In the 18th session of #MultimodalWeekly , we have two exciting speakers who will share their work in LLM evaluation and medical AI.

My second paper on fair PCA accepted to @NeurIPS 2023 (and my first NeurIPS!)

“Fair Streaming Principal Component Analysis: Statistical and Algorithmic Viewpoint”
Junghyun Lee, Hanseul Cho, Se-Young Yun, Chulhee Yun
KAIST AI

Stay tuned for our camera-ready version.

🧵(1/4)

I'll not be at ICML this year, but my co-first author Junsu Kim will present MV-MWM! Please our poster session at Tuesday.

Also, Junsu is looking for a potential internship or visitor position, please contact him if you're interested in working with him!

I will present Multi-View Masked World Models (MV-MWM) for visual robotic manipulation.

Please visit #ICML2023 poster session at 7/25 (Tue) 2:00-3:30pm!

Exciting to share our paper titled 'Generalizable Lightweight Proxy for Robust NAS against Diverse Perturbations' in @AdvMLICML #ICML2023 with
Minseon Kim Sung Ju Hwang.

Please come and check our paper on Friday!
More information in the thread 💻 (1/6)

Are open-sourced LLMs really good? 👀

We introduce FLASK🧪, a fine-grained evaluation based on skill sets! Even SOTA open-sourced LLMs such as LLaMA2 Chat 70B lag behind proprietary LLMs for some abilities. 🤯

Paper: arxiv.org/abs/2307.10928
Demo: kaistai.github.io/FLASK

I'd also like to talk if you're interested in Chain of Thought Fine-tuning, Zero-shot Generalization and Few-shot adaptation! Come visit :)

I'm going to present our Expert Language Model paper at ICML 2023! Come to Exhibit Hall 1 (7.25 2:00PM - 3:30PM) if you're interested in either Instruction Tuning or Expert LMs!

1/ Our paper 'Towards standardizing Korean Grammatical Error Correction: Datasets and Annotation' will be live after 30mins! Please check it out at the poster session 3!

Paper: arxiv.org/pdf/2210.14389…
Work led by: @soyoungyoon97
Thread below. 👇

at KAIST School of Computing

New fact checking dataset alert 🚨 AmbiFC: Fact-Checking Ambiguous Claims with Evidence 🤔 arxiv.org/abs/2104.00640…

Introducing 📸SelFee📸, an iterative self-revising LLM empowered by self-feedback generation. 13B SelFee achieves 103% ChatGPT performance in the Vicuna evaluation setting.

Project Blog: kaistai.github.io/SelFee/
Demo: kaistai.github.io/SelFee/demo
Github: github.com/kaistAI/SelFee

🎉Super happy to share that our paper 'FactKG: Fact Verification via Reasoning on Knowledge Graphs' has been accepted to #ACL2023NLP main conference!

Work with Sungjin Park, Yeonsu Kwon, Yohan Jo, James Thorne, Edward Choi from KAIST and Amazon

Paper: arxiv.org/abs/2305.06590

Excited to announce our keynote speaker Jong Chul Ye (bispl.weebly.com), from KAIST AI and his lecture on Diffusion Models for Inverse Problems in Medical Imaging

🧐Looking to pretrain fine-grained visual models with tons of unlabeled and uncurated open-sets?
Check out our #CVPR2023 🇨🇦paper about OpenSSL (Open-Set Self-Supervised Learning)!

📖: arxiv.org/abs/2303.11101
🧑‍💻: github.com/sungnyun/opens…

Welcome to Sangdoo Yun from NAVER, presenting about research activities and #HyperCLOVAX #languagemodel at the first of our 2023 industry networking events

New York University and KAIST AI are building a new NYU-KAIST campus in NYC. as part of this initiative, Alice Oh started NYU-KAIST Talk Series on Language Models. i was honoured to be the first speaker on this series and talked about LLM: youtube.com/watch?v=289wOY…

Scaling 📈 the total # of tasks during instruction tuning has been known to unlock new abilities in LMs. However, we find that an LM trained on a single task outperforms an LM trained on 300+ tasks on unseen tasks 🤯
📝:arxiv.org/abs/2302.03202
1/8

FEVER 6 will be hosted at eaclmeeting this - come join us in Croatia!