Yuxin Wen

@ywen99

Safety Research @OpenAI | PhD @umdcs advised by @tomgoldsteincs

yuxinwenrick.github.io

Joined December 2021

900 Following

675 Followers

Replies Reposts Media

Yuxin Wen
@ywen99
Sep 17, 2024
📢Think removing watermarks is easy? Maybe not this time! Join our NeurIPS competition for a chance to win from a $6,000 prize pool and test your methods at removing our carefully prepared watermarks!
Furong Huang
@furongh
Sep 16, 2024
🚨 Join the Challenge! 🚨 We're excited to announce the NeurIPS competition "Erasing the Invisible: A Stress-Test Challenge for Image Watermarks" running from September 16 to November 5. This is your chance to test your skills in a cutting-edge domain and win a share of our $6000
Yuxin Wen
@ywen99
Jun 17, 2024
We introduce Goldfish Loss, a simple yet effective method for preventing verbatim memorization. By dropping certain tokens when calculating the loss during training, the model avoids memorizing the training data, even under our super stress tests.
Tom Goldstein
@tomgoldsteincs
Jun 17, 2024
LLMs can memorize training data, causing copyright/privacy risks. Goldfish loss is a nifty trick for training an LLM without memorizing training data. I can train a 7B model on the opening of Harry Potter for 100 gradient steps in a row, and the model still doesn't memorize.
Yuxin Wen
@ywen99
Apr 3, 2024
Very interesting paper! Sharing a similar threat model but with a different focus, our recent paper (arxiv.org/abs/2404.01231), also titled Privacy Backdoors 🤗, achieves strong membership inference performance through poisoning pre-trained models.
Florian Tramèr
@florian_tramer
Apr 3, 2024
If you download a pretrained model you have to trust that the developer did not backdoor it. We know backdoors break model integrity. But what about privacy? With Shanglun Feng we introduce 𝐩𝐫𝐢𝐯𝐚𝐜𝐲 𝐛𝐚𝐜𝐤𝐝𝐨𝐨𝐫𝐬: pretrained models that steal your finetuning data! 🧵
1
Yuxin Wen
@ywen99
Dec 5, 2023
I'll be at #NeurIPS2023 next week! Very excited to present Tree-ring (poster session 1, #126) and PEZ (poster session 4, #606)! Feel free to DM me if you would like to chat about ML privacy and security or even just want to grab some hotpot🍲 together!
3
Yuxin Wen
@ywen99
Oct 19, 2023
Jonas is definitely one of the most excellent researchers I have ever worked with. If you're passionate about ML privacy, safety, and efficiency, don't hesitate to reach out!
Jonas Geiping
@jonasgeiping
Oct 19, 2023
Replying to @jonasgeiping
We're also looking to hire PhD students interested in machine learning! You can find more information about joining my group on my webpage: jonasgeiping.github.io

Yuxin Wen

Yuxin Wen’s posts