Black-box attack:黑盒攻击(对模型不了解,对训练集不了解或了解很少) Query-based attack:查询攻击(通过提供图片给模型,并根据模型反馈信息对图片进行进一步调整) Transfer-based attack:迁移攻击(通过攻击surrogate model,计算出某张图片的扰动,再将扰动后的图片直接攻击victim model) 2.2 Transfer adversarial attack C...
Learning Transferable Features with Deep Adaptation Networks - 本文作者是清华大学的 Mingsheng Long 也是Domain Adaptation 的专家,可从 Google Scholar 上看出他的近乎所有文章都在研究这个问题。这篇文章包括作者后续的文章里都用到了一个叫作 max mean discrepancies (MMD) 定义为 Source Target Domain 的 feature...
We demonstrate that the simplepre-trainingtask of predicting which caption goes with which image is an efficient and scalable way to learn SOTA image representations from scratch on a dataset of 400 million (image, text) pairs collected from the internet. After pre-training, natural language is ...
CtrlFormer: Learning Transferable State Representation for Visual Control via Transformer CtrlFormer ICML22 jointly learns self-attention mechanisms between visual tokens and policy tokens among different control tasks, where multitask representation can be learned and transferred without catastrophic forgetting...
We propose an end-to-end multi-stage encoder-decoder network for learning the residuals of morphing process to detect attacks. Leveraging the residuals, we learn an efficient classifier using cross-entropy loss and asymmetric loss. The use of asymmetric loss in our approach is motivated by ...
To learn generalizable prompts 所以把 prompt 设计为 instance-conditional 的。 Example prompt structure 为prompt 加上一个跟当前图像相关的特征以提高泛化性能。具体来说,先用 Image Encoder 计算当前图像的 feature,然后通过一个 Meta-Net 把 feature 映射到 prompt 的特征空间,加到 prompt 上面。
20181212 arXiv Learning Transferable Adversarial Examples via Ghost Networks Use ghost networks to learn transferrable adversarial examples 使用ghost网络来学习可迁移的对抗样本 20181211 arXiv Adversarial Transfer Learning A survey on adversarial domain adaptation 一个关于对抗迁移的综述,特别用在doma...
it is important to verify thatFashionCLIPdoes not only learn a dataset (e.g., an “Armani collection”), but genuine transferable concepts, such as “skirt”, “sleeves”, etc. Taking inspiration from CLIP, our two initial benchmarks will test howFashionCLIPgoes from text to image, and ...
This repository contains the code for paperLearning Transferable Adversarial Examples via Ghost Networks. In this paper, we propose Ghost Networks to efficiently learn transferable adversarial examples. The key principle of ghost networks is to perturb an existing model, which potentially generates a huge...
PromptFL: Let Federated Participants Cooperatively Learn Prompts Instead of Models -- Federated Learning in Age of Foundation Model 作者同时也提到,上述这些研究要么假设用户设备上存在大模型,这在资源受限的跨设备环境下可能不可行,要么由于依赖需要发送敏感数据的云大模型API而面临隐私问题,因此目前联邦+大模型...