The DeACL framework consists of two stages. In the first stage, DeACL performs standard self-supervised learning (SSL) to obtain a non-robust encoder. In the second stage, the pretrained encoder acts as a teacher model, generating pseudo-targets to guide supervised adversarial training (AT) on a student model.

2022年7月22日 · With this said, this work discards prior practices of directly introducing AT to SSL frameworks and proposed a two-stage framework termed Decoupled Adversarial Contrastive Learning (DeACL). Extensive experimental results demonstrate that our DeACL achieves SOTA self-supervised adversarial robustness while significantly reducing the training ...

Extensive experimental results demonstrate that our DeACL achieves SOTA self-supervised adversarial robustness while significantly reducing the training time, which validates its effectiveness and efficiency.

2023年3月9日 · 本文提出deacl，一种将对抗训练和自监督学习分为两个阶段的框架。首先进行非鲁棒的ssl以获取预训练编码器，然后通过伪监督at优化编码器，提高模型的健壮性。这种方法减少了直接结合ssl和at带来的计算复杂度增加问题。

During training, to generate adversarial examples, we use 10-step l∞ PGD attack with ε = 8/255 and the whole encoder fθ with linear classifier will be trained and updated for 25 epochs. The learning rate schedule is the same as that of SLF. Setup for DeACL at first stage.

[TMLR'24] Official code for "ProFeAT: Projected Feature Adversarial Training for Self-Supervised Learning of Robust Representations". - ProFeAT/train_deacl.sh at main · val-iisc/ProFeAT

2023年3月2日 · The DeACL framework consists of two stages. In the first stage, DeACL performs standard self-supervised learning (SSL) to obtain a non-robust encoder. In the second stage, the pretrained encoder acts as a teacher model, generating pseudo-targets to guide supervised adversarial training (AT) on a student model.

Tandem Repeats Finder (TRF) 是一个用于分析 DNA 序列中串联重复序列的程序。串联重复序列是指 DNA 中两个或多个相邻的、近似重复的核苷酸模式。TRF 可以帮助用户定位和展示这些重复序列。用户只需提交一个 FASTA…

2019年4月21日 · trf软件是基因组注释中常用于检测序列中串联重复序列的软件，无需安装，使用简单方便。 1. 重复序列分为串联重复序列和散在重复序列（转座子）； 串联重复序列又包含卫星序列 >...

The proposed DeACL has two advantages: (a) enabling flexible configuration for the two sub-problems; (b) requiring much fewer computation resources. Extensive experiments demonstrate that DeACL achieves SOTA robustness while significantly reducing the training time.