenge in effectively adapting to the evolving tactics of cyber attackers. To address this challenge, we propose a multistage artificial intelligence enabled framework for intrusion detection in network traffic, capable of handling zero-day, out-of-distribution, and adversarial evasion attacks. Our framework comprises three sequential deep neural network (DNN) architectures: one for the classifier and two for specific autoencoders, designed to effectively detect both known attack patterns and novel, previously unseen samples. We introduce an innovative transfer learning technique where specific combinations of neurons and layers in the DNN architectures are frozen during one-shot learning to enhance the framework’s robustness to novel attacks. To validate the effectiveness of our framework, we conducted extensive experimentation using publicly available benchmark intrusion detection data sets. Leveraging the one-shot learning approach in the transfer learning component of the framework, we demonstrate continuous improvement in detection accuracy for both known and novel network traffic patterns. The results demonstrate the effectiveness of the multiple stages in the framework by achieving, on average, 98.5% accuracy in detecting various attacks.

Cybersecurity at the edge requires fast computing in energy-constrained environments. Decision trees can provide an explainable solution for network intrusion detection with high detection accuracy at the packet level. However, their hardware implementation needs to support efficient real-time operation. In this paper, we propose a spatially distributed decision tree for network intrusion detection, using memristor-based chiplet leaves. Each chiplet processes an input by comparing it to a predefined boundary stored in the memristor cell and provides a binary output to select one of the interconnected leaves on the lower level, with an estimated power consumption in a 130nm node design of 389$\mu$W. The delay is 2.5$\mu$s for one inference decision. This chiplet approach is reconfigurable and in line with the natural architecture of decision trees. It also supports the prototyping with known good dies, overcoming the non-idealities challenge prevalent in memristor technologies. Our memristor-based decision trees show high intrusion detection accuracy of 82%, 84%, and 73% on the benchmark UNSW, CIC-IDS, and ACI-IoT datasets respectively, considering 6-bit device precision in one memristor vs. three memristor per boundary configurations. This distributed approach opens the way to utilizing memristor technology despite device defects for applications in need of local real-time computing.

Network intrusion detection systems (NIDSs) to detect malicious attacks continue to meet challenges. NIDS are often developed offline while they face auto-generated port scan infiltration attempts, resulting in a significant time lag from adversarial adaption to NIDS response. To address these challenges, we use hypergraphs (HGs) focused on Internet protocol (IP) addresses and destination ports to capture evolving patterns of port scan attacks. The derived set of HG-based metrics are then used to train an ensemble machine learning (ML)-based NIDS that allows for real-time adaption in monitoring and detecting port scanning activities, other types of attacks, and adversarial intrusions at high accuracy, precision and recall performances. This ML adapting NIDS was developed through the combination of 1) intrusion examples; 2) NIDS update rules; 3) attack threshold choices to trigger NIDS retraining requests; and 4) a production environment with no prior knowledge of the nature of network traffic. 40 scenarios were auto-generated to evaluate the ML ensemble NIDS comprising three tree-based models. The resulting ML ensemble NIDS was extended and evaluated with the CIC-IDS2017 dataset. Results show that under the model settings of an Update-ALL-NIDS rule (specifically retrain and update all the three models upon the same NIDS retraining request) the proposed ML ensemble NIDS evolved intelligently and produced the best results with nearly 100% detection performance throughout the simulation.

Open Set Recognition (OSR) is the ability of a machine learning (ML) algorithm to classify the known and recognize the unknown. In other words, OSR enables novelty detection in classification algorithms. This broader approach is critical to detect new types of attacks, including zero-days, thereby improving the effectiveness and efficiency of various MLenabled mission-critical systems, such as cyber-physical, facial recognition, spam filtering, and cyber defense systems such as intrusion detection systems (IDS). In ML algorithms, like deep learning (DL) classifiers, hyperparameters control the learning process; their values affect other model parameters, such as weights and biases, which affect the performance of OSR algorithms. Moreover, OSR introduces additional parameters, making DL classifiers bigger and training them more computationally intensive. Determining the suitable set of hyperparameters and parameters is a computationally expensive task. Alternative OSR algorithms have demonstrated promising results on image datasets, but only limited studies have been performed in the context of IDS. This paper proposes OpenSetPerf, an empirical investigation of three prominent OSR algorithms using a current, real-world network intrusion detection systems (NIDS) benchmark dataset to discover the relationship between the DL-based OSR algorithm’s hyperparameter values and their performance. OpenSetperf evaluates these algorithms using quantitative studies with widely used ML performance evaluation metrics.

Traditional machine learning models used for network intrusion detection systems rely on vast amounts of network traffic data with expertly engineered features. The abundance of computational and expert resources at the enterprise level allow for the employment of such models; however, these resources quickly dwindle in edge network scenarios. As Internet of Battlefield Things (IoBT) networks become common place in tactical environments, there is a need for improved and distributed models trained without these enterprise resources. Transfer learning – which allows us to take information learned in one domain and apply it to another – provides one way to create and distribute these models towards the edge. Using neural networks, we demonstrate the feasibility of transfer learning for intrusion detection using only raw network traffic in computationally limited environments. Our results show that with a transferred one-dimensional convolutional neural network model combined with a retrained random forest model, we obtain over 96% accuracy with only 5000 training samples on edge devices with an edge training time of approximately 67 s.

Adapting modern approaches for network intrusion detection is becoming critical, given the rapid technological advancement and adversarial attack rates. Therefore, packet-based methods utilizing payload data are gaining much popularity due to their effectiveness in detecting certain attacks. However, packet-based approaches suffer from a lack of standardization, resulting in incomparability and reproducibility issues. Unlike flow-based datasets, no standard labeled dataset exists, forcing researchers to follow bespoke labeling pipelines for individual approaches. Without a standardized baseline, proposed approaches cannot be compared and evaluated with each other. One cannot gauge whether the proposed approach is a methodological advancement or is just being benefited from the proprietary interpretation of the dataset. Addressing comparability and reproducibility issues, we introduce Payload-Byte, an open-source tool for extracting and labeling network packets in this work. Payload-Byte utilizes metadata information and labels raw traffic captures of modern intrusion detection datasets in a generalized manner. Moreover, we transformed the labeled data into a byte-wise feature vector that can be utilized for training machine learning models. The whole cycle of processing and labeling is explicitly stated in this work. Furthermore, source code and processed data are made publicly available so that it may act as a standardized baseline for future research work. Lastly, we present a brief comparative analysis of machine learning models trained on packet-based and flow-based data.

Methods for sequential decision-making are often built upon a foundational assumption that the underlying decision process is stationary. This limits the application of such methods because real-world problems are often subject to changes due to external factors (\textit{passive} non-stationarity), changes induced by interactions with the system itself (\textit{active} non-stationarity), or both (\textit{hybrid} non-stationarity). In this work, we take the first steps towards the fundamental challenge of on-policy and off-policy evaluation amidst structured changes due to active, passive, or hybrid non-stationarity. Towards this goal, we make a \textit{higher-order stationarity} assumption such that non-stationarity results in changes over time, but the way changes happen is fixed. We propose, OPEN, an algorithm that uses a double application of counterfactual reasoning and a novel importance-weighted instrument-variable regression to obtain both a lower bias and a lower variance estimate of the structure in the changes of a policy's past performances. Finally, we show promising results on how OPEN can be used to predict future performances for several domains inspired by real-world applications that exhibit non-stationarity.

Adversarial machine learning, a technique which seeks to deceive machine learning (ML) models, threatens the utility and reliability of ML systems. This is particularly relevant in critical ML implementations such as those found in Network Intrusion Detection Systems (NIDS). This paper considers the impact of adversarial influence on NIDS and proposes ways to improve ML based systems. Specifically, we consider five feature robustness metrics to determine which features in a model are most vulnerable, and four defense methods. These methods are tested on six ML models with four adversarial sample generation techniques. Our results show that across different models and adversarial generation techniques, there is limited consistency in vulnerable features or in effectiveness of defense method.

Feature engineering and selection is a critical step in the implementation of any machine learning system. In application areas such as intrusion detection for cybersecurity, this task is made more complicated by the diverse data types and ranges presented in both raw data packets and derived data fields. Additionally, the time and context specific nature of the data requires domain expertise to properly engineer the features while minimizing any potential information loss. Many previous efforts in this area naively apply techniques for feature engineering that are successful in image recognition applications. In this work, we use network packet dataflows from the Defense Research and Engineering Network (DREN) and the Engineer Research and Development Center's (ERDC) high performance computing systems to experimentally analyze various methods of feature engineering. The results of this research provide insight on the suitability of the features for machine learning based cybersecurity applications.