Ovarian cancer histotype classification is challenging because of substantial morphological heterogeneity and subtle subtype-specific features. This work presents a controlled evaluation of high-resolution pathology representations and slide-level aggregation methods for automated classification of five ovarian cancer subtypes from whole-slide histopathology images. Precomputed CONCH patch embeddings were aggregated using mean pooling, max pooling, gated attention-based multiple-instance learning, and a max-pooling cascade. The models were evaluated on 513 non-TMA whole-slide images from UBC-OCEAN using leakage-controlled five-fold cross-validation, with each slide receiving exactly one out-of-fold prediction. Gated ABMIL achieved a balanced accuracy of 81.39% and a macro-F1 score of 81.96%. Max pooling produced the highest numerical performance, with a balanced accuracy of 81.44%, macro-F1 of 82.59% (95% CI: 78.84-86.26%), macro-AUROC of 96.99%, and macro-AUPRC of 90.79%. However, paired slide-level bootstrap comparisons found no statistically significant differences among the CONCH aggregation strategies after Holm correction. Compared with the EfficientNet-B0 thumbnail baseline, CONCH max pooling improved macro-F1 by 30.30 percentage points and balanced accuracy by 25.70 percentage points, with both paired bootstrap confidence intervals excluding zero. Attention weights enabled visualization of influential patches, although these regions were not independently validated by pathologists. The findings show that high-resolution pathology foundation-model representations support ovarian cancer subtyping, while greater aggregation complexity does not necessarily improve performance. External multi-institutional validation is required before clinical generalizability can be established.

Signature-based IDS is becoming more difficult due to the ever-increasing amount of encrypted network traffic. Hence, the demand for interpretable and lightweight intrusion detection techniques becomes essential. In this paper, we propose Graph-Temporal Adaptive Intrusion Detection-Non-Parametric (GTAID-NP), which is a novel graph-temporal framework that exploits degree-based graph structure, temporal periodicity extracted using Fast Fourier Transform, and non-parametric log-odds estimation. We evaluate our proposed model on the BCCC-DarkNet-2025 benchmark dataset that contains 22,795 flow records, among which 6,317 are encrypted and 16,478 are not encrypted, which are expressed by 427 features. An extensive ablation study on 192 experiment setups was done to analyze the influence of bin granularity, total-variation smoothing, feature selection strategy, top-K feature selection threshold, graph augmentation, temporal periodicity extraction, and leakage-guard correction based on an 80:20 stratified train-test split. Among all experimental setups, the best setup achieves an AUC of 0.878. The results suggest that fine granular binning, weak smoothing, and joint consideration of graph-structure and temporal properties contribute to better detection performance. We also conduct robustness evaluation of GTAID-NP through five different random seed runs, which achieve a mean AUC of 0.9074 ± 0.0036, showing the stable performance of GTAID-NP on various train/test splits. While the ensemble methods outperform our model on the benchmark score, GTAID-NP can be regarded as a transparent and lightweight approach for intrusion detection on encrypted traffic. Future research works include adversarial robustness, federated inference, and adaptive online learning.

Industrial Control Systems (ICS) and Modbus-enabled networks are facing escalating threats from sophisticated cyber-attacks, while current Intrusion Detection Systems (IDS) struggle to identify intricate and adaptive attacks. This study envisions an ensemble learning-based IDS for Modbus-enabled industrial networks using a real-like Modbus 2023 dataset for industrial networks. The proposed IDS combines four base classifiers, namely K-Nearest Neighbors (KNN), Support Vector Machine (SVM), Random Forest (RF), and Adaptive Boosting (AdaBoost), using the stack ensemble framework, where Logistic Regression acts as the meta-classifier. Preprocessing involved PCAP capture and attack log synchronization, feature normalization, and one-hot encoding for balanced and accurate model training. Experimental evaluation demonstrated that the ensemble model has a 99.78% detection accuracy while outperforming the base individual models in terms of precision, recall, and F1-score. The results indicate the efficiency of ensemble learning for enhanced accuracy detection and false-positive reduction for Modbus networks. Future research will consider real-time testing, feature elimination, and explainable AI for higher operational deployment and scalability. 

Rapid progress in generative artificial intelligence (AI) technologies has brought forth stupendous challenges in differentiating AI-written text from human text. The Naturalness Score, a composite measure that considers lexical diversity, syntactic complexity, sentiment variability, and grammatical faults, is a new idea that emerged from this study. The Naturalness score is part of a larger machine learning framework, although it does have an individual classifier called the Naturalness-Based Logistic Regression Classifier or NLRC. The NLRC model was analyzed against a large, diverse corpus of nearly 45,000 text samples, most of which were student essays, articles, and web-scraped content. The proposed model outperformed all existing baseline models with an accuracy of 96.41%, precision of 0.98, recall of 0.95, and F1 score of 0.96. The high areas under the receiver operating characteristic curve (AUC=1.00) and precision-recall curve (AUC-PR) also indicate the effectiveness of the model in differentiating AI generated from human-written text. The proposed approach offers several advantages including increased detection accuracy, resilience against AI-generated content, cross-domain applicability, and interpretability. The research has implications for applying such models in schools, although it also calls for future research on the implications of the rapidly changing landscape of AI-generated content which it states. It emphasizes the importance of these findings in developing robust and adaptive detection systems to ensure the integrity of academic assessments, thereby preventing the misuse of AI tools.