Program

1. Mohamed Bouguessa, Maroun Haddad TrackGAE: Tracking Dynamic Community Evolution with Graph Autoencoders
   Tracking the evolution of communities (or clusters) in dynamic networks is a critical challenge in numerous applications, including social network analysis, biological systems, and financial modeling. Existing methods primarily focus on node membership overlap while overlooking structural and attribute-based transformations, leading to inconsistencies when clusters undergo structural or attribute changes. To address these limitations, we propose TrackGAE, a two-phase deep learning framework that leverages graph autoencoders to generate temporal representations of clusters and construct evolutionary sequences that preserve community identity. In the first phase, a Temporal Graph Autoencoder extracts structural and attribute-aware cluster embeddings. In the second phase, a Clustering Graph Autoencoder refines these embeddings using a proposed Deep-Pruning mechanism to generate high-quality cluster sequences. TrackGAE captures node membership, attributes, and structure, enabling accurate tracking of dynamic clusters over time. Preliminary results on the Yelp dataset demonstrate the suitability of our approach.
2. Ayan Chatterjee,Barbara Ikica,Babak Ravandi, John Palowitch Transfer Learning for Temporal Link Prediction
   Link prediction on graphs has applications spanning from recommender systems to drug discovery. Temporal link prediction (TLP) refers to predicting future links in a temporally evolving graph and adds additional complexity related to the dynamic nature of graphs. State-of-the-art TLP models incorporate memory modules alongside graph neural networks to learn both the temporal mechanisms of incoming nodes and the evolving graph topology. However, memory modules only store information about nodes seen at train time, and hence, such models cannot be directly transferred to entirely new graphs at test time and deployment. In this work, we study a new transfer learning task for temporal link prediction, and develop transfer-effective methods for memory-laden models. Specifically, motivated by recent research showing the informativeness of structural signals for the TLP task, we augment a structural mapping module to the existing TLP model architectures, which learns a mapping from graph structural (topological) features to memory embeddings. Our work paves the way for a memory-free foundation model for TLP.
3. Nilanjana Debnath,Unnikrishnan Cheramangalath HT-Graph : Heterogeneous Continuous-Time Dynamic Graph Representation Learning using Neighbor-store with Restart
   Graphs are integral to model real-world complex systems like social networks, citation networks, transaction networks etc. Real-world graphs are mostly heterogeneous, continuously evolving dynamic graphs (i.e. Heterogeneous continuous-time dynamic graph – HCTDG). Modeling HCTDGs requires effective representation learning, which is difficult because of their entangled structural and temporal dependencies. We introduce HT-Graph, a novel framework aimed to improve link prediction in HCTDG graphs. HT-Graph addresses scalability and computational inefficiencies while focusing on enhancing link prediction accuracy. As formation of new links between nodes depends on their neighborhood, HT-Graph introduces a neighbor-aware memory module (i.e. memory module with a neighbor-store) that stores and updates local neighborhood information of each node efficiently for faster calculation of structural information, eliminating redundant computations required for traditional neighborhood sampling. To introduce parallelism, we used a neighbor-aware restarter to restart the training at any timestamp using interaction history. During training, the restarter module resets memory states at multiple timestamps and learns to mimic the encoder through the knowledge distillation process. This eliminates sequential dependencies, enabling HT-Graph to capture temporal dynamics and structural heterogeneity while ensuring scalability. HT-Graph outperforms state-of-the-art models in link prediction, providing higher scalability, efficiency, and better predictive performance even with limited data.
4. Bishal Lakha,Alessio Barboni,Massimiliano Lupo Pasini,Janet Layne,Edoardo Serra, Prasanna Balaprakash HT-Graph : Assessing the Few-Shot Learning Capabilities of Large Language Models for Graph-Based Novel Molecule Generation
   This paper investigates the potential of large language models (LLMs) to generate novel molecular structures (in the form of a graph) in a few-shot learning environment that observes the structures of several existing molecules without relying on any ground truth of how the molecule should be generated. Unlike traditional approaches that optimize existing molecules—by taking a seed structure and incrementally improving its properties—we leverage the intrinsic creativity of LLMs to learn and capture the underlying distribution of molecular data. Our method involves providing only a handful of exemplar molecules from the computationally-derived QM9 dataset as input, without explicit target outputs, to prompt the LLM to generate chemically plausible and diverse molecular structures. Experimental evaluations demonstrate that even with minimal input data, the LLM consistently produces molecules that exhibit correct chemical properties and structural similarity to known compounds. These findings suggest that LLMs are capable of determining complex molecular patterns and generalizing beyond the provided examples, thus opening new avenues for de novo molecular design. The proposed framework not only challenges the conventional optimization paradigm but also highlights the potential of LLMs as a generative tool in computational chemistry and drug discovery.
5. Bing Xue A Heterogeneous Graph and Multi-Feature Fusion Based Framework for Smart Contract Vulnerability Detection
   In recent years, graph neural networks have demon- strated strong capabilities in processing graph-structured data and have made significant progress in the field of smart contract vulnerability detection. This paper proposes a new smart contract vulnerability detection framework——HF-Sec. The framework first automatically generates heterogeneous contract graphs from the source code of Ethereum smart contracts to represent the control flow and function call relationships of the code. Then, by using a multi-source attention mechanism, the framework is able to synthesize features from different sources to capture key information from multiple perspectives. In addition, HF- Sec utilizes Fast Graph Transformer Networks and Kolmogorov- Arnold Networks to automatically extract mission-critical meta- paths and enhance the interpretability of the model. We per- formed experimental validation on a mixed dataset containing 423 contracts with vulnerabilities and 2742 contracts without vulnerabilities. The experimental results show that HF-Sec can significantly improve the accuracy of smart contract vulnerability detection, which is better than the methods based on machine learning or traditional analysis techniques. Through a series of ablation experiments, we further verified the importance of various key components in HF-Sec to improve the detection accuracy.