Abstract:Fine-tuning large language models on private data for downstream applications poses significant privacy risks in potentially exposing sensitive information. Several popular community platforms now offer convenient distribution of a large variety of pre-trained models, allowing anyone to publish without rigorous verification. This scenario creates a privacy threat, as pre-trained models can be intentionally crafted to compromise the privacy of fine-tuning datasets. In this study, we introduce a novel poisoning technique that uses model-unlearning as an attack tool. This approach manipulates a pre-trained language model to increase the leakage of private data during the fine-tuning process. Our method enhances both membership inference and data extraction attacks while preserving model utility. Experimental results across different models, datasets, and fine-tuning setups demonstrate that our attacks significantly surpass baseline performance. This work serves as a cautionary note for users who download pre-trained models from unverified sources, highlighting the potential risks involved.
Abstract:Many state-of-the-art RGB-T trackers have achieved remarkable results through modality fusion. However, these trackers often either overlook temporal information or fail to fully utilize it, resulting in an ineffective balance between multi-modal and temporal information. To address this issue, we propose a novel Cross Fusion RGB-T Tracking architecture (CFBT) that ensures the full participation of multiple modalities in tracking while dynamically fusing temporal information. The effectiveness of CFBT relies on three newly designed cross spatio-temporal information fusion modules: Cross Spatio-Temporal Augmentation Fusion (CSTAF), Cross Spatio-Temporal Complementarity Fusion (CSTCF), and Dual-Stream Spatio-Temporal Adapter (DSTA). CSTAF employs a cross-attention mechanism to enhance the feature representation of the template comprehensively. CSTCF utilizes complementary information between different branches to enhance target features and suppress background features. DSTA adopts the adapter concept to adaptively fuse complementary information from multiple branches within the transformer layer, using the RGB modality as a medium. These ingenious fusions of multiple perspectives introduce only less than 0.3\% of the total modal parameters, but they indeed enable an efficient balance between multi-modal and temporal information. Extensive experiments on three popular RGB-T tracking benchmarks demonstrate that our method achieves new state-of-the-art performance.
Abstract:In distributed learning settings, models are iteratively updated with shared gradients computed from potentially sensitive user data. While previous work has studied various privacy risks of sharing gradients, our paper aims to provide a systematic approach to analyze private information leakage from gradients. We present a unified game-based framework that encompasses a broad range of attacks including attribute, property, distributional, and user disclosures. We investigate how different uncertainties of the adversary affect their inferential power via extensive experiments on five datasets across various data modalities. Our results demonstrate the inefficacy of solely relying on data aggregation to achieve privacy against inference attacks in distributed learning. We further evaluate five types of defenses, namely, gradient pruning, signed gradient descent, adversarial perturbations, variational information bottleneck, and differential privacy, under both static and adaptive adversary settings. We provide an information-theoretic view for analyzing the effectiveness of these defenses against inference from gradients. Finally, we introduce a method for auditing attribute inference privacy, improving the empirical estimation of worst-case privacy through crafting adversarial canary records.
Abstract:Referring Video Object Segmentation is an emerging multi-modal task that aims to segment objects in the video given a natural language expression. In this work, we build two instance-centric models and fuse predicted results from frame-level and instance-level. First, we introduce instance mask into the DETR-based model for query initialization to achieve temporal enhancement and employ SAM for spatial refinement. Secondly, we build an instance retrieval model conducting binary instance mask classification whether the instance is referred. Finally, we fuse predicted results and our method achieved a score of 52.67 J&F in the validation phase and 60.36 J&F in the test phase, securing the final ranking of 3rd place in the 6-th LSVOS Challenge RVOS Track.
Abstract:Iris recognition is widely used in high-security scenarios due to its stability and distinctiveness. However, the acquisition of iris images typically requires near-infrared illumination and near-infrared band filters, leading to significant and consistent differences in imaging across devices. This underscores the importance of developing cross-domain capabilities in iris anti-spoofing methods. Despite this need, there is no dataset available that comprehensively evaluates the generalization ability of the iris anti-spoofing task. To address this gap, we propose the IrisGeneral dataset, which includes 10 subsets, belonging to 7 databases, published by 4 institutions, collected with 6 types of devices. IrisGeneral is designed with three protocols, aimed at evaluating average performance, cross-racial generalization, and cross-device generalization of iris anti-spoofing models. To tackle the challenge of integrating multiple sub-datasets in IrisGeneral, we employ multiple parameter sets to learn from the various subsets. Specifically, we utilize the Mixture of Experts (MoE) to fit complex data distributions using multiple sub-neural networks. To further enhance the generalization capabilities, we introduce a novel method Masked-MoE (MMoE). It randomly masks a portion of tokens for some experts and requires their outputs to be similar to the unmasked experts, which improves the generalization ability and effectively mitigates the overfitting issue produced by MoE. We selected ResNet50, VIT-B/16, CLIP, and FLIP as representative models and benchmarked them on the IrisGeneral dataset. Experimental results demonstrate that our proposed MMoE with CLIP achieves the best performance on IrisGeneral.
Abstract:Contrastive Language-Image Pre-training (CLIP), which excels at abstracting open-world representations across domains and modalities, has become a foundation for a variety of vision and multimodal tasks. However, recent studies reveal that CLIP has severe visual shortcomings, such as which can hardly distinguish orientation, quantity, color, structure, etc. These visual shortcomings also limit the perception capabilities of multimodal large language models (MLLMs) built on CLIP. The main reason could be that the image-text pairs used to train CLIP are inherently biased, due to the lack of the distinctiveness of the text and the diversity of images. In this work, we present a simple post-training approach for CLIP models, which largely overcomes its visual shortcomings via a self-supervised diffusion process. We introduce DIVA, which uses the DIffusion model as a Visual Assistant for CLIP. Specifically, DIVA leverages generative feedback from text-to-image diffusion models to optimize CLIP representations, with only images (without corresponding text). We demonstrate that DIVA improves CLIP's performance on the challenging MMVP-VLM benchmark which assesses fine-grained visual abilities to a large extent (e.g., 3-7%), and enhances the performance of MLLMs and vision models on multimodal understanding and segmentation tasks. Extensive evaluation on 29 image classification and retrieval benchmarks confirms that our framework preserves CLIP's strong zero-shot capabilities. The code will be available at https://github.com/baaivision/DIVA.
Abstract:Portrait editing is challenging for existing techniques due to difficulties in preserving subject features like identity. In this paper, we propose a training-based method leveraging auto-generated paired data to learn desired editing while ensuring the preservation of unchanged subject features. Specifically, we design a data generation process to create reasonably good training pairs for desired editing at low cost. Based on these pairs, we introduce a Multi-Conditioned Diffusion Model to effectively learn the editing direction and preserve subject features. During inference, our model produces accurate editing mask that can guide the inference process to further preserve detailed subject features. Experiments on costume editing and cartoon expression editing show that our method achieves state-of-the-art quality, quantitatively and qualitatively.
Abstract:The Diffusion models, widely used for image generation, face significant challenges related to their broad applicability due to prolonged inference times and high memory demands. Efficient Post-Training Quantization (PTQ) is crucial to address these issues in traditional models. Unlike those models, diffusion models critically rely on the time-step $t$ for effective multi-round denoising. Typically, $t$ from the finite set $\{1, \ldots, T\}$ is encoded into a hypersensitive temporal feature by several modules, entirely independent of the sampling data. However, existing PTQ methods do not optimize these modules individually. Instead, they employ unsuitable reconstruction objectives and complex calibration methods, leading to significant disturbances in the temporal feature and denoising trajectory. To address these challenges, we introduce a novel quantization framework: 1)~TIB-based Maintenance: Based on our innovative Temporal Information Block~(TIB) definition, Temporal Information-aware Reconstruction~(TIAR) and Finite Set Calibration~(FSC) are developed to efficiently align full precision temporal features. 2)~Cache-based Maintenance: Instead of indirect and complex optimization for the related modules, pre-computing and caching quantized counterparts of temporal features are developed to minimize errors. 3)~Disturbance-aware Selection: Employ temporal feature errors to guide a fine-grained selection for superior maintenance. This framework preserves most of the temporal information and ensures high-quality end-to-end generation. Extensive testing on various datasets and diffusion models confirms our superior results. Notably, our approach closely matches the performance of the full-precision model under 4-bit quantization. Furthermore, the quantized SD-XL model achieves hardware acceleration of 2.20$\times$ on CPU and 5.76$\times$ on GPU demonstrating its efficiency.
Abstract:Randomized smoothing is a defensive technique to achieve enhanced robustness against adversarial examples which are small input perturbations that degrade the performance of neural network models. Conventional randomized smoothing adds random noise with a fixed noise level for every input sample to smooth out adversarial perturbations. This paper proposes a new variational framework that uses a per-sample noise level suitable for each input by introducing a noise level selector. Our experimental results demonstrate enhancement of empirical robustness against adversarial attacks. We also provide and analyze the certified robustness for our sample-wise smoothing method.
Abstract:Biosignal-based hand gesture classification is an important component of effective human-machine interaction. For multimodal biosignal sensing, the modalities often face data loss due to missing channels in the data which can adversely affect the gesture classification performance. To make the classifiers robust to missing channels in the data, this paper proposes using Random Channel Ablation (RChA) during the training process. Ultrasound and force myography (FMG) data were acquired from the forearm for 12 hand gestures over 2 subjects. The resulting multimodal data had 16 total channels, 8 for each modality. The proposed method was applied to convolutional neural network architecture, and compared with baseline, imputation, and oracle methods. Using 5-fold cross-validation for the two subjects, on average, 12.2% and 24.5% improvement was observed for gesture classification with up to 4 and 8 missing channels respectively compared to the baseline. Notably, the proposed method is also robust to an increase in the number of missing channels compared to other methods. These results show the efficacy of using random channel ablation to improve classifier robustness for multimodal and multi-channel biosignal-based hand gesture classification.