iiinova — where your digital self keeps evolving

Come to our #NeurIPS #Competition on Erasing The Invisible Workshop on Stress Testing Image Watermarks today, Dec 15, 1:30-4:30pm PST, West Meeting Room 208. erasinginvisible.github.io/workshop.html

030

This is Michael-Andrei Panaitescu-Liess presenting!

030

Mitigation? We show that adaptive methods (e.g., methods that know information about the watermarking scheme) can improve the detection performance under watermarking. 4/4

On the other hand, watermarking reduces the efficacy of training data detection methods. 3/4

100

On the one hand, watermarking reduces the generation of copyrighted text. 2/4

100

In this paper, we showed that watermarking can be a double-edged sword for copyright regulators since 1⃣ it promotes compliance during generation time 2⃣ but can make training time copyright violations harder to detect. 1/4

100

Can Watermarking Large Language Models Prevent Copyrighted Text Generation and Hide Training Data?

Congratulations to Michael-Andrei Panaitescu-Liess, Zora Che, Bang An, Yuancheng Xu, Pankayaraj Pathmanathan, Souradip Chakraborty, Sicheng Zhu, Tom Goldstein for winning the best paper award at the AdvML Frontiers workshop at #NeurIPS2024. Paper: arxiv.org/abs/2407.17417 A short thread 👇

arxiv.org

Large Language Models (LLMs) have demonstrated impressive capabilities in generating diverse and contextually rich text. However, concerns regarding copyright infringement arise as LLMs may inadverten...

180

I hope NeurIPS and the broader academic community take this as a wake-up call to address the biases and systemic issues that enable such comments to go unchallenged. We must do better. 16/n

020

Racism has no place in academia, and incidents like this tarnish the principles of inclusion and respect that we, as a global research community, should uphold. 15/n

I regret that this happened at NeurIPS. I regret that this happened in my research community—a place I have cherished and contributed to for over 14 years. I regret that this happened at MIT, an institution of excellence and aspiration for many Chinese scholars. 14/n

What is most heartbreaking is that Professor Picard couldn’t even acknowledge something as simple as: “Most Chinese scholars are honest and upright.” Instead, she focused on the singular exception and added, “Of course, with this one exception in this case” in her response. 13/n

2.Even if the student’s school didn’t teach ethics (which is false for schools in China), other sources like family and community often instill strong ethical values. Ignoring this nuance is careless and reinforces stereotypes. 12/n

There are glaring logical flaws in this argument: 1.If the student cheated, why would their excuse about ethics education be taken at face value? A serious scholar would investigate the claim before making it a central part of their argument. 11/n

110

Professor Picard reinforced her remarks by quoting the student’s excuse —that ethics wasn’t taught in their school—and generalized this as a broader issue with Chinese education. This statement is both factually incorrect and deeply offensive. 10/n

This was a generous and high-EQ question, offering Professor Picard an opportunity to reconsider or clarify her comments. Unfortunately, she doubled down instead. 9/n

Are you calling out the student’s nationality because you find most Chinese scholars honest, and the fact that the cheating student was Chinese is rare? Is that why you emphasized nationality? 8/n

What made this incident worse was how it unfolded during the Q&A session. A Chinese attendee asked a professional and thoughtfully articulated question. She began by thanking Professor Picard for her talk and posed this question: 7/n

This needs to change. Asians, like everyone else, have the right to speak out and demand accountability when racism occurs. We will ensure that being racist against Asians has consequences, including here, Professor Picard. 6/n

This choice perpetuates harmful stereotypes about Chinese scholars and reflects a broader bias against Asians, often rooted in the assumption that we “work hard, avoid conflict, and don’t push back.” 5/n

First, it was entirely unnecessary to mention the student’s nationality when discussing an incident of cheating. The point about academic integrity could have been made without emphasizing nationality. Yet, Professor Picard chose to highlight it. 4/n

I deeply respect Professor Picard’s scholarship and contributions to the field. However, her comments during the talk reflected a deeply troubling and racist view of Chinese scholars. This was not just inappropriate but also profoundly disheartening. 3/n

After learning more, I feel compelled to address what I witnessed during an invited talk at NeurIPS 2024 by Professor Rosalind Picard. 2/n

140

Furong Huang @furongh.bsky.social · 07/12/2024

I saw a slide circulating on social media last night while working on a deadline. I didn’t comment immediately because I wanted to understand the full context before speaking. x.com/xwang_lk/sta... 1/n

x.com

180

🌟 Excited to be in Vancouver for #NeurIPS2024 next week! 🌟 Let’s connect about: - Research 🧠: 🔍 watermarks 🤖 LLM alignment & adversarial robustness 🧩 Multi-agent learning & causality in recommendations 📊 Difficulty profiling for benchmarking LLM performance - PhD openings at UMD 🎓

0100

As someone who calls herself an AI explorer, always keeping up with the latest technology and trends, today an incredibly engaging 15 min conversation between ChatGPT and a 5 year-old was an eye-opener. AI has quietly transformed our everyday lives in ways we may not know yet. 3/3

000

AI even taught him about black holes, the speed of light, and how bacteria can be both helpful and harmful. The best part? It kept him curious, engaged, and learning in ways I’d never think to explain so simply. 🧠✨ 2/3

110

Leo had the most fascinating chat with AI today! He said things I’ve never heard him say before, like: 🌍 “I live on Earth.” 🐙 Asking if the kraken is real. 🚗💨 “Can Sonic run faster than a car?” ☀️🔥 Comparing meteors to the Sun’s heat. 1/3

In such cases, acknowledging the validity of the question is important. More importantly, proposing a solution or fix strengthens the rebuttal and demonstrates the authors’ commitment to improving the work. And it will make the rebuttal stronger. 2/2

010

Assuming the reviewer is not malicious, the shared goal between authors and reviewers is to improve the paper and benefit the research community. When faced with a challenging question, it is often a sign that the reviewer has raised a valid and important point. 1/2

100

10/n I hope these tips can help make the task a bit more manageable. Best of luck, and I’m rooting for your success! 💪

000

9/n …concern and emphasizing its resolution in your rebuttal. •Understanding the reviewer’s intention and the underlying reasons for their comments is key to crafting an effective response. Writing a rebuttal is a challenging process that requires balancing details with the big picture.

100

8/n 6.Balance Details with the Big Picture: •Address every small concern to ensure reviewers don’t reject your paper for overlooking specific points. •Simultaneously, prioritize the big picture: Identify the primary reason a reviewer might give your paper a low score. Focus on addressing that …

100

7/n •Rebut or Explain: Explain why you did (or didn’t) take a particular approach, providing logical & well-supported reasoning. •Provide Solutions: If valid, acknowledge reviewer’s point and offer a solution. This may involve proposing changes or even conducting additional experiments if feasible.

100

6/n 5.Handle Tricky Questions Strategically: For complex or challenging questions, consider the following approach: •Acknowledge or Clarify: Start by acknowledging the reviewer’s concern or asking for clarification if needed to demonstrate understanding.

200

5/n 4.Remind and Address: •Highlight positive feedback from the reviewers to reinforce their favorable opinions. •Address all negative points directly and thoroughly so the AC has no reason to reject your paper.

100

4/n 3. Be Polite but Firm: •Maintain a professional and respectful tone. •Make your claims clear and confidently defend your position.

100

3/n •Make your rebuttal easy to follow by organizing it into bullet points, with each addressing one specific concern. •Use BLUF (Bottom Line Up Front): Begin each bullet or paragraph with a concise summary of your main point, followed by supporting details.

100

2/n 2.Keep It Clear and Concise: The reviewing committee, including reviewers and ACs, has very limited time. They likely won’t dig into every detail. This makes it crucial to: •Show that you’ve taken the time to address concerns with additional experiments or analyses.

100

1/n 1.Adopt the Right Mindset: Remember, your rebuttal is primarily for the reviewer and the Area Chair. The AC will read both the reviews and your rebuttal to make the final decision. Your goal is to convince them that your paper addresses all concerns raised and deserves acceptance.

100

I know there are already plenty of tips out there on how to write an effective rebuttal, but I thought I’d share mine as well. I’m not claiming to be an expert or to have a perfect success rate, but I hope these suggestions might be helpful for anyone who could use them.

1162

To all my students: Thank you for making this journey so meaningful. Watching you grow and succeed has been the greatest reward of my career. 💖 Go out there and shine! ✨

060

Sicheng is the author of AutoDAN: Auto Interpretable Jailbreak of LLMs, Auto Pseudo-Harmful Prompt Generation in LLMs, PerceptionCLIP, Robustness from Out-of-distribution Data via Equivariant Domain Translator, generalization benefit of model invariance, and more. 🌟 Website: schzhu.github.io

schzhu.github.io

Sicheng Zhu

150

Sicheng Zhu: Sicheng’s research focuses on trustworthy machine learning, including robustness and generalization. His recent work focuses on alignment of large language models, including jailbreak attacks and automated red-teaming for false refusals.

100

Yuhang Zhou Personal Website

Author of Concept-level Spurious Correlations in LLMs, Multi-Stage Balanced Distillation under Long-Tailed Data, GFairHint: Individual Fairness in GNNs, and more.) 🌟 Website: tonyzhou98.github.io

tonyzhou98.github.io

100

Yuhang Zhou: Yuhang’s research centers on NLP and computational social science, focusing on developing robust models in large language models (LLMs). He also has extensive experience working on responsbile AI/ML, designing algorithms that are fair and trustworthy. Co-advised with Wei Ai.

100

Yuancheng Xu is a PhD student at the University of Maryland working on trustworthy machine learning.

Yuancheng is the author of GenARM: token-level reward guidance for test time LLM alignment, Shadowcast: poisoning VLMs, ELBERT: Adapting Static Fairness to Sequential Decision-Making, Decision Boundary Dynamics for Adversarial Robustness, and more. 🌟 Website: yuancheng-xu.github.io

yuancheng-xu.github.io

Yuancheng Xu

100