In this supplementary material, we provide more details regarding implementation details in Appendix B, more analysis of ERDA in Appendix C, full experimental results in Appendix D, and studies on parameters in Appendix E.

Our code and pre-trained models will be released.

Embeddings have become a cornerstone in the functionality of large language models (LLMs) due to their ability to transform text data into rich, dense numerical representations that capture semantic and syntactic properties. These embedding vector databases serve as the long-term memory of LLMs, enabling efficient handling of a wide range of natural language processing tasks. However, the surge in popularity of embedding vector databases in LLMs has been accompanied by significant concerns about privacy leakage. Embedding vector databases are particularly vulnerable to embedding inversion attacks, where adversaries can exploit the embeddings to reverse-engineer and extract sensitive information from the original text data. Existing defense mechanisms have shown limitations, often struggling to balance security with the performance of downstream tasks. To address these challenges, we introduce Eguard, a novel defense mechanism designed to mitigate embedding inversion attacks. Eguard employs a transformer-based projection network and text mutual information optimization to safeguard embeddings while preserving the utility of LLMs. Our approach significantly reduces privacy risks, protecting over 95% of tokens from inversion while maintaining high performance across downstream tasks consistent with original embeddings.

Weight-preserving model editing techniques heavily rely on the scoping mechanism that decides when to apply an edit to the base model. These scoping mechanisms utilize distance functions in the representation space to ascertain the scope of the edit. In this work, we show that distance-based scoping functions grapple with lexical biases leading to issues such as misfires with irrelevant prompts that share similar lexical characteristics. To address this problem, we introduce, Projector Editor Networks for Model Editing (PENME),is a model editing approach that employs a compact adapter with a projection network trained via a contrastive learning objective. We demonstrate the efficacy of PENME in achieving superior results while being compute efficient and flexible to adapt across model architectures.

Utilization of intracranial electroencephalography (iEEG) is rapidly increasing for clinical and brain-computer interface applications. iEEG facilitates the recording of neural activity with high spatial and temporal resolution, making it a desirable neuroimaging modality for studying neural dynamics. Despite its benefits, iEEG faces challenges such as inter-subject variability in electrode implantation, which makes the development of unified neural decoder models across different patients difficult. In this research, we introduce a novel decoder model that is robust to inter-subject electrode implantation variability. We call this model RISE-iEEG, which stands for Robust Inter-Subject Electrode Implantation Variability iEEG Classifier. RISE-iEEG employs a deep neural network structure preceded by a patient-specific projection network. The projection network maps the neural data of individual patients onto a common low-dimensional space, compensating for the implantation variability. In other words, we developed an iEEG decoder model that can be applied across multiple patients' data without requiring the coordinates of electrode for each patient. The performance of RISE-iEEG across multiple datasets, including the Audio-Visual dataset, Music Reconstruction dataset, and Upper-Limb Movement dataset, surpasses that of state-of-the-art iEEG decoder models such as HTNet and EEGNet. Our analysis shows that the performance of RISE-iEEG is 10\% higher than that of HTNet and EEGNet in terms of F1 score, with an average F1 score of 83\%, which is the highest result among the evaluation methods defined. Furthermore, the analysis of projection network weights in the Music Reconstruction dataset across patients suggests that the Superior Temporal lobe serves as the primary encoding neural node. This finding aligns with the auditory processing physiology.

In personalized federated learning (PFL), it is widely recognized that achieving both high model generalization and effective personalization poses a significant challenge due to their conflicting nature. As a result, existing PFL methods can only manage a trade-off between these two objectives. This raises an interesting question: Is it feasible to develop a model capable of achieving both objectives simultaneously? Our paper presents an affirmative answer, and the key lies in the observation that deep models inherently exhibit hierarchical architectures, which produce representations with various levels of generalization and personalization at different stages. A straightforward approach stemming from this observation is to select multiple representations from these layers and combine them to concurrently achieve generalization and personalization. However, the number of candidate representations is commonly huge, which makes this method infeasible due to high computational costs.To address this problem, we propose DualFed, a new method that can directly yield dual representations correspond to generalization and personalization respectively, thereby simplifying the optimization task. Specifically, DualFed inserts a personalized projection network between the encoder and classifier. The pre-projection representations are able to capture generalized information shareable across clients, and the post-projection representations are effective to capture task-specific information on local clients. This design minimizes the mutual interference between generalization and personalization, thereby achieving a win-win situation. Extensive experiments show that DualFed can outperform other FL methods. Code is available at https://github.com/GuogangZhu/DualFed.

Unsupervised graph-level anomaly detection (UGAD) has received remarkable performance in various critical disciplines, such as chemistry analysis and bioinformatics. Existing UGAD paradigms often adopt data augmentation techniques to construct multiple views, and then employ different strategies to obtain representations from different views for jointly conducting UGAD. However, most previous works only considered the relationship between nodes/graphs from a limited receptive field, resulting in some key structure patterns and feature information being neglected. In addition, most existing methods consider different views separately in a parallel manner, which is not able to explore the inter-relationship across different views directly. Thus, a method with a larger receptive field that can explore the inter-relationship across different views directly is in need. In this paper, we propose a novel Simplified Transformer with Cross-View Attention for Unsupervised Graph-level Anomaly Detection, namely, CVTGAD. To increase the receptive field, we construct a simplified transformer-based module, exploiting the relationship between nodes/graphs from both intra-graph and inter-graph perspectives. Furthermore, we design a cross-view attention mechanism to directly exploit the view co-occurrence between different views, bridging the inter-view gap at node level and graph level. To the best of our knowledge, this is the first work to apply transformer and cross attention to UGAD, which realizes graph neural network and transformer working collaboratively. Extensive experiments on 15 real-world datasets of 3 fields demonstrate the superiority of CVTGAD on the UGAD task. The code is available at \url{https://github.com/jindongli-Ai/CVTGAD}.

Federated Learning (FL) is a machine learning paradigm that allows decentralized clients to learn collaboratively without sharing their private data. However, excessive computation and communication demands pose challenges to current FL frameworks, especially when training large-scale models. To prevent these issues from hindering the deployment of FL systems, we propose a lightweight framework where clients jointly learn to fuse the representations generated by multiple fixed pre-trained models rather than training a large-scale model from scratch. This leads us to a more practical FL problem by considering how to capture more client-specific and class-relevant information from the pre-trained models and jointly improve each client's ability to exploit those off-the-shelf models. In this work, we design a Federated Prototype-wise Contrastive Learning (FedPCL) approach which shares knowledge across clients through their class prototypes and builds client-specific representations in a prototype-wise contrastive manner. Sharing prototypes rather than learnable model parameters allows each client to fuse the representations in a personalized way while keeping the shared knowledge in a compact form for efficient communication. We perform a thorough evaluation of the proposed FedPCL in the lightweight framework, measuring and visualizing its ability to fuse various pre-trained models on popular FL datasets.