Mohan Timilsina

Yang Yang, Mohan Timilsina, Edward Curry. Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing. 2025.

Recently, there has been more interest in Decentralized Data-Sharing (DDS) because of the introduction of Dataspace 4.0. DDS is becoming increasingly popular as a safe, open, and effective way for many parties to data-sharing. Unlike conventional, centralized methods, DDS has several benefits, such as better knowledge exchange, higher accessibility and interoperability, and data privacy and security. The paper covers DDS’s advantages, including improved resilience, higher security, increased privacy, and improved interoperability. DDS gives people and organizations more control and ownership over their data while reducing the dangers of centralized data management. In this survey, we highlight promising technologies for DDS in Dataspace 4.0, including Federated Learning (FL), blockchain, decentralized file systems, semantic web and knowledge representation, and Peer-to-Peer (P2P) networks. We highlight the challenges, opportunities, and future directions of technology enabling further DDS advancement in Industry 4.0.

Almost every community Question-Answering (cQA) platform has the pressing need of enhancing user experience by presenting dedicated displays, connecting potential answerers with open questions and revitalizing the material in their archives. In doing so, it is crucial to understand the profile of their community members, especially as it relates to their demographics. In this realm, variables such as age and gender have shown to be particularly promising for managing content. For instance, they make it easier to connect questions posted by one generation that are more likely to be answered by individuals from the previous generation. This paper advances the current body of knowledge in this area by exploring the performance of nineteen frontier transformer-based models (e.g., BERT and ELECTRA) on age recognition across a large-scale collection of cQA members. In effect, the best encoder (LongFormer) finished with an accuracy of 78.61% (F1-Score of 0.7424) by taking full-questions and answers into account. Unlike gender recognition, our outcomes do not show a noticeable difference between cased and uncased models. But on the other hand, they confirm that the transition from one age group to the other is smooth, and thus boundary individuals pose a tough challenge to discriminant models built on top of frontier machine learning approaches.

The increasing importance of audio-based healthcare diagnostics, particularly in chronic respiratory problems, has prompted the development of novel approaches for disease classification using the full spectrum analysis of cough sounds. This paper presents an innovative study exploring the potential of employing supervised and semi-supervised learning methodologies for disease categorization based on cough audio samples. Specifically, this study focuses on scenarios characterized by a shortage of annotated data related to chronic diseases. The objectives of our study involved the utilization of standard machine learning algorithms for direct classification based on embeddings, as well as the integration of Graph Neural Networks (GNNs) on the KNN graph. Preliminary results indicate that GNN models consistently outperformed traditional classifiers. For instance, with just 1 % of the data, GAT and GCN achieved AUC PR values of 0.84 and 0.87, respectively, surpassing all traditional methods. The superiority mentioned above was sustained even when the data fraction was augmented to 3% and 5%. The usefulness of graph neural networks (GNNs) was further supported by comprehensive performance measures, wherein the graph convolutional network (GCN) exhibited exceptional preci-sion and PR (precision-recall). In summary, the AudioVecDiagnosis framework presents a promising opportunity for further investigation in audio-based healthcare diagnostics. It provides an optimum approach for situations with a limited availability of labeled data.

PURPOSE: Stratifying patients with cancer according to risk of relapse can personalize their care. In this work, we provide an answer to the following research question: How to use machine learning to estimate probability of relapse in patients with early-stage non–small-cell lung cancer (NSCLC)? MATERIALS AND METHODS: For predicting relapse in 1,387 patients with early-stage (I-II) NSCLC from the Spanish Lung Cancer Group data (average age 65.7 years, female 24.8%, male 75.2%), we train tabular and graph machine learning models. We generate automatic explanations for the predictions of such models. For models trained on tabular data, we adopt SHapley Additive exPlanations local explanations to gauge how each patient feature contributes to the predicted outcome. We explain graph machine learning predictions with an example-based method that highlights influential past patients. RESULTS: Machine learning models trained on tabular data exhibit a 76% accuracy for the random forest model at predicting relapse evaluated with a 10-fold cross-validation (the model was trained 10 times with different independent sets of patients in test, train, and validation sets, and the reported metrics are averaged over these 10 test sets). Graph machine learning reaches 68% accuracy over a held-out test set of 200 patients, calibrated on a held-out set of 100 patients. CONCLUSION: Our results show that machine learning models trained on tabular and graph data can enable objective, personalized, and reproducible prediction of relapse and, therefore, disease outcome in patients with early-stage NSCLC. With further prospective and multisite validation, and additional radiological and molecular data, this prognostic model could potentially serve as a predictive decision support tool for deciding the use of adjuvant treatments in early-stage lung cancer.