Huan Sun (OSU)
@hhsun1
Followers
3,934
Following
506
Media
24
Statuses
538
Associate Professor (with Tenure) in CSE, endowed CoE Innovation Scholar, The Ohio State University (NLP and Data Mining)
The Ohio State University
Joined March 2012
Don't wanna be here?
Send us removal request.
Explore trending content on Musk Viewer
Elon
• 931256 Tweets
#गल_काटे_सो_कुफर_कसाई
• 170666 Tweets
Michigan
• 161554 Tweets
#UFC307
• 127040 Tweets
Vandy
• 120254 Tweets
Watch Sant RampalJi YouTube
• 119253 Tweets
Tennessee
• 112383 Tweets
Dodgers
• 107418 Tweets
Vanderbilt
• 84022 Tweets
Dark MAGA
• 81758 Tweets
トッキュウジャー
• 80796 Tweets
#BUS_KnockKnockKnock_Korat
• 74482 Tweets
Nico
• 73472 Tweets
Pereira
• 71472 Tweets
Bama
• 71222 Tweets
Ohtani
• 70989 Tweets
ドジャース
• 66207 Tweets
プリキュア
• 49368 Tweets
Khalil
• 47556 Tweets
Arkansas
• 37667 Tweets
DONBELLE ASAP RESURGENCE
• 36863 Tweets
Aldo
• 33358 Tweets
ポストシーズン
• 30748 Tweets
Bautista
• 26629 Tweets
Save Sanatan Dharma
• 26347 Tweets
Pennington
• 22963 Tweets
Ifigenia Martínez
• 18954 Tweets
Machado
• 18888 Tweets
京都大賞典
• 17903 Tweets
Poatan
• 14451 Tweets
大谷さん
• 14355 Tweets
山本由伸
• 11684 Tweets
Pinned Tweet
Thanks
@_akhaliq
for sharing our work. Very proud to introduce my star student
@BoshiWang2
's new work
@osunlp
: Grokked Transformers are Implicit Reasoners: A Mechanistic Journey to the Edge of Generalization
Can transformers reason? Are transformers fundamentally limited in
Grokked Transformers are Implicit Reasoners
A Mechanistic Journey to the Edge of Generalization
We study whether transformers can learn to implicitly reason over parametric knowledge, a skill that even the most capable language models struggle with. Focusing on two
7
49
307
13
79
357
Official now! Excited to be promoted to Associate Professor with tenure
@OhioStateCSE
@OSUengineering
Sincerest thanks to my advisors, mentors, collaborators, and colleagues who wrote support letters for me! A warm thank you to students in
@osunlp
and to my family
@ysu_nlp
. ☺️
24
2
318
The Department of Computer Science and Engineering () at The Ohio State University
@OSUengineering
has *20* tenure-track faculty positions open 👏. Retweet is appreciated!
@AcademicJobs
@ajobsonline
@csfacultyjobs
Check more:👇
9
159
289
Claude 3 Opus gets better than GPT-4 on chemistry!
We tested it on our recently released benchmark (SMolInstruct: ) that has a variety of tasks, including name conversion, property prediction, molecule captioning, molecule generation, forward synthesis,
Large Language Models for Chemistry: Chemistry plays a crucial role in many domains like drug discovery and material science. While LLMs exhibit remarkable capabilities on various NLP tasks, existing work shows their performance on chemistry tasks is discouragingly low.
4
31
110
3
28
125
Large Language Models for Chemistry: Chemistry plays a crucial role in many domains like drug discovery and material science. While LLMs exhibit remarkable capabilities on various NLP tasks, existing work shows their performance on chemistry tasks is discouragingly low.
4
31
110
Finally got a bit time to introduce our recent work on learning to generate adversarial suffixes: : Our generative model, named AmpleGCG, captures the distribution of adversarial suffixes given a harmful query and enables rapid generation of hundreds of
4
35
102
Honored to receive a
@GoogleAI
Research Scholar Award 2022 to explore pre-trained language models for reasoning. Many thanks to my sponsors (esp.
@congyu
@JiaMShen
) and students at
@osunlp
(esp.
@BoshiWang2
& Xiang Deng).
4
3
94
Chain-of-thought prompting has been shown very effective for large LMs like GPT-3 & PaLM to solve tasks that require multi-step reasoning. Our
#EMNLP2022
work investigates how to iteratively prompt much smaller pre-trained LMs (e.g., BART-large) to develop a chain of thought.👇👇
1
19
92
Join us for the Workshop on Natural Language Processing for Programming
@Nlp4Prog
on Aug 6th (9AM-6PM ET) at
#ACL2021NLP
, featuring a stellar multi-disciplinary lineup of speakers from PL/NLP/HCI/Robotics. Check out the program here
@osunlp
1
22
92
Hiring multi Ph.D. students this cycle in areas:
#LLM
train/eval, trustworthiness of LLMs incl. privacy & safety, LLM for biomedicine/chemistry. see below for representative work. I won't be
#EMNLP23
, but pls talk to my (former) students there
@xiangyue96
@BoshiWang2
@RonZiruChen
5
23
91
Congratulations! Our acceptance rate this time is approximating 100% 😂
Congratulations to the OSU NLP students and faculty who contributed to 6 papers at ACL and CVPR 2022! Topics range from question answering, semantic parsing, language modeling, and vision-language navigation.
1
3
43
2
0
71
How to augment language models
#languagemodel
with the ability to reason over multiple, possibly hybrid, contexts (e.g., for question answering
#QuestionAnswering
)? Check out our work (ReasonBert: Pre-trained to Reason) at
#EMNLP2021
#NLProc
Link:
6
7
69
Our work that studies grokked transformers on reasoning and their generalization behaviors is accepted to
#NeurIPS2024
@NeurIPSConf
. Coincidentally, there is an exciting panel moderated by
@murraycampbell
(thanks) on AI reasoning that discussed about the definition of reasoning
Thanks
@_akhaliq
for sharing our work. Very proud to introduce my star student
@BoshiWang2
's new work
@osunlp
: Grokked Transformers are Implicit Reasoners: A Mechanistic Journey to the Edge of Generalization
Can transformers reason? Are transformers fundamentally limited in
13
79
357
3
10
72
A great pleasure for
@osunlp
to host an NLP rising star
@ManlingLi_
from UIUC. Thanks for the great talk on multimedia information extraction! Looking forward to possible collaborations! Manling is on the faculty job market; Check out her profile if your department is hiring!
1
6
60
Super honored to receive the
#NSFCAREER
award to study interactive QA & its applications in the clinical domain:
3
1
58
This seems very related to our recent Grokked Transformer paper: , where we find that transformers can induce and apply latent rules (which can be regarded as functions in this case) after grokking with different levels of generalization for different
New paper, surprising result:
We finetune an LLM on just (x,y) pairs from an unknown function f. Remarkably, the LLM can:
a) Define f in code
b) Invert f
c) Compose f
—without in-context examples or chain-of-thought.
So reasoning occurs non-transparently in weights/activations!
34
220
2K
2
4
56
We
@osunlp
had a blast at the Midwest Speech and Language Days Symposium! Many thanks to colleagues at UMich for organizing this great event!
P.S. How could we miss our wonderful
@BoyuGouNLP
in the first picture? Must take a second one, though with a smaller group!
1
6
52
🏆🏆🏆CONGRATS to
@osunlp
students
@yugu_nlp
@XiangDeng1
@xiangyue96
@BoshiWang2
& wonderful collaborators for winning 1 Outstanding Paper Award and 2 Honorable Mentions!
#ACL2023NLP
#NLProc
I won't be at ACL, as i've got a lovely infant to take care of! Happy to chat online! 😁
📢 Super proud that OSU NLP group has three papers receiving paper awards at
#ACL2023NLP
-- 1 Outstanding Paper and 2 Honorable Mentions:
1
5
45
3
3
48
Wow, 128 H100 GPUs are going to be added to
@osc
the Ohio Supercomputer Center! The brewing news is finally out! Consider joining or collaborating with OSU as your next step!😜
Advanced GPU compute is vital to cutting-edge AI. Thrilled that 128 H100 GPUs are being added to the Ohio Supercomputer Center, a major and timely investment by the State of Ohio and
@OhioState
. The new cluster will substantially propel the research at
@osunlp
and OSU at large.
1
6
64
2
0
48
Introducing TableLlama and TableInstruct: the first open-source generalist
#LLMs
and instruction tuning dataset for a variety of table-based tasks
#tables
. See details below and here
We construct TableInstruct, a comprehensive table-based instruction tuning dataset that covers a variety of real-world tables and realistic tasks. We include 14 datasets of 11 tasks in total, with both in-domain and out-of-domain evaluation settings.
1
1
12
1
4
46
We were so excited to host
@lileics
yesterday
@OSU
TDAI
@OSUbigdata
! Lei introduced his team's exciting work: ALGO, ToolDec, SALAM, and discussed about LLMs cannot self-correct reasoning by
@jefffhj
& insights on weak to strong learning!
@osunlp
had lots of fun chatting with Lei!
🚀🤖 Excited to drop some serious "Buckeye Brainpower" at
#OSU
! 🌰📚 Join me as I unravel the secrets of "Self-assisting and Cooperative Large Language Models" at 12pm in POM301.
👾 I'll be spilling the beans on how LLMs can write code like it's a poetry slam and LLMs cooperate!
0
3
36
1
5
46
OpenAI o1 vs. grokked transformer on complex comparative reasoning:
Can OpenAI o1 infer John is younger than Rick in this figure?
Still no better than random guess!
Can OpenAI o1 tackle hard reasoning problems? We tested it on the complex reasoning task in our Grokked Transformers paper. It turns out that o1-preview also struggles a lot like earlier LLMs; on the other hand, a grokked transformer can nail it near-perfectly.
14
78
539
4
7
46
We have released a new version of our work on LLMs for chemistry with more in-depth analysis: ! In addition to Claude 3 Opus outperforming GPT-4, several other interesting discoveries:
(1) We use a previous chemical instruction tuning dataset to train same
🌟 News: We've just released a new version of LlaSMol () with some fascinating 🔍findings!
(1) We evaluate Claude 3 Opus on our chemistry instruction dataset SMolInstruct, and it outshines GPT-4 on these chemistry tasks. However, our smaller,
1
13
33
1
9
42
My students/collaborators and I will present two papers in the poster sessions
#ACL2024NLP
on Tue/Wed:
1. "AttributionBench: How Hard is Automatic Attribution Evaluation?" --> Findings In-Person Poster Session 3 (August 13, 12:15 pm).
led by
📢📢Glad to share our recent work: "AttributionBench: How Hard is Automatic Attribution Evaluation?" accepted at ACL2024 Findings!
📰Paper:
🤗Dataset:
🔗Website:
4
29
73
0
11
43
Excited to win the Best Paper Award from
#BIBM2021
(IEEE Conf. on
#Bioinformatics
&
#Biomedicine
)
#bionlp
#clinicalnlp
#nlproc
Plz check out our dataset for testing the generalizability of Clinical
#QuestionAnswering
models: Code:
How to improve and test the generalizability of Clinical QA models? Check out our paper, which received the Best Paper Award of
#BIBM2021
:
1
3
24
2
3
42
Consider submitting to the Natural Language Interface Workshop
#acl2020nlp
(deadline 04/20). Speakers: Percy Liang & Monica Lam (Stanford), Joyce Chai (UMichigan),
@LukeZettlemoyer
(UW), Imed Zitouni (Google) & more.
#NLProc
#NLIworkshop
@yusuOSU
@scottyih
@AhmedAwadallahMSR
1
13
42
LLMs like
#ChatGPT
have got impressively high accuracies on complx reasoning tasks like
#GSM8K
, but do they really have a deep understanding of the truth? can they actually defend their own correct answer? Our paper finds NO! They can be easily misled by user's invalid arguments.
Are LLMs reasoning based on deep understandings of truth and logic? Can LLMs hold & defend their own "reasoning"?
Our
#EMNLP23
findings paper () explores testing LLMs' reasoning by engaging them in a debate that probes deeper into their understanding.
3
36
111
1
8
39
Proud advisor moment: My very first student Ziyu Yao
@ZiyuYao
at OSU CSE
@OSUengineering
will join the CS department at George Mason University as an Assistant Professor this coming Fall! So rewarding to see students grow to be independent PIs! Many thanks to all collaborators!
I am thrilled to be joining the CS department at George Mason University as an Assistant Professor in Fall 2021! I will continue my research on
#NLProc
and
#AI
. Reach out if you are interested in doing a PhD or collaborating with me!
@GeorgeMasonU
@GMUCompSci
36
16
428
1
1
37
Two papers led by students/alumni from
@osunlp
got selected as Oral by
#CVPR2024
...Big honor! Another piece of great news that came yesterday is
@LingboMo
won the competitive Graduate Research Award of CSE in 2024 (the annual award to the best graduating students). This is the
Excited that we have two papers selected as Oral at
#CVPR2024
(90 orals in total, 0.8%). Congrats to all the students and collaborators, and see you in Seattle!
- BioCLIP: A Vision Foundation Model for the Tree of Life () led by
@samstevens6860
Lisa Wu
-
2
15
112
1
3
38
Continuing our efforts on automatically generating adversarial prompts to red-team LLMs efficiently, we are releasing AmpleGCG-plus: a stronger generative model of adversarial suffixes (Higher ASR in fewer sampling attempts and essentially a shorter time), which is a follow-up to
Thrilled to announce that AmpleGCG has been accepted to COLM and the main takes can be found in the thread below. Very glad to see that our released models have become valuable resources to study LLMs vulnerabilities.
🚀 Today, we're even more excited to introduce AmpleGCG-plus,
3
22
38
0
6
38
@osunlp
TacoBot team is excited to win the third place in the 1st Alexa Prize TaskBot Challenge. Many congratulations to the top two performers (GRILLBot from University of Glasgow and Twiz from FCT NOVA) and the other two finalists (QuakerBot from UPenn and GauchoBot from UCSB).
Alexa Prize TaskBot Challenge announced its top three university teams—which addresses AI challenges through the fusion of multiple AI techniques. Learn more about the winning work from teams' from
@UofGlasgow
,
@FCTNOVA
, and
@OhioState
.
#AlexaPrize
2
15
49
2
10
36
📢📢Instruction tuning with hybrid Chain-of-Thought and Program-of-Thought rationales on diverse datasets ➡️an average gain of 13-29% over open-source SoTA at different model scales on 9 math datasets. Check details below👇👇
Excited to introduce our latest math generalist model MAmmoTH 🦣, built through instruction tuning. We proposed hybrid "chain-of-thought" & "program-of-thought" training to supercharge LLMs' math reasoning capabilities. 🦣 beats the open SoTA by 20+% on many datasets like MATH.
9
41
256
0
10
36
One should not simply judge the amount of effort based on the number of examples in the (online/human) evaluation experiments: (1) We spent a HUGE amount of effort developing our online evaluation tool and releasing it online responsibly , with everyone on
@ysu_nlp
@_akhaliq
Thanks
@ysu_nlp
, I really appreciate your clarification and agree that SeeAct incorporates online evaluations of the Mind2Web framework, but only covers a subset consisting 50 examples. I think our intention was to emphasize that the core evaluations and analyses within SeeAct is
1
0
0
1
3
35
Honored to be an endowed College of Engineering Innovation Scholar! All credits go to my students at
@osunlp
, colleagues, collaborators, and mentors. Thank you all so much!
Endowed & distinguished faculty positions are essential to achieving ever-increasing excellence in our missions of teaching, research & societal impact. Learn about the 17 earning such distinction this year:
2
3
30
3
3
35
Dr. William Wang
@WilliamWangNLP
sharing his recent exciting work on enhancing in-context learning with a Bayesian latent variable approach and synergizing LLMs with symbolic solvers, as well as his vision on compositional and collaborative AI! Many thanks for visiting OSU
0
5
35
We were very glad to host
@xwang_lk
Dr. Xin (Eric) Wang yesterday! Eric talked about his recent work on multimodal embodied agents, commonsense reasoning in object navigation, grounding, as well as his vision for future research. Many thanks to OSU TDAI
@OSUbigdata
for the
0
1
34
We finally released SeeAct v0.1.0 (), a multimodal generalist web agent. Kudos to the student leads
@boyuan__zheng
@BoyuGouNLP
, who have been tirelessly working on the project in the past 6 months!
I first tried it on a real task I found very annoying at
🚀Releasing SeeAct v0.1.0, generalist web agents at your fingertips!
💻Github Repo:
SeeAct enables everyone to effortlessly run web agents based on GPT-4V (and more) with just one click. We hope it can help make the complex Internet a more accessible
2
47
178
0
7
32
Excited to attend
#CVPR2024
! I will be at the oral session for MMMU (THU 20 JUN 9:30am: Orals 3B Vision, Language, and Reasoning). In addition, happy to talk about agents, reasoning/planning, auto red-teaming, and LLMs/Agents for Chemistry. Recent work on these topics:
🥰Attending
#CVPR2024
and presenting 🏆Award Candidate Paper MMMU!
DM is open, drop me a message if you'd like to chat about
#multimodal
,
#LLMs
,
#evaluation
or
#GenAI
in general!
TUE 18 JUN 3:30pm:
New frontiers for zero-shot Image Captioning Evaluation (NICE)
THU 20 JUN
0
19
94
0
10
33
Web agents now *with eyes*: Entering SeeAct, a multimodal agent based on GPT-4V! With our online eval. tool, SeeAct can successfully complete up to *50%* of tasks on live websites, if oracle action grounding is given! led by
@boyuan__zheng
@BoyuGouNLP
@ysu_nlp
See demo&details👇
Generalist web agents may get here sooner than we thought---introducing SeeAct, a multimodal web agent built on GPT-4V(ision).
What's this all about?
> Back in June 2023, when we released Mind2Web () and envisioned generalist web agent, a language agent
19
149
655
0
5
33
Second call for papers! Please consider submitting your work to our workshop on NLP for Programming at ACL 2021 (
@Nlp4Prog
@aclmeeting
)! w/ Royi Lachmy,
@ZiyuYao
@gregd_nlp
, Milos Gligoric,
@jessyjli
@BoredRayMooney
@gneubig
@yusuOSU
@hhsun1
@rtsarfaty
Interested in using natural language processing💬 to assist computer programming💻? Consider submit to our workshop on NLP for Programming (NLP4Prog) at
@aclmeeting
2021! 🤩🤩🤩Call for paper at Deadline: April 26, 2021🗓️
2
20
45
0
16
32
Honored to be invited! Please consider submitting your great work to the workshop! Its going to be a wonderful event… Plus New Orleans would be great to visit in Dec! :-)
Excited to announce the amazing speakers who will share their work & perspective on Table Representation Learning
@TrlWorkshop
on 2 Dec at
@NeurIPSConf
-- most in-person 🔥!
>> Submissions due in 2 weeks! CFP and info here: .
More updates follow soon!
0
10
23
0
3
30
I will give a short talk about how we developed a bot that won 3rd place in the first Alexa Prize TaskBot challenge
@AmazonScience
at
#NAACL2022
#DLG4NLP
workshop () at 3:45PM PT. If you are curious or will participate in future events, happy to chat more!
2
2
31
Check out our
#NAACL2024
tutorial: Combating Security and Privacy Issues in the Era of Large Language Models. Happening now.
Muhao, Chaowei, Huan, Leon, Anima and I are presenting a tutorial on
Combating Security and Privacy Issues in the Era of Large Language Models
at
#NAACL2024
in room Don Alberto 4.
@muhao_chen
@ChaoweiX
@hhsun1
@LeonDerczynski
@AnimaAnandkumar
0
10
62
0
6
30
Two pieces of our
@osunlp
work got interviewed recently, by
@wosunews
,
@yahoonewsUK
,
@OhioStateNews
etc: (1) Generalist web agent, check
@ysu_nlp
's interview : (2) LLMs can't defend its (correct) answer, showing limit. &safety risks
0
6
29
😂😂Missed this earlier: "This paper [AmpleGCG: ] flew under the radar because it’s from a less known PhD student; but it may be the best gradient-based attack right now.” (from ) uh-oh, it hurts; we will try to be better known in the
Alert 🚨🚨：
We have released our adversarial suffixes generators (AmpleGCG-series models) on Huggingface ().
Our generated adversarial suffixes on AdvBench and MaliciousInstruct can be accessed via Google form ().
In light of ethical
0
18
40
1
6
29
2022 ACM SIGMOD Research Highlight Award: As
@SIGMODConf
2022 gets started, I realize that the
@sigmod
record of research highlights (Mar issue) has been released! Our paper introduced below on Table Understanding was honored to be selected!👇👇🏆🏆
2
2
26
I was often asked "how's Columbus" by prospective students & faculty members. It's *surprisingly great*! (I was mainly on the West Coast before.) Just googled some fun facts that you may not know about Columbus: Come join us!
@osunlp
@OSUengineering
0
2
26
Wondering about what makes Chain-of-Thought prompting work? An empirical study of what aspects matter of a CoT rationale 👇
#NLProc
#AI
Summary: Valid reasoning steps in demonstration might not be as important as one would have imagined! Relevance & coherence are the key aspects.
Chain-of-Thought (CoT) prompting improves large LMs on complex reasoning. But what aspects of the CoT prompt matter? Are valid reasoning steps important?
Check out our paper: ! Work w/
@sewon__min
,
@xiangdeng1
,
@mickeysjm
, You Wu,
@LukeZettlemoyer
,
@hhsun1
11
64
321
0
8
25
Eunsol
@eunsolc
and I will host a "Birds of a Feather" session on Question Answering
#QuestionAnswering
at
#ACL2021NLP
on Aug 2nd (11AM PT / 2PM ET). Please come join us to discuss anything related!!!😀
#NLProc
@osunlp
The
#ACL2021NLP
Academic Inclusion committee (cc
@khyathi_chandu
@lifu_huang
@seirasto
) is glad to announce the following "Birds of a Feather" sessions. We hope to see you come with your questions/comments/ suggestions for the various
#NLProc
research themes.
2
5
33
0
3
25
Please check out our
#acl2020nlp
work: A graph-based framework with model rationales for medical relation prediction. Looking forward to meeting everyone online & more discussions on
#ExplainableAI
, especially for
#ClinicalNLP
tasks!
#NLProc
1/5 Very happy to share our ACL 2020 paper "Rationalizing Medical Relation Prediction from Corpus-level Statistics" with Jennifer Lee, Simon Lin, and
@hhsun1
at
#acl2020nlp
Paper:
#nlproc
1
0
9
0
2
23
#ACL2023NLP
Come join us (
@BoshiWang2
@sewon__min
@XiangDeng1
& others) to discuss one of our Honorable Mentions: "Towards Understanding Chain-of-Thought Prompting: An Empirical Study of What Matters." Session 6, July 12, 09:00-10:30, Room: Frontenac Ballroom and Queen's Quay.
Chain-of-Thought (CoT) prompting improves large LMs on complex reasoning. But what aspects of the CoT prompt matter? Are valid reasoning steps important?
Check out our paper: ! Work w/
@sewon__min
,
@xiangdeng1
,
@mickeysjm
, You Wu,
@LukeZettlemoyer
,
@hhsun1
11
64
321
0
0
22
So, y'know there's an expert-level AGI (bg: ); how can we track the progress towards that? Despite existing multimodal benchmarks, how can we measure *expert-level* perception, knowledge & reasoning abilities? Enter tough
#MMMU
! 👇
🚀 Introducing MMMU, a Massive Multi-discipline Multimodal Understanding and Reasoning Benchmark for Expert AGI.
🧐 Highlights of the MMMU benchmark:
> 11.5K meticulously collected multimodal questions from college exams, quizzes, and textbooks
>
19
184
748
0
2
21
Can LLMs do self-correction (or self-reflection)? Do we want to choose self-correction and the more complex *tree search* over the simpler *generating-and-reranking* strategy? Key findings in our
#ACL2024NLP
paper (camera-ready: ) include: (1) iterative
We have updated our Arxiv with the
#ACL2024NLP
camera-ready version! Besides tree search, our work shares a few insights with recent work on analyzing self-correction (e.g., by
@Dongwei__Jiang
, by
@RyoKamoi
): (1/4)
2
11
28
2
5
21
📢Please check out our workshop this Saturday, featuring an amazing lineup of speakers (Prof. Eneko Agirre
@enekoagirre
Prof. Heng Ji
@elgreco_winter
Dr. Hoifung Poon
@hoifungpoon
Prof. Yizhou Sun
@YizhouSun
) and also quite a few interesting papers!
News: our workshop "Wise-Supervision 2022" (Weak, Indirect & Self Supervision) is taking place this Saturday (2pm London; Nov. 5, 2022) virtually with zoom: . No Registration. Let's gather to see how these leading minds are "cutting the edge" of this area.
0
1
8
1
2
21
This is not one of my main research topics, but I've been enjoying learning/exploring new things. Thanks to my student and collaborators for teaching me! New
#ACL2021NLP
Findings paper: "Differential Privacy for Text Analytics via Natural Text Sanitization".
#DifferentialPrivacy
1
1
20
Happening now.. come check out Yu’s talk on Language Agents!
It's the last day of the conference, and Bangkok has so much to explore. i know, i know.. but if you still have some appetite for a (hopefully thought-provoking) talk, I will talk about web/GUI agents tomorrow at 2 pm in the SpLU-RoboNLP workshop. I will cover:
1. Why the
1
18
70
0
2
19
Great job, everyone
@osunlp
! Many thanks to our collaborators! Topics cover grounding LLMs to real-world environments, understanding Chain-of-Thought prompting, differential privacy for text generation, semantic parsing, KBQA, zero-shot relation extraction.
#NLProc
#ACL2023
0
3
20
One of the exciting efforts
@OhioState
on building foundation models for science: Check out *BioCLIP* led by
@ysu_nlp
and team, a vision FM for biology, with strong zero/few-shot results and their curated TreeOfLife-10M dataset! Also, check out the mentioned Postdoc opportunity!
Introducing BioCLIP: A Vision Foundation Model for the Tree of Life
A foundation model that strongly generalizes on the tree of life (2M+ species), outperforming OpenAI CLIP by 18% in zero-shot classification, and supports open-ended classification over
9
94
440
0
6
20
This is 5-7PM today (Dec 14) at the poster session
#NeurIPS2023
Come join us to chat about Mind2Web, MagicBrush, web agents and multimodality, and attending conferences with a baby!!
#NeurIPS2023
We will present at the 5-7 pm session tmr:
> MagicBrush: A Manually Annotated Dataset for Instruction-Guided Image Editing
> Mind2Web: Towards a Generalist Agent for the Web
Come chat about agents, multimodality, and
1
8
44
1
0
20
Thank you for featuring our TableLlama. More details can be found here:
Meet The Experts: SambaNova is excited to feature TableLlama - a large open-source generalist model for tables based on
@AIatMeta
‘s llama model and trained on the TableInstruct dataset. TableInstruct is a large-scale instruction tuning dataset with diverse, realistic tasks
1
6
23
0
3
19
Fine-tuning
#GPT
(large language model) with
#DifferentialPrivacy
: A simple and effective recipe to generate synthetic data for downstream tasks, to protect the privacy of the original data.
@osunlp
📢📢Want to share your textual data outside your org or team but worry about privacy leakage? Check out our new preprint "Synthetic Text Generation with Differential Privacy: A Simple and Practical Recipe"() 👇👇👇
1
6
37
0
0
19
Booking travels in real life is tedious and a task I bet not many people like. Can LLMs understand your request for flight booking? What about complex ones like this? "If I am leaving on April 23rd, I need a flight after 8PM so I can make it after work. However, if I depart on
0
5
19
Our OSU
#NLProc
group summary for 2022 is here. 🥂🥂Wish everyone a Healthy and Happy 2023!
⚡️OSU
#NLProc
group highlights 2022⚡️
It was truly a year of challenges, but looking back now, it has also been extremely rewarding. It's so joyful to see the growth of the group, and we are very grateful for all the help and support along the way!
Most proud moments:
1
13
31
0
1
18
The fast development of language agents is both exciting and scary! There is a large gap between the speed and scale of their development and our understanding of their safety risks. In what real-world scenarios can language agents be potentially attacked? What attack strategies
🚀 Language agents fueled by LLMs are rapidly advancing. Their capability is being further capitalized by connecting to a wide range of external components such as databases, tools, the Internet, robotic embodiment, etc. However, our understanding of their safety risks lags much
0
17
57
0
8
18
#MMMU
has got an upgrade (with augmented options, vision-only input setting, etc).... See MMMU-Pro 👇👇
🚀 Introducing MMMU-Pro: A more robust version of MMMU
After launching MMMU, we received valuable feedback from the community:
1️⃣ Some questions were answerable without even seeing the images.
2️⃣ Models didn’t always "know" the answer but found shortcuts
3
32
146
0
2
18
Thank you so much for having me! Appreciate all the questions and discussions. Very enlightening. 👍👍👍
0
1
18
HUGE congratulations to Xiang Yue
@xiangyue96
@osunlp
on the very competitive Exemplary Graduate Student Researcher Award (1 out of 21 nominees across the College)! His work studies privacy-preserving
#LLMs
, attributions by LLMs, etc.
#NLProc
#proudadvisor
Congrats to CSE PhD Student Xiang Yue for winning the College of Engineering Graduate Award for Exemplary Graduate Student Researcher! Read story:
0
3
10
0
1
17
So proud of you
@xiangyue96
! it is bitter-sweet for ph.d. advisor to see their students graduate and leave the group. 👍👍😂😂 Wish all graduates the very best on their next chapter!
0
0
17
Together with researchers at Nationwide Children's Hospital, we have done a series of work to enhance clinical note querying and understanding, including deep query suggestion, clinical phrase mining, medical relation prediction, and clinical QA.
#clinicalNLP
#biomedicalNLP
🚨 In our latest paper at ACI Open 👉To improve the medical text search, and overcome current search limitations, we developed and proposed an artificial intelligence (AI) tool “DeepSuggest” to expand queries using medical notes.
@ACI_Journal
#nlp
#EHR
0
1
4
4
0
17
Many thanks to the amazing organizers for making it happen!!!
The 1st edition of the Table Representation Learning workshop at
#NeurIPS22
was a great success 💯!! We've seen the start of an excited community spanning multiple areas. Thanks to everyone involved! We look forward to the next edition: feedback? -> Link in thread! 1/6
1
7
26
0
0
15
Check out the multiple Ph.D. and postdoc positions available at
@TheNingLab
at OSU.
@osunlp
and
@TheNingLab
are long-term collaborators on LLMs for Chemistry and Biomedicine. Consider joining OSU! Retweet appreciated! 🙏🤝
📢We are
#hiring
! NingLab currently (as of May 1, 2024) has multiple openings for Ph.D. students, postdocs, and research staff. If interested, please contact Dr. Ning (ning.104
@osu
.edu). More information is available here:
#ArtificialIntelligence
#AI
#ML
0
5
13
0
3
15
Excited to share our new work on learning semantic parsers from user interaction at
#emnlp2020
check more here:
Thanks to all collaborators!
@ZiyuYao
Yiqi Tang
@scottyih
@yusuOSU
lifelong train your language interfaces? Minimize privacy risks while still able to improve from use? Reduce bootstrapping cost to 1/10? What if one says you can have it all, plus nice theoretical analysis? Check our new
#emnlp2020
paper w/
@ZiyuYao
Yiqi Tang
@scottyih
@hhsun1
0
0
5
0
0
14
Excited to share this work on editing tree-structured data accepted by
#ICLR21
! Thanks to
@ZiyuYao
@frankxu2004
@pengchengyin
@gneubig
. My awesome student
@ZiyuYao
is on the faculty job market. Please check out her profile () if your department is hiring!
Often times in real scenarios we need to *iteratively* edit tree-structured data (e.g., ASTs of computer programs). We are happy to share our
#ICLR21
paper (w/
@frankxu2004
@pengchengyin
@hhsun1
@gneubig
) in this topic:
Code:
1/6
3
21
100
0
0
13
Fantastic talk…Really inspiring and thought-provoking! Thank you!!! I am planning to play part of it in my NLP class this Fall.
In nervous anticipation of my
#ACL2021NLP
keynote, I recorded myself giving my talk, and I've posted that version for people who can't attend the live event. The video has high-quality captions, and you don't particularly need video to follow along:
10
73
363
1
1
11
Happening now. Welcome to check out our work and join the zoom/poster session!
In 10 min, the great Xiang Deng (OSU) will present “Structure-Grounded Pretraining for Text-to-SQL” at
#NAACL2021
! To robustify semantic parsing, we pretrain a text+table representation model using table structure objectives.
📘Paper:
📜Code: coming soon!
1
3
13
0
0
12
Frustrated at a Text2SQL parser being imperfect & always getting errors? See our error correction model
#ACL2023
Two highlights: (1) better to do edits at SQL clause level rather than token level; (2) represent SQL (& edits) as python to better leverage LMs of code!
#NLProc
0
0
9
When asked about what tasks can be more interesting to evaluate LLMs for chemistry,
@JoshuaSchrier
(thank you!) kindly forwarded me this report from the first LLM hackathon for materials and chemistry organized by
@BenBlaiszik
(original tweet:
@hhsun1
Good question: Maybe nobody knows (yet), but some idea may come from: & (& of course the voluminous literature on data extraction)
0
0
5
6
0
11
Super exciting! Congratulations!!!
Super honored and excited to be appointed as Distinguished Assistant Professor of Engineering Inclusive Excellence by
@OSUengineering
! It will support me to continue pushing towards the democratization of AI in research and teaching.
4
4
137
0
0
11
Our paper on learning semantic parsers from user interaction will be presented tonight at 9PM EST in Gather-4F. My awesome student Ziyu
@ZiyuYao
(Presidential Fellowship awardee; EEIS rising star) will be on the job market this year. Please reach out for more discussions!
Our paper will be presented in Gather Session 4F (Nov 17th 9PM EST/6PM PST/2AM UTC +1d) at
#emnlp2020
. Hope to see you there!😉
0
3
11
0
1
10
Many thanks to the organizers for a fantastic job! Looking forward to it!
Excited to share the full program: 🔥.
Beyond the announced talks, we'll have a panel w/
@hhsun1
@eisenjulian
@FrankRHutter
@rasbt
@elgreco_winter
@gneubig
, and 2 poster sessions (sneakpreview of pitches: ).
See you in New Orleans!
1
10
30
0
1
10
Pls consider submitting your work to the Wise-Supervision Workshop
#AKBC2022
: Weak, Indirect and Self Supervision for Knowledge Extraction! Deadline: Oct 15th.
welcome submit to the workshop co-organized by
@barbara_plank
@ Benjamin Roth,
@hongming110
@hhsun1
@lifu_huang
@muhao_chen
0
4
13
0
0
10
ReasonBERT () got merged to the PrimeQA repo led by
@aviaviavi__
! Interestingly, it was verified as reproducible by Arvan,
@NatalieParde
@luisggpina
in "Reproducibility in Computational Linguistics: Is Source Code Enough?"(only 2 out of 8 rand. selected)
Let's congratulate and thank
@XiangDeng1
from
@hhsun1
's
@osunlp
@OhioState
for his contributions on few shot learning with "ReasonBERT" into PrimeQA. FSL with ReasonBERT gives significant boost over vanilla PLMs e.g. RoBERTa. Give it a shot today:
0
3
13
1
2
10
@vldb2021
this week, we are discussing our work on
"TURL: Table Understanding through Representation Learning"
Paper:
Recorded video:
Also feel free to email us if you have any questions/comments.
2
1
10
If you are at
#Tapia2021
@Tapia_con
, come meet Ohio State faculty
@OSUengineering
! Feel free to ping Prof. Atanas (Nasko) Rountev, Prof. Chris Stewart, me or others to chat during or after the conference.
@OSUengineering
is one of the academic supporters this year.
@cmdit
0
2
10
@addwater233333
@ysu_nlp
@_akhaliq
@BoshiWang2
@osunlp
Thanks for the question! A very good and important one! We have some discussions on Page 7 in the paper that you might find interesting. Attached a screenshot for your convenience.
My further thoughts are: 1. If one tends to believe in data scaling, our findings suggest that it
3
2
9
📢📢Check out the new work by
@ysu_nlp
and students
@osunlp
on grounding language models with strong results. Excited about future work it enables!
❓How to ground language models like ChatGPT to real-world environments
📢📢We present Pangu, a generic neuro-symbolic framework for grounded language understanding, where a symbolic agent and a neural LM work in a concerted way.
➡️ SOTA on KBQA & few-shot KBQA with Codex (1/n)
4
52
195
0
1
9
Wish everyone a happy and healthy new year! Here our OSU
#NLProc
group did a brief annual summary. We are so grateful to our students for their hard work, mentors and colleagues for their support. Fingers crossed that we get to see more friends and good luck to all of us in 2022!
OSU
#NLProc
group annual summary: 2021 was challenging for everyone, but we are grateful that many wonderful things still happened during this difficult time thanks to the diligent work of our students and collaborators. Share several achievements that we are most proud of: [1/9]
2
1
21
0
0
9
@_akhaliq
Thank you very much for sharing🙏. Feel free to check my summary here:
Thanks
@_akhaliq
for sharing our work. Very proud to introduce my star student
@BoshiWang2
's new work
@osunlp
: Grokked Transformers are Implicit Reasoners: A Mechanistic Journey to the Edge of Generalization
Can transformers reason? Are transformers fundamentally limited in
13
79
357
0
0
9
Since the beginning of pandemic, we have been developing a COVID FAQ dataset. It is now available! Frederick is going to present it at the poster session NOW.
#EMNLP2021
#NLProc
Please join the discussion. Underline link:
(1/6) Happy to share our new
#EMNLP2021
work “COUGH: A Challenge Dataset and Models for COVID-19 FAQ Retrieval”. Check out our work if interested in high-quality resources for information/FAQ retrieval or
#NLProc
research for combating
#COVID19
.
1
2
21
0
2
9
@CabannesVivien
@f_charton
@KempeLab
@_Vassim
@alicey_ang
@AIatMeta
Interesting work! Will read it in detail later. Quite related to it, especially "transformers are likely to develop *inner circuits* specially dedicated to multistep reasoning," our recent work (…) discovers generalizing circuits for implicit multistep
Thanks
@_akhaliq
for sharing our work. Very proud to introduce my star student
@BoshiWang2
's new work
@osunlp
: Grokked Transformers are Implicit Reasoners: A Mechanistic Journey to the Edge of Generalization
Can transformers reason? Are transformers fundamentally limited in
13
79
357
0
1
8
@bansalg_
@tongshuangwu
@besanushi
@ecekamar
@dsweld
Gagan, great work! We recently worked on explaining medical relation prediction (), and hope to further improve the human evaluation part (e.g., how to evaluate the effectiveness of explanations). Could you let us know if you have any comment on such tasks?
1
0
8
#siamsdm19
Congratulations to the Best Paper Award winners at SDM19! More photos are available here:
0
1
7
“eCeLLM: Generalizing Large Language Models for E-commerce from Large-scale, High-quality Instruction Data” () got accepted by
#ICML2024
, led by awesome students
@peng1230248
@xinyiling_
from
@ningx005
’s group
@TheNingLab
and
@RonZiruChen
from
@osunlp
!
Congratulations to Bo Peng (
@peng1230248
) and Xinyi Ling (
@xinyiling_
), and our collaborators Ron Chen (
@RonZiruChen
) and Dr. Huan Sun (
@hhsun1
) from
@osunlp
, for the eCeLLM paper () accepted by
#icml2024
!
This is a testimonial of productive collaboration
0
0
7
0
1
8
Curious about recent progress in semantic parsing over knowledge bases? 🤔
Check out our latest survey paper on KBQA: Knowledge Base Question Answering: A Semantic Parsing Perspective (), w/
@vardaanpahuja
,
@gong_cheng
,
@ysu_nlp
2
13
52
0
0
8
So excited to be part of the NSF AI institute ICICLE! Looking forward to building natural language and conversational interfaces to help democratize AI and collaborating with all team members. Check out our website and video: and
0
0
8
"When is Tree Search Useful for LLM Planning? It Depends on the Discriminator"
#ACL2024NLP
details in earlier tweet:
0
1
8
👍👍👍👏👏👏
Congrats to OSU NLP students for landing exciting research internships at Google Research (Xiang Deng), Google Brain (
@vardaanpahuja
), MSR (
@xiangyue96
), IBM-MIT AI Lab (
@zhenwang9102
), IBM Research (Tianshu Zhang), NIH (
@bernaaaljg
), and Amazon (
@LingboMo
). Have a fun summer!
2
1
30
0
0
8
Thanks for tweeting! Yep, please check out our SIGKDD'21 tutorial if you are interested in table understanding.
A must-read for any newcomer to the field of table understanding: "From Tables to Knowledge: Recent Advances in Table Understanding" by
@jay_mlr
@szeke
@hhsun1
@muhao_chen
1
0
14
0
2
7