Add to Chrome
Add to Firefox
Add to EdgeWeilong Wang
Discovering Universal Semantic Triggers for Text-to-Image Synthesis
Feb 12, 2024
Recently text-to-image models have gained widespread attention in the community due to their controllable and high-quality generation ability. However, the robustness of such models and their potential ethical issues have not been fully explored. In this paper, we introduce Universal Semantic Trigger, a meaningless token sequence that can be added at any location within the input text yet can induce generated images towards a preset semantic target.To thoroughly investigate it, we propose Semantic Gradient-based Search (SGS) framework. SGS automatically discovers the potential universal semantic triggers based on the given semantic targets. Furthermore, we design evaluation metrics to comprehensively evaluate semantic shift of images caused by these triggers. And our empirical analyses reveal that the mainstream open-source text-to-image models are vulnerable to our triggers, which could pose significant ethical threats. Our work contributes to a further understanding of text-to-image synthesis and helps users to automatically auditing their models before deployment.
Efficient Recruitment Strategy for Collaborative Mobile Crowd Sensing Based on GCN Trustworthiness Prediction
Jun 07, 2023
Collaborative Mobile Crowd Sensing (CMCS) enhances data quality and coverage by promoting teamwork in task sensing, with worker recruitment representing a complex multi-objective optimization problem. Existing strategies mainly focus on the characteristics of workers themselves, neglecting the asymmetric trust relationships between them, which affects the rationality of task utility evaluation. To address this, this paper first employs the Mini-Batch K-Means clustering algorithm and deploys edge servers to enable efficient distributed worker recruitment. Historical data and task requirements are utilized to obtain workers' ability types and distances. A trust-directed graph in the worker's social network is input into the Graph Convolutional Network (GCN) framework for training, capturing asymmetric trustworthiness between worker pairs. Privacy leakage is prevented in CMCS scenarios through high trust values between workers. Ultimately, an undirected recruitment graph is constructed using workers' abilities, trust values, and distance weights, transforming the worker recruitment problem into a Maximum Weight Average Subgraph Problem (MWASP). A Tabu Search Recruitment (TSR) algorithm is proposed to rationally recruit a balanced multi-objective optimal task utility worker set for each task. Extensive simulation experiments on four real-world datasets demonstrate the effectiveness of the proposed strategy, outperforming other strategies.
NCL: Textual Backdoor Defense Using Noise-augmented Contrastive Learning
Mar 03, 2023
At present, backdoor attacks attract attention as they do great harm to deep learning models. The adversary poisons the training data making the model being injected with a backdoor after being trained unconsciously by victims using the poisoned dataset. In the field of text, however, existing works do not provide sufficient defense against backdoor attacks. In this paper, we propose a Noise-augmented Contrastive Learning (NCL) framework to defend against textual backdoor attacks when training models with untrustworthy data. With the aim of mitigating the mapping between triggers and the target label, we add appropriate noise perturbing possible backdoor triggers, augment the training dataset, and then pull homology samples in the feature space utilizing contrastive learning objective. Experiments demonstrate the effectiveness of our method in defending three types of textual backdoor attacks, outperforming the prior works.