Influence Maximization (IM) is a critical task in social network analysis, yet its application to large-scale networks is often hampered by computational complexity and the challenge of effectively identifying influential communities. Existing community-based IM algorithms frequently rely on traditional detection methods that may not scale well or adequately capture influence propagation dynamics. While deep learning has advanced community detection, the explicit integration of influence properties into these models remains a significant gap. This paper introduces the Deep Learning-based Community-aware Influence Maximization (DLCIM) algorithm, a novel approach that synergistically combines deep learning for community detection with a tailored focus on influence diffusion characteristics. DLCIM employs an autoencoder framework with a novel influence-aware modularity maximization objective to learn node representations that are sensitive to information propagation. Subsequently, it filters important communities and allocates seed quotas proportionally, followed by efficient seed selection using established techniques within these communities. Comprehensive experiments on real-world datasets demonstrate that DLCIM achieves superior influence spread and competitive computational efficiency compared to state-of-the-art baseline algorithms.

Merge conflicts in Distributed Version Control Systems (DVCS) like Git are a persistent challenge in the software development lifecycle. If not handled properly or overlooked, they can lead to issues like hindering collaboration and introducing errors. While automated resolution methods exist, prevailing approaches—such as multi-class classification and direct code generation—often suffer from limited interpretability, demanding substantial manual effort to refine predictions, and risk producing subtly flawed code. Critically, existing research often overlooks a prevalent conflict type: adjacent-line conflicts, where independent edits to contiguous lines are flagged by tools like Git. Our empirical analysis reveals that these make up a substantial portion of all conflicts. Moreover, they can often be resolved using simple patterns.Motivated by these limitations and empirical findings, we propose a novel approach: modeling merge conflict resolution as edit script selection. Instead of predicting abstract categories or generating code from scratch, our method makes a binary decision for each atomic line-level edit script contributing to the conflict: accept or reject. Our method inherently makes the reasoning behind proposed solutions transparent, as decisions directly correspond to individual, developer-authored code modifications. It also aligns closely with how developers naturally approach conflict analysis by considering each change in context. Our method applies to the vast majority (94.18%) of conflicts that can be correctly resolved without introducing any novel code lines—that is, lines not present in any of the parent versions being merged; this selection process directly yields the resolved code by applying the chosen subset of existing edits. We developed EditFusion, a deep learning model that performs edit script selection by leveraging semantic embeddings and edit metadata. Extensive evaluation on large-scale, real-world datasets demonstrates both the prevalence of adjacent-line conflicts and EditFusion’s superior performance in accurately resolving conflicts compared to baselines. Our work represents an attempt towards more transparent, intuitive, and practical automated merge conflict resolution.

Object detection with the capacity to incrementally adapt to new domains is a crucial yet relatively under-explored research topic. The catastrophic forgetting problem presents a significant challenge to achieve this goal, where the model’s performance improves quickly in new conditions but deteriorates sharply in old ones after several incremental learning sessions. Drawing on recent discoveries in visual memories of the human brain, we introduce the Topology-Preserving Domain Incremental Object Detection (TP-DIOD) approach, which aims to address the catastrophic forgetting problem by extracting the topological structure of the feature space learned by the Convolutional Neural Network (CNN) model and preserving this topology during the subsequent incremental learning sessions. Specifically, we model the feature space topology using the self-organizing map (SOM) and construct an anchor image set based on the centroid vectors of the SOM nodes to memorize the feature space topology. We then develop the anchor loss function to penalize the topological changes of the feature space during the subsequent incremental learning sessions. Experimental evaluations on two sets of datasets demonstrate the effectiveness of the proposed TP-DIOD method in mitigating the catastrophic forgetting problem and achieving high accuracy on both old and new domain datasets.

Face Super-resolution (FSR) task works for generating high-resolution (HR) face images from the corresponding low-resolution (LR) inputs, which has received a lot of attentions because of the wide application prospects. However, due to the diversity of facial texture and the difficulty of reconstructing detailed content from degraded images, FSR technology is still far away from being solved. In this paper, we propose a novel and effective face super-resolution framework based on Transformer, namely Interconnected Dual Pyramid Transformer (IDPT). Instead of straightly stacking cascaded feature reconstruction blocks, the proposed IDPT designs the pyramid encoder/decoder Transformer architecture to extract coarse and detailed facial textures respectively, while the relationship between the dual pyramid Transformers is further explored by a bottom pyramid feature extractor. The pyramid encoder/decoder structure is devised to adapt various characteristics of textures in different spatial spaces hierarchically. A novel fusing modulation module is inserted in each spatial layer to guide the refinement of detailed texture by the corresponding coarse texture, while fusing the shallow-layer coarse feature and corresponding deep-layer detailed feature simultaneously. Extensive experiments and visualizations on various datasets demonstrate the superiority of the proposed method for face super-resolution tasks.