Evaluating the quality of published medical research with ChatGPT

Evaluating the quality of published research is time-consuming but important for departmental evaluations, appointments, and promotions. Previous research has shown that ChatGPT can score articles for research quality, with the results correlating positively with an indicator of quality in all fields except Clinical Medicine. This article investigates this anomaly with the largest dataset yet and a more detailed analysis. The results showed that ChatGPT 4o-mini scores for articles submitted to the UK's Research Excellence Framework (REF) 2021 Unit of Assessment (UoA) 1 Clinical Medicine correlated positively (r=0.134, n=9872) with departmental mean REF scores, against a theoretical maximum correlation of r=0.226 (due to the departmental averaging involved). At the departmental level, mean ChatGPT scores correlated more strongly with departmental mean REF scores (r=0.395, n=31). For the 100 journals with the most articles in UoA 1, their mean ChatGPT score correlated strongly with their REF score (r=0.495) but negatively with their citation rate (r=-0.148). Journal and departmental anomalies in these results point to ChatGPT being ineffective at assessing the quality of research in prestigious medical journals or research directly affecting human health, or both. Nevertheless, the results give evidence of ChatGPT's ability to assess research quality overall for Clinical Medicine, so now there is evidence of its ability in all academic fields.

Towards the generation of hierarchical attack models from cybersecurity vulnerabilities using language models

This paper investigates the use of a pre-trained language model and siamese network to discern sibling relationships between text-based cybersecurity vulnerability data. The ultimate purpose of the approach presented in this paper is towards the construction of hierarchical attack models based on a set of text descriptions characterising potential/observed vulnerabilities in a given system. Due to the nature of the data, and the uncertainty sensitive environment in which the problem is presented, a practically oriented soft computing approach is necessary. Therefore, a key focus of this work is to investigate practical questions surrounding the reliability of predicted links towards the construction of such models, to which end conceptual and practical challenges and solutions associated with the proposed approach are outlined, such as dataset complexity and stability of predictions. Accordingly, the contributions of this paper focus on producing neural networks using a pre-trained language model for predicting sibling relationships between cybersecurity vulnerabilities, then outlining how to apply this capability towards the generation of hierarchical attack models. In addition, two data sampling mechanisms for tackling data complexity, and a consensus mechanism for reducing the amount of false positive predictions are outlined. Each of these approaches is compared and contrasted using empirical results from three sets of cybersecurity data to determine their effectiveness.

Take History as a Mirror in Heterogeneous Federated Learning

Federated Learning (FL) allows several clients to cooperatively train machine learning models without disclosing the raw data. In practice, due to the system and statistical heterogeneity among devices, synchronous FL often encounters the straggler effect. In contrast, asynchronous FL can mitigate this problem, making it suitable for scenarios involving numerous participants. However, Non-IID data and stale models present significant challenges to asynchronous FL, as they would diminish the practicality of the global model and even lead to training failures. In this work, we propose a novel asynchronous FL framework called Federated Historical Learning (FedHist), which effectively addresses the challenges posed by both Non-IID data and gradient staleness. FedHist enhances the stability of local gradients by performing weighted fusion with historical global gradients cached on the server. Relying on hindsight, it assigns aggregation weights to each participant in a multi-dimensional manner during each communication round. To further enhance the efficiency and stability of the training process, we introduce an intelligent $\ell_2$-norm amplification scheme, which dynamically regulates the learning progress based on the $\ell_2$-norms of the submitted gradients. Extensive experiments demonstrate that FedHist outperforms state-of-the-art methods in terms of convergence performance and test accuracy.