July 25 ~ 26, 2026, Toronto, Canada
Abhradeep Chatterjee, NTT DATA Services, United States
Modern enterprise operations face a compounding failure surface created by microservices sprawl, hybrid cloud dependencies, and continuous delivery. Traditional AIOps pipelines detect anomalies but often stop short of trustworthy, auditable actions, leaving the highest-cost minutes of an incident—triage, correlation, and root-cause analysis—largely manual. This paper presents an agentic, closed-loop AIOps architecture that couples event intelligence with evidence-aware reasoning, policy-guarded action execution, and continuous learning from outcomes. The design unifies multi-source telemetry ingestion, causal-graph correlation, retrieval-augmented runbook planning, risk-scored remediation with human-in-the-loop controls, and SLO-governed feedback. We define an evaluation protocol spanning detection quality, diagnostic latency, action safety, and operator load, and provide a simulation harness to compare alerting, classic AIOps, and agentic AIOps. Simulation results across three scenarios show improved triage and mitigation speed while keeping unsafe actions near-zero via policy gating.
AIOps, agentic systems, incident management, root cause analysis, SLO governance.
Yusuf Abubakar Kutigi, Abdullahi Muhammad Bashir, Mohammed Abdulmalik Danlami and Adabara Nasiru Usman, 1University of Maiduguri, Maiduguri, Nigeria, 2,3,4Federal University of Technology Minna
The medical datasets are often confronted with the problems of class imbalance and redundancy of the features, which could affect the quality of classification prediction. In this study, the performance of four approaches to the Synthetic Minority Oversampling Technique (SMOTE)—SMOTE-ENN, Borderline SMOTE, ADASYN, and SMOTE-Tomek Links—has been studied together with feature selection and the XGBoost classifier in order to predict breast cancer and heart disease. The used datasets were processed to exclude noises, to make the distribution even, and to select the most important features. The analysis of the model has been conducted based on several criteria, such as accuracy, precision, recall, F1-score, and Cohens Kappa. For the heart disease data, the best results were obtained for the SMOTE-ENN approach, with accuracy, recall, and F1-score equal to 56.15%, 44.50%, and 27.46% correspondingly, that allowed detecting minority class cases. At the same time, for the breast cancer data, all other approaches provided better results, including accuracy, recall, F1-score, and Kappa of 96.49%, 100%, 97.30%, and 92.31% respectively.
SMOTE, feature selection, XGBoost, class imbalance, breast cancer, heart disease, machine learning, medical diagnosis.
Sri Sowmya Nemani, Independent Researcher, USA
Organizations continue to face increasingly sophisticated cyber threats that bypass traditional security controls. Security Operations Centres (SOCs) rely on detection engineering to identify malicious activity through logs, network telemetry, endpoint data, and authentication events. The MITRE ATT&CK framework provides a common language for understanding adversary behaviours and mapping security detections to real-world attack techniques. This paper explains how to build an ATT&CK-aligned detection program using practical case studies. The paper also discusses ATT&CK coverage, detection gaps, and applications across different industries. Results show that ATT&CK-aligned detections can improve threat visibility and security monitoring.
Detection Engineering, MITRE ATT&CK, Security Monitoring, SOC Operations, SIEM, Threat Detection, Cyber Defense, Security Analytics, Threat Hunting.
Hasan Abdulrahman, Northern Technical University, Iraq
Social-media steganography is a hostile channel problem rather than a conventional pixel editing problem. When a user uploads an im-age to Facebook, the platform may resize the image, recompress it as JPEG, alter chroma information, remove metadata, quantize colors, and apply proprietary optimization. These operations are not designed as attacks, yet they routinely erase payloads produced by fragile spatial and transform-domain steganographic schemes. This paper presents FB-StegNet, a deep neural framework for robust data hiding in color im-ages transmitted through Facebook like processing. The proposed sys-tem combines a dense binary message encoder, a residual CNN cover fea-ture extractor, a diflerentiable Facebook simulation layer, and a texture aware attention embedding network. The simulator is the main security oriented design element: it exposes the encoder and decoder during train-ing to randomized JPEG compression, downsampling upsampling, color quantization, and signal dependent noise. As a result, the learned embed-ding is optimized for recovery after a distribution of social-media trans-formations rather than for clean-image reconstruction alone. FBStegNet is formulated with a joint objective that balances bit recovery, visual dis-tortion, structural fidelity, and adaptive embedding energy. Additional, this work specifies a reproducible evaluation protocol using steganogra-phy image benchmarks, Facebook style simulated channels, real upload download tests, payload metrics, perceptual quality measures, and ste-ganalysis checks. FBStegNet therefore provides a practical and extensi-ble design for robust secret message communication through lossy social media image pipelines.
Information hiding Social media Deep learning Image steganography .
Ayawo Désiré Dandji1and Nadia Baaziz1 1Department of Computer Science and Engineering, University of Quebec in Outaouais (UQO), Gatineau (Quebec), Canada
The rapid growth of visual databases calls for efficient Content-Based Image Retrieval (CBIR). Texture descriptors are central to these systems; however, their performance often degrades under geometric image transformations, particularly rotation. This paper presents a CBIR framework designed to compare handcrafted and deep texture features for rotation-invariant retrieval. A hybrid approach combines Local Binary Patterns (LBP) with the Stationary Wavelet Transform (SWT) to extract compact, multi-scale descriptors robust to orientation variability. In parallel, a transfer learning strategy leverages intermediate layers of pre-trained convolutional neural networks (VGG16 and ResNet50) with multi-angle feature aggregation to extract rotation-robust deep descriptors. Experiments on benchmark texture datasets (Outex and Kylberg) show that the deep transfer-learning approach achieves higher recall at the cost of larger descriptor dimensionality and greater computational and memory demands, whereas the proposed hybrid descriptor provides a favorable trade-off between accuracy, compactness, and computational efficiency, making it well-suited for resource-constrained applications.
CBIR, handcrafted texture feature, rotation invariance, transfer learning.
Gavin Du1, Andrew Park2,1Sage Hill School, 20402 Newport Coast Dr, Newport Coast, CA 92657 2University of California, Irvine, Irvine, CA 92697
Post-traumatic stress disorder remains difficult to monitor because care often depends on self-reported symptoms rather than continuous physiological observation. To address this problem, this paper presents VitalLink, a mobile and hardware-integrated system designed to support veterans and other users affected by PTSD through real-time monitoring, caregiver alerts, and AI-assisted guidance [1]. The system combines a Flutter mobile application, Firebase backend services, and a Raspberry Pi-based hardware layer equipped with sensors for heart rate and motion tracking. Local processing and language-model-based interpretation helps transform raw sensor readings into useful insights while preserving responsiveness [2]. Key implementation challenges included sensor reliability, secure account permissions, and balancing AI capability with hardware limits.
PTSD, Veterans, Artificial Intelligence, Healthcare.