Hangwei Qian

Composite materials design requires understanding complex microstructural characteristics, necessitating the integration of heterogeneous data sources with artificial intelligence. Current multimodal learning frameworks are mostly developed for crystalline or polymer systems with discrete structure-property mappings and well-defined structural graphs, but fail to model the continuous and nonlinear composite design spaces under data scarcity. Here we present ordinality as a core principle when building multimodal representations for composite materials. We propose ORDinal-aware imagE-tabulaR (ORDER) alignment to integrate microstructures with tabular material descriptors and apply physics-based surrogate signals to eliminate the need for full property annotation. ORDER ensures similar target properties occupy nearby regions in the latent space, preserving the continuous nature of composite properties and enabling meaningful interpolation between sparsely observed designs. Evaluated on nanofiber and carbon fiber composite datasets, ORDER consistently outperforms alignment-oriented and property-aware baselines across property prediction, cross-modal retrieval, and microstructure generation tasks.

Graph Prompt Learning (GPL) effectively bridges the gap between graph pre-training and downstream tasks, addressing the bottleneck in knowledge transfer. However, its susceptibility to backdoor attacks during pre-training remains largely unexplored. This work systematically investigates the backdoor risks in GPL and introduces TrojanGPL-a cross-context backdoor attack that embeds triggers solely through manipulations of the pre-trained encoder, independent of downstream knowledge. Our analysis reveals that jointly optimizing trigger graphs with the prompt mechanism not only narrows the gap between upstream and downstream main tasks but also aligns main and backdoor learning objectives, facilitating seamless backdoor transfer from pre-trained models to diverse downstream applications. To counter this threat, we propose ES-Detector, a defense mechanism that identifies backdoored graph samples by detecting discriminative embedding shifts upon the removal of important nodes. Specifically, when clean and backdoored graphs are predicted as the same class, clean graphs exhibit stable embedding changes, whereas backdoored graphs show significant deviations from clean ones. Extensive experiments on 5 benchmark datasets and 3 GPL frameworks demonstrate that TrojanGPL achieves an attack success rate exceeding 90% in different cross-context scenarios while minimally impacting the performance of the main task. Meanwhile, ES-Detector reduces the attack success rate by at least 60% across all the test settings, validating its practical robustness without prior knowledge of attack patterns.

Next Point-of-Interest (POI) recommendation task focuses on mining user behavioral preference patterns from historical check-ins to provide personalized suggestions for the next destination. Existing methods primarily rely on shallow contextual information and handcrafted feature interactions to predict the next POI. However, the inherent sparsity and complexity of user mobility patterns limit the computational capacity of non-reasoning models to capture deep intent, while large language models (LLMs) perform suboptimally because they lack a deep understanding of semantic IDs (SIDs) when SIDs are trained separately. To address these limitations, we propose Think2Go, a novel generative next POI recommendation framework, which enhances the model's comprehension of SID representations and explores diverse spatial-temporal patterns via test-time computational scaling. We unify supervised fine-tuning (SFT) and reinforcement learning (RL)-based reasoning within a single architecture, enabling joint optimization of memorization and adaptive reasoning to better retain user behavior patterns while exploring diverse user preferences. To further calibrate policy optimization in adaptive reasoning, we propose two advantage weighting mechanisms that integrate (1) prompt epistemic uncertainty, estimated via kernel density methods to assess the spatial-temporal periodic pattern alignment between queries and user history, promoting increased exploration under high epistemic uncertainty; and (2) reward-informed advantage scaling, captured by normalizing rewards against their maxima to adapt update magnitudes, thereby improving training stability and mitigating overfitting to noisy signals. This joint calibration forms an implicit curriculum learning strategy, delivering fine-grained, instance-aware policy updates that prevent entropy collapse and support robust exploration. Extensive experiments conducted on three real-world datasets demonstrate that Think2Go exhibits strong generalization capabilities and enhances the LLM's understanding of SIDs.

Fully Test-Time Adaptation (FTTA) addresses domain shifts without access to source data and training protocols of the pre-trained models. Traditional strategies that align source and target feature distributions are infeasible in FTTA due to the absence of training data and unpredictable target domains. In this work, we exploit a dual perspective on FTTA, and propose Agnostic FTTA (AFTTA) as a novel formulation that enables the usage of off-the-shelf domain transformations during test-time to enable direct generalization to unforeseeable target data. To address this, we develop an uncover-and-unlearn approach. First, we uncover potential unwanted shifts between source and target domains by simulating them through predefined mappings and consider them as nuisances. Then, during test-time prediction, the model is enforced to unlearn these nuisances by regularizing the consequent shifts in latent representations and label predictions. Specifically, a mutual information-based criterion is devised and applied to guide nuisances unlearning in the feature space and encourage confident and consistent prediction in label space. Our proposed approach explicitly addresses agnostic domain shifts, enabling superior model generalization under FTTA constraints. Extensive experiments on various tasks, involving corruption and style shifts, demonstrate that our method consistently outperforms existing approaches.

Graph Prompt Learning (GPL) bridges significant disparities between pretraining and downstream applications to alleviate the knowledge transfer bottleneck in real-world graph learning. While GPL offers superior effectiveness in graph knowledge transfer and computational efficiency, the security risks posed by backdoor poisoning effects embedded in pretrained models remain largely unexplored. Our study provides a comprehensive analysis of GPL's vulnerability to backdoor attacks. We introduce CrossBA, the first cross-context backdoor attack against GPL, which manipulates only the pretraining phase without requiring knowledge of downstream applications. Our investigation reveals both theoretically and empirically that tuning trigger graphs, combined with prompt transformations, can seamlessly transfer the backdoor threat from pretrained encoders to downstream applications.Through extensive experiments involving 3 representative GPL methods across 5 distinct cross-context scenarios and 5 benchmark datasets of node and graph classification tasks, we demonstrate that CrossBA consistently achieves high attack success rates while preserving the functionality of downstream applications over clean input. We also explore potential countermeasures against CrossBA and conclude that current defenses are insufficient to mitigate CrossBA. Our study highlights the persistent backdoor threats to GPL systems, raising trustworthiness concerns in the practices of GPL techniques.