Chun Chen

Defang Chen, Jian-Ping Mei, Hailin Zhang et al.

Knowledge distillation aims to compress a powerful yet cumbersome teacher model into a lightweight student model without much sacrifice of performance. For this purpose, various approaches have been proposed over the past few years, generally with elaborately designed knowledge representations, which in turn increase the difficulty of model development and interpretation. In contrast, we empirically show that a simple knowledge distillation technique is enough to significantly narrow down the teacher-student performance gap. We directly reuse the discriminative classifier from the pre-trained teacher model for student inference and train a student encoder through feature alignment with a single $\ell_2$ loss. In this way, the student model is able to achieve exactly the same performance as the teacher model provided that their extracted features are perfectly aligned. An additional projector is developed to help the student encoder match with the teacher classifier, which renders our technique applicable to various teacher and student architectures. Extensive experiments demonstrate that our technique achieves state-of-the-art results at the modest cost of compression ratio due to the added projector.

Wujie Sun, Defang Chen, Can Wang et al.

Although diffusion model has shown great potential for generating higher quality images than GANs, slow sampling speed hinders its wide application in practice. Progressive distillation is thus proposed for fast sampling by progressively aligning output images of $N$-step teacher sampler with $N/2$-step student sampler. In this paper, we argue that this distillation-based accelerating method can be further improved, especially for few-step samplers, with our proposed \textbf{C}lassifier-based \textbf{F}eature \textbf{D}istillation (CFD). Instead of aligning output images, we distill teacher's sharpened feature distribution into the student with a dataset-independent classifier, making the student focus on those important features to improve performance. We also introduce a dataset-oriented loss to further optimize the model. Experiments on CIFAR-10 show the superiority of our method in achieving high quality and fast sampling. Code is provided at \url{https://github.com/zju-SWJ/RCFD}.

Yu He, Boheng Li, Liu Liu et al.

Membership Inference Attacks (MIAs) aim to predict whether a data sample belongs to the model's training set or not. Although prior research has extensively explored MIAs in Large Language Models (LLMs), they typically require accessing to complete output logits (\ie, \textit{logits-based attacks}), which are usually not available in practice. In this paper, we study the vulnerability of pre-trained LLMs to MIAs in the \textit{label-only setting}, where the adversary can only access generated tokens (text). We first reveal that existing label-only MIAs have minor effects in attacking pre-trained LLMs, although they are highly effective in inferring fine-tuning datasets used for personalized LLMs. We find that their failure stems from two main reasons, including better generalization and overly coarse perturbation. Specifically, due to the extensive pre-training corpora and exposing each sample only a few times, LLMs exhibit minimal robustness differences between members and non-members. This makes token-level perturbations too coarse to capture such differences. To alleviate these problems, we propose \textbf{PETAL}: a label-only membership inference attack based on \textbf{PE}r-\textbf{T}oken sem\textbf{A}ntic simi\textbf{L}arity. Specifically, PETAL leverages token-level semantic similarity to approximate output probabilities and subsequently calculate the perplexity. It finally exposes membership based on the common assumption that members are `better' memorized and have smaller perplexity. We conduct extensive experiments on the WikiMIA benchmark and the more challenging MIMIR benchmark. Empirically, our PETAL performs better than the extensions of existing label-only attacks against personalized LLMs and even on par with other advanced logit-based attacks across all metrics on five prevalent open-source LLMs.

Defang Chen, Jian-Ping Mei, Yuan Zhang et al.

Knowledge distillation is a technique to enhance the generalization ability of a student model by exploiting outputs from a teacher model. Recently, feature-map based variants explore knowledge transfer between manually assigned teacher-student pairs in intermediate layers for further improvement. However, layer semantics may vary in different neural networks and semantic mismatch in manual layer associations will lead to performance degeneration due to negative regularization. To address this issue, we propose Semantic Calibration for cross-layer Knowledge Distillation (SemCKD), which automatically assigns proper target layers of the teacher model for each student layer with an attention mechanism. With a learned attention distribution, each student layer distills knowledge contained in multiple teacher layers rather than a specific intermediate layer for appropriate cross-layer supervision. We further provide theoretical analysis of the association weights and conduct extensive experiments to demonstrate the effectiveness of our approach. Code is avaliable at \url{https://github.com/DefangChen/SemCKD}.

Yuke Hu, Zheng Li, Zhihao Liu et al.

Vision-Language Models (VLMs), built on pre-trained vision encoders and large language models (LLMs), have shown exceptional multi-modal understanding and dialog capabilities, positioning them as catalysts for the next technological revolution. However, while most VLM research focuses on enhancing multi-modal interaction, the risks of data misuse and leakage have been largely unexplored. This prompts the need for a comprehensive investigation of such risks in VLMs. In this paper, we conduct the first analysis of misuse and leakage detection in VLMs through the lens of membership inference attack (MIA). In specific, we focus on the instruction tuning data of VLMs, which is more likely to contain sensitive or unauthorized information. To address the limitation of existing MIA methods, we introduce a novel approach that infers membership based on a set of samples and their sensitivity to temperature, a unique parameter in VLMs. Based on this, we propose four membership inference methods, each tailored to different levels of background knowledge, ultimately arriving at the most challenging scenario. Our comprehensive evaluations show that these methods can accurately determine membership status, e.g., achieving an AUC greater than 0.8 targeting a small set consisting of only 5 samples on LLaVA.

Shen Han, Zhiyao Zhou, Jiawei Chen et al.

Graph Neural Networks (GNNs) have become a prominent approach for learning from graph-structured data. However, their effectiveness can be significantly compromised when the graph structure is suboptimal. To address this issue, Graph Structure Learning (GSL) has emerged as a promising technique that refines node connections adaptively. Nevertheless, we identify two key limitations in existing GSL methods: 1) Most methods primarily focus on node similarity to construct relationships, while overlooking the quality of node information. Blindly connecting low-quality nodes and aggregating their ambiguous information can degrade the performance of other nodes. 2) The constructed graph structures are often constrained to be symmetric, which may limit the model's flexibility and effectiveness. To overcome these limitations, we propose an Uncertainty-aware Graph Structure Learning (UnGSL) strategy. UnGSL estimates the uncertainty of node information and utilizes it to adjust the strength of directional connections, where the influence of nodes with high uncertainty is adaptively reduced. Importantly, UnGSL serves as a plug-in module that can be seamlessly integrated into existing GSL methods with minimal additional computational cost. In our experiments, we implement UnGSL into six representative GSL methods, demonstrating consistent performance improvements.