Blind image quality assessment (BIQA) models heavily depend on expensive human subjective ratings. To reduce this dependency, full-reference IQA (FR-IQA) metrics are used as surrogates for generating pseudo-labels. However, FR-IQA-based methods introduce substantial label noise that degrades assessment accuracy. In this research, an adaptive transition matrix learning-based BIQA (ATML-BIQA) method is proposed that learns from synthetically degraded images and multiple annotators without requiring human labels or suffering from noisy label effects. The method operates in four stages: first, distorted images are synthesized from high-quality references and paired randomly. Eight FR-IQA models then assign pseudo-binary labels indicating relative perceptual quality. Second, a CNN-based BIQA model is trained on these binary labels to generate Bayesian optimal labels. Third, an adaptive instance transition matrix (AITM) is designed to model the relationship between noisy pseudo-labels and optimal labels, capturing the correlation between noisy annotations and latent ground truth for effective label noise correction. Finally, the BIQA model is retrained with corrected labels to produce final quality scores. This research significantly reduces the reliance of BIQA on expensive manual annotations and overcomes the accuracy limitations of traditional proxy labels by adaptively modeling annotation noise. Experimental results demonstrate that while enhancing model generalization, this method provides an efficient and robust new paradigm for quality assessment tasks in unsupervised or weakly supervised environments.

Few-shot knowledge graph completion (FKGC) aims to infer missing triples within long-tail relations by leveraging a limited number of reference instances. Existing FKGC models struggle to effectively distinguish informative neighbors from noisy ones during the aggregation of neighborhood information for central entities. Moreover, in the matching and prediction phase, they typically rely solely on entity pair similarity, which often leads to biased predictions when the reference triples are unevenly distributed. To address these challenges, MhAMM, a novel FKGC model, is proposed based on multi-head attention matching. In the neighborhood aggregation stage, MhAMM introduces a multi-head attention mechanism tailored to the sparsity characteristics of FKGC tasks, which effectively amplifies the attention weights of informative neighbors while suppressing the influence of noisy ones, thereby improving the encoding quality of central entities. In the matching stage, a multidimensional matching network is designed, which integrates both the entity pair similarity score and a triple plausibility score computed via a fully connected neural network. These two complementary scores jointly enhance the overall matching performance. Extensive experiments on public datasets demonstrate that MhAMM consistently achieves significant improvements across multiple evaluation metrics, verifying the effectiveness and robustness of the proposed model.

Modifying a code segment may give rise to a consistency issue when the code segment belongs to a clone group comprising closely similar code segments. Recent studies have demonstrated that such consistent changes can incur extra maintenance costs when clones are checked for consistency and introduce defects if developers forget to change clones consistently when needed. To address this problem, researchers have proposed an approach to predict clone consistency in advance with handcrafted attributes, notably using machine learning methods. Although these attributes can help predict clone consistency to some extent, the capability of such an approach is generally weak and unsatisfactory in practice. Such limitations in capability are especially severe at a project’s infancy stage when there is not sufficient within-project data to model clone consistency behavior, and cross-project data have not been helpful in supporting prediction. In this paper, we propose the Clone Hierarchical Attention Neural Network (CHANN) to represent code clones and their evolution by adopting a hierarchical perspective of code, context, and code evolution, and thus enhancing the effectiveness of clone consistency prediction. To assess the effectiveness of CHANN, we conduct experiments on the dataset collected from eight open-source projects. The experimental results show that CHANN is highly effective in predicting clone consistency, and the precision, recall, and F-measure attained in prediction are around 82%. These findings support our hypothesis that the hierarchical neural network can help developers predict clone consistency effectively in the case of cross-project incubation when insufficient data are available at the early stage of software development.