High-level automation is increasingly critical in AI, driven by rapid advances in large language models (LLMs) and AI agents. However, LLMs, despite their general reasoning power, struggle significantly in specialized, data-sensitive tasks such as designing Graph Neural Networks (GNNs). This difficulty arises from (1) the inherent knowledge gaps in modeling the intricate, varying relationships between graph properties and suitable architectures and (2) the external noise from misleading descriptive inputs, often resulting in generic or even misleading model suggestions. Achieving proficiency in designing data-aware models—defined as the meta-level capability to systematically accumulate, interpret, and apply data-specific design knowledge—remains challenging for existing automated approaches, due to their inefficient construction and application of meta-knowledge. To achieve meta-level proficiency, we propose DesiGNN, a knowledge-centered framework that systematically converts past model design experience into structured, fine-grained knowledge priors well-suited for meta-learning with LLMs. To account for the inherent variability and external noise, DesiGNN aligns empirical property filtering from extensive benchmarks with adaptive elicitation of literature insights via LLMs. By constructing a solid meta-knowledge between unseen graph understanding and known effective architecture patterns, DesiGNN can deliver top-5.77% initial model proposals for unseen datasets within seconds and achieve consistently superior performance with minimal search cost compared to baselines.

In recent years, dynamic knowledge graph embedding (DKGE) has been widely studied to deal with large-scale dynamic knowledge graphs (DKG). The core idea is to encode dynamic information within DKGs into embedding vectors and decode them for various downstream tasks on the DKGs. Plenty of contributions have been made to this field. Full retraining DKG models additionally encode temporal information for higher performance, while neighboring retraining models view time data as dynamic changes in graph topology for better efficiency. However, existing approaches within these categories suffer from either effectiveness-insufficient or efficiency-insufficient issues. Recent contributions in graph area propose solutions to selectively retrain the models by choosing training data following certain criteria, but the majority of selective retraining models are designed for homogeneous graphs. The heterogeneous graph information and large graph sizes make it improper to transfer methods across scenarios. In this paper, we propose an efficient selective incremental training framework for DKGE, namely SIT. Given a restriction on training data size, we select a set of important triples instead of all triples in the DKG to improve training efficiency. In detail, we design a novel importance criteria considering DKGE model parameters, historical embedding and graph topology. Extensive experiments on open-source datasets demonstrate the effectiveness and efficiency of the SIT framework against different DKGE models.

In recent years, the design and transfer of neural network models have been widely studied due to their exceptional performance and capabilities. However, the complex nature of datasets and the vast architecture space pose significant challenges for both manual and automated algorithms in creating high-performance models. Inspired by researchers who design, train, and document the performance of various models across different datasets, this paper introduces a novel schema that transforms the benchmark data into a Knowledge Benchmark Graph (KBG), which primarily stores the facts in the form of performance(data, model). Constructing the KBG facilitates the structured storage of design knowledge, aiding subsequent model design and transfer. However, it is a non-trivial task to retrieve or design suitable neural networks based on the KBG, as real-world data are often off the records. To tackle this challenge, we propose transferring existing models stored in KBG by establishing correlations between unseen and previously seen datasets. Given that measuring dataset similarity is a complex and open-ended issue, we explore the potential for evaluating the correctness of the similarity function. Then, we further integrate the KBG with Large Language Models (LLMs), assisting LLMs to think and retrieve existing model knowledge in a manner akin to humans when designing or transferring models. We demonstrate our method specifically in the context of Graph Neural Network (GNN) architecture design, constructing a KBG (with 26,206 models, 211,669 performance records, and 2,540,064 facts) and validating the effectiveness of leveraging the KBG to promote GNN architecture design.

Node classification is a key task in temporal graph learning (TGL). Real-life temporal graphs often introduce new node classes over time, but existing TGL methods assume a fixed set of classes. This assumption brings limitations, as updating models with full data is costly, while focusing only on new classes results in forgetting old ones. Graph continual learning (GCL) methods mitigate forgetting using old-class subsets but fail to account for their evolution. We define this novel problem as temporal graph continual learning (TGCL), which focuses on efficiently maintaining up-to-date knowledge of old classes. To tackle TGCL, we propose a selective learning framework that substitutes the old-class data with its subsets, Learning Towards the Future (LTF). We derive an upper bound on the error caused by such replacement and transform it into objectives for selecting and learning subsets that minimize classification error while preserving the distribution of the full old-class data. Experiments on three real-world datasets show that LTF effectively addresses the TGCL challenge.

Temporal link prediction in dynamic graphs is a critical task with applications in diverse domains such as social networks, recommendation systems, and e-commerce platforms. While existing Temporal Graph Neural Networks (T-GNNs) have achieved notable success by leveraging complex architectures to model temporal and structural dependencies, they often suffer from scalability and efficiency challenges due to high computational overhead. In this paper, we propose EAGLE, a lightweight framework that integrates short-term temporal recency and long-term global structural patterns. EAGLE consists of a time-aware module that aggregates information from a node's most recent neighbors to reflect its immediate preferences, and a structure-aware module that leverages temporal personalized PageRank to capture the influence of globally important nodes. To balance these attributes, EAGLE employs an adaptive weighting mechanism to dynamically adjust their contributions based on data characteristics. Also, EAGLE eliminates the need for complex multi-hop message passing or memory-intensive mechanisms, enabling significant improvements in efficiency. Extensive experiments on seven real-world temporal graphs demonstrate that EAGLE consistently achieves superior performance against state-of-the-art T-GNNs in both effectiveness and efficiency, delivering more than a 50× speedup over effective transformer-based T-GNNs.

Recently, continual learning (CL) has attracted much attention due to its widespread applications in the real world. Given a set of data sets sequentially, continual learning aims to achieve good performance on the new data sets while avoiding deterioration in performance on the old data sets. Despite the success, most CL models follow the supervised setting, which limits their potential in data scarcity cases. Thus, some pioneering works study unsupervised CL (UCL) to discuss what CL tricks suit the unsupervised setting. However, their advancements lack in-depth analysis of the characteristics of UCL, especially the lack of attention to the use of old data. We identify that using old data sets is essential for improving the UCL model performance while existing works ignore them. Unfortunately, given a limited data storage budget, it is a nontrivial task to select representative data and effectively replay them without label assistance. To further improve the UCL performance, we present a new method in this paper, named Effective Data Selection and Replay (EDSR) for UCL. Specifi-cally, we analyze that entropy can be an effective data selection metric, where representative data usually exhibit the highest entropy in the representation space. Then, to balance the model stability for old data and the plasticity for new data, we adopt a strategy of replaying those stored representative data with a noise-enhanced knowledge distillation process. The empirical study demonstrates the outstanding performance of EDSR on benchmark computer vision data sets. Especially, EDSR shows strong resistance to forgetting old data knowledge while maintaining high accuracy. The implementation is publicly available at https://github.com/LeeJarvis996/edsr_project/tree/main/EDSR.

Recently, incremental learning has attracted a lot of interest in both research communities and industries. Generally, given a series of data sets sequentially, it tries to achieve good performance on the new data set while maintaining not bad performance on the old ones. Despite the recent success of incremental learning, existing works mainly assume that the coming data set is from the feature space of old ones, i.e., homogeneous feature space. And they adopt one feature extractor to forcibly project different feature spaces into one space. However, this assumption is hard to hold in real-world scenarios. Especially, the attributes of tables may sequentially increase in tabular learning. Thus, classic incremental learning models may hinder their effectiveness. In this paper, we propose a new method, incremental tabular learning on heterogeneous feature space (ILEAHE) to solve this issue. We first propose the ideas that feature extractors should be decomposed into shared and specific extractors to process the shared and specific features across different data sets respectively. Then, we propose a novel measurement named discriminative ability to measure specific extractors. Thus, two kinds of extractors can be discriminated and the specific extractor will more focus on those domain-specific features. We further demonstrate the effectiveness of ILEAHE through empirical studies.