A Prelude to NeurIPS 2023 — Spotlight & Exploring Generative AI Frontiers
One thing to look forward to in December (among other things) is the NeurIPS conference. Like previous years there are lots of good papers, tutorials, workshops, posters and discussions — one wants to be everywhere ! Of course, the laws pf physics kick in and we have to choose. My focus this year is, of course, Generative AI with clarity on Guardrails, Augmentation, …. Two of my relevant blogs — Generative AI Guardrails [here] and AGI/LLM Evaluation [here].
And, the NeurIPS 2023 is so advanced that they have orals on 5D Vision — forget about 3D and 4D ! (Sorry, couldn’t resist)
This blog is the prelude before the conference, I plan to publish the aftermath (Published here : Unboxing NeurIPS 2023) , followup with pointers to talks and summary — in my github https://github.com/xsankar/Awesome-AGI in the appropriate repository
[Update 12.18.23] Churned through 3584 papers ! Extremely, extremely rewarding — just reading the abstracts. Have the (not so) short list. Working on the blog and the summary GitHub.
1The Essentials
1.1. LLM App Development — Tutorial
Mon 11 Dec 11:45 a.m. PST — 2:15 p.m. PST : The must attend tutorial by Andrew Ng !! I don’t know how the experience would be for remote attendees. And I have a meeting that I can’t sneak out of ;o(
With a statement like this, how can we not attend ? “We will share best practices for integrating LLMs into more complex software systems, evaluating and continually improving their performance, and enhancing their safety”
- Another must attend, even though it is 6:45 AM on a Saturday !
- I have high hopes for this workshop
1.2. LLM theory/Explanations
- An important topic, to start off the conference !
- Tue 12 Dec 8 a.m. PST — 8:45 a.m. PST Oral 1B
- I have been keeping an eye on the Decoding Trust paper
- Should be interesting
1.3. LLM Evaluation & Benchmarks
- https://neurips.cc/virtual/2023/poster/73431
- Poster : https://neurips.cc/media/PosterPDFs/NeurIPS%202023/73431.png?t=1699498575.2411695
- Paper : https://openreview.net/forum?id=vTrRq6vCQH
1.4. Threats & Vulnerabilities
- https://neurips.cc/virtual/2023/workshop/66550
- Workshop Website : https://neurips2023-bugs.github.io/
- It has a picture from my favorite Eastwood movie !
1.5. Data
- https://neurips.cc/virtual/2023/competition/66587
- Competition Page : https://big-ann-benchmarks.com/neurips23.html
1.6. Competitions
There are a few interesting competitions. The focus of the sessions is the lessons learned — good to understand.
This year I missed them — too many things to do; but plan to attempt next year ! Just now added a calendar notification on July 15 2024.
Fri 15 Dec : Large Language Model Efficiency Challenge
- Very relevant, in these days of huuuge LLMs !
Sat 16 Dec : The Trojan Detection Challenge
- Should be interesting. There is also a poster
2The Poster Sessions
- I haven’t yet caught up with the poster sessions. There are tons of papers on each day — the navigation takes time. For each poster there is the abstract, the 5 min video, the poster itself and the OpenRiview paper — lots to read!
- I will write more in my aftermath blog
Starting My list:
- TrojLLM: A Black-box Trojan Prompt Attack on Large Language Models https://github.com/UCF-ML-Research/TrojLLM; poster https://neurips.cc/media/PosterPDFs/NeurIPS%202023/71224.png?t=1697497280.3117414
3The esoteric — Generative Horizons: Bridging the Gap between Imagination and Reality at NeurIPS 2023
There are some very interesting futuristic sessions. I will probably watch the videos and read the summary later.
1st/2nd are on Dec 16 & the 3rd/4th on Dec 15
- https://neurips.cc/virtual/2023/workshop/66522
- Workshop website : https://mathai2023.github.io/
- https://neurips.cc/virtual/2023/workshop/66535
- Website : https://sites.google.com/view/infocog-neurips-2023
What did I miss ? for the LLM world ?
