(special thanks to Chris Kanich for his guidance and support)
We enjoyed and were inspired by MSW and Mini-MSW events we attended. After consulting the MSW steering committee, we decided to create our own event in a similar style this year because there is no MSW or mini-MSW this year.
The Midwest Security & Privacy Meeting at Notre Dame will be held on October 26th, 2024. We aim to provide great opportunities for bringing people together for interactive activities (e.g., speed advising and interactive group discussions).
This is a free event open to anyone who works on topics related to security and privacy. The goal is to help connect the security/privacy researchers in the Midwest region for future collaborations and networking. Similar to MSW and mini-MSW events, we de-emphasize presenting past work, and we encourage sharing new research ideas for new collaborations.
If you wish to attend this event, please register for it (free) at https://forms.gle/STATiYZNSZNpCR6j9. Registration will be closed at noon on Tuesday!
Below, please find the tentative agenda for the meeting. All events will be held at the University of Notre Dame. Please note that Notre Dame is in the Eastern Time zone, and all times are in ET.
Disclaimer: This will be a one-time event held in 2024 only. While we are inspired by MSW/mini-MSW and have tried to keep the format similar (so that the momentum of such gathering is continued), we and this event are independent and not affiliated with the past or future MSW/mini-MSW events.
We hope MSW or mini-MSW will resume in 2025 to foster security and privacy research in the Midwest area. For future MSW or mini-MSW events, please check the official MSW website.
| Time (ET) | Event | Location Hayes-Healy (HH) Center |
|---|---|---|
| 10:30AM-11:00AM | Arrival and Registration Please arrive early to pick up your badge. | Hallway outside HH127 |
| 11:00AM-11:10AM | Welcome and Overview Meeting organizers: Taeho Jung, Jaroslaw Nabrzyski, Fang Liu, Fanny Ye | HH127 |
| 11:10AM-12:00PM | Keynote: The Memory Trap: Can AI Truly Forget? Haixu Tang, Indiana University, Bloomington In the rapidly advancing field of large language models (LLMs), the challenge of catastrophic forgetting (CF) is well documented: as these models are fine-tuned on new data, they tend to overwrite prior knowledge, seemingly losing access to previously learned information. Here, I will introduce our recent findings that may complicate this narrative. We devised the Janus attack, demonstrating that fine-tuning of LLM may recover the seemingly “forgotten” information, including personally identifiable information (PII). Our discovery shows that while models seem to forget at a surface level, the underlying parameters may still retain traces of sensitive information that can be retrieved through targeted attacks based on fine-tuning or in-context learning (ICL). The challenge thus lies in understanding when LLMs truly forget versus when information is simply dormant within their neural architecture. I will further discuss the implications of our findings, in particular about privacy and the ethical deployment of AI systems, prompting a deeper examination of how memory is encoded and how it can truly be erased from LLMs.  Dr. Haixu Tang is currently a Professor in Department of Computer Science and the Director of Data Science Academic Programs in the Luddy School of Informatics, Computing and Engineering at Indiana University Bloomington. His researchinterests include Computational Biology, Privacy-enhancing technologies and AI/ML. He received the NSF CAREER Award in 2007, the IU Outstanding Junior Faculty Award in 2009, the PET award for his work in genome privacy in 2011, and the RECOMB Test-of-Time Award in 2013. He is one of the organizers of the iDASH Genome Privacy Workshops. | HH127 |
| 12:00PM-1:30PM | Lunch Lunch is on your own. There are a number of dining options on campus and on Eddy Street Commons (5-minute walk away) | |
| 1:30PM-2:40PM | Break-out Discussion Sessions We will have 6 rooms with group discussion sessions for people with the same interests to spread awareness of each other’s research for collaboration and networking. AI/ML: HH127, chaired by Fang Liu Privacy-Enhancing Technologies: HH125, chaired by Binghui (Alan) Wang Software Security: HH117, chaired by Joanna Cecilia da Silva Santos System Security: HH129, chaired by Jarek Nabrzyski Cryptography: HH229, chaired by Changhao Chenli Usable Security/Privacy: HH231, chaired by Sid Stamm | HH117, HH125, HH127 HH129, HH229, HH231 |
| 2:40PM-3:00PM | Coffee Break Coffee and light snacks will be provided. | Hallway outside HH127 |
| 3:00PM-4:30PM | Speed Advising Students will meet with faculty/staff/postdoc researchers from other universities for 10-15 minutes. Mentors will stay in HH129, and students will come to them. When not meeting a mentor, students are encouraged to network with each other in the hallway outside HH127. | HH129 |
| 4:30PM-4:40PM | Concluding Remarks Meeting organizers: Taeho Jung, Jaroslaw Nabrzyski, Fang Liu, Fanny Ye | HH127 |