Bootstrap The Original Latent: Learning a Private Model from a Black-box Model
Not on the map. Why?
Abstract (arXiv)
In this paper, considering the balance of data/model privacy of model owners and user needs, we propose a new setting called Back-Propagated Black-Box Adaptation (BPBA) for users to better train their private models via the guidance of the back-propagated results of a Black-box foundation/source model. Our setting can ease the usage of foundation/source models as well as prevent the leakage and misuse of foundation/source models. Moreover, we also propose a new training strategy called Bootstrap The Original Latent (BTOL) to fully utilize the foundation/source models. Our strategy consists of a domain adapter and a freeze-and-thaw strategy. We apply our BTOL under BPBA and Black-box UDA settings on three different datasets. Experiments show that our strategy is efficient and robust in various settings without manual augmentations.
Key facts. Cited by 1 paper in the map (data through 7 Oct 2026). Builds on 29 of its 40 references that are in the map. First posted on arXiv 7 Mar 2023.
Builds on
29 of 40 references.
References · 29
- 2021Learning Transferable Visual Models From Natural Language Supervisioncited by 33,456
- 2015U-Net: Convolutional Networks for Biomedical Image Segmentationcited by 19,604
- 2022Training language models to follow instructions with human feedbackcited by 15,712
- 2019Momentum Contrast for Unsupervised Visual Representation Learningcited by 7,482
- 2018MobileNetV2: Inverted Residuals and Linear Bottleneckscited by 4,773
- 2020Bootstrap your own latent: A new approach to self-supervised Learningcited by 4,712
- 2018Encoder-Decoder with Atrous Separable Convolution for Semantic Image Segmentationcited by 3,210
- 2020Exploring Simple Siamese Representation Learningcited by 2,853
- 2014Unsupervised Domain Adaptation by Backpropagationcited by 1,906
- 2020Understanding Contrastive Representation Learning through Alignment and Uniformity on the Hyperspherecited by 1,393
- 2020Do We Really Need to Access the Source Data? Source Hypothesis Transfer for Unsupervised Domain Adaptationcited by 844
- 2018ADVENT: Adversarial Entropy Minimization for Domain Adaptation in Semantic Segmentationcited by 623
17 more
- 2021The Medical Segmentation Decathloncited by 620
- 2022Image as a Foreign Language: BEiT Pretraining for All Vision and Vision-Language Taskscited by 510
- 2021Domain Adaptation for Medical Image Analysis: A Surveycited by 351
- 2021ERNIE 3.0: Large-scale Knowledge Enhanced Pre-training for Language Understanding and Generationcited by 258
- 2022Black-Box Tuning for Language-Model-as-a-Servicecited by 254
- 2019REFUGE Challenge: A Unified Framework for Evaluating Automated Methods for Glaucoma Assessment from Fundus Photographscited by 226
- 2016Multivariate mixture model for myocardium segmentation combining multi-source imagescited by 133
- 2020Test-Time Adaptable Neural Networks for Robust Medical Image Segmentationcited by 102
- 2021Source-Free Domain Adaptive Fundus Image Segmentation with Denoised Pseudo-Labelingcited by 61
- 2020Collaborative Unsupervised Domain Adaptation for Medical Image Diagnosiscited by 60
- 2017Unsupervised Reverse Domain Adaptation for Synthetic Medical Images via Adversarial Trainingcited by 57
- 2020Source-Relaxed Domain Adaptation for Image Segmentationcited by 52
- 2021DINE: Domain Adaptation from Single and Multiple Black-box Predictorscited by 51
- 2021Unsupervised Domain Adaptation of Black-Box Source Modelscited by 32
- 2022Rethinking Alignment and Uniformity in Unsupervised Semantic Segmentationcited by 10
- 2023Aggregation of Disentanglement: Reconsidering Domain Variations in Domain Generalizationcited by 8
- 2023FTSO: Effective NAS via First Topology Second Operatorcited by 3
Built on by
Citing papers per year: 2023: 1.
1 paper.