Lianhui Qin @ ICLR 2024
@Lianhuiq
Followers
4,225
Following
399
Media
16
Statuses
397
Incoming Assistant Professor at UCSD CSE. Currently postdoc at AI2 Mosaic. NLP, ML, AI. I’m recruiting PhD students.
Seattle
Joined October 2018
Don't wanna be here?
Send us removal request.
Explore trending content on Musk Viewer
#MetGala
• 2575608 Tweets
Rafah
• 2427023 Tweets
#스키즈_멧갈라_축하해
• 361313 Tweets
ariana
• 261454 Tweets
Zendaya
• 218711 Tweets
Lana
• 190153 Tweets
#الاهلي_الهلال
• 181682 Tweets
Manchester United
• 166535 Tweets
Superman
• 128149 Tweets
مالكوم
• 111700 Tweets
Crystal Palace
• 108479 Tweets
連休明け
• 101493 Tweets
Tyla
• 96517 Tweets
Rihanna
• 83026 Tweets
GW明け
• 80210 Tweets
メットガラ
• 73472 Tweets
Olise
• 73111 Tweets
Casemiro
• 71795 Tweets
#CRYMUN
• 64625 Tweets
Katy Perry
• 60208 Tweets
gigi
• 57996 Tweets
Garden of Time
• 52273 Tweets
Macklemore
• 51764 Tweets
emma chamberlain
• 49298 Tweets
bruna marquezine
• 47020 Tweets
Cardi
• 45194 Tweets
Kendall
• 44078 Tweets
Onana
• 35636 Tweets
Man U
• 35116 Tweets
Anna Wintour
• 33572 Tweets
Blake Lively
• 27478 Tweets
Kylie
• 27423 Tweets
taylor russell
• 27362 Tweets
休み明け
• 24721 Tweets
Hades 2
• 23207 Tweets
Law Roach
• 23066 Tweets
سمير عثمان
• 18676 Tweets
Mateta
• 17418 Tweets
ゴールデンウィーク明け
• 10540 Tweets
Pinned Tweet
🚀 PhD Opportunities in AI
@ucsd_cse
for Fall '24🌞
I'm recruiting PhD students with a passion for Large Language Models (LLMs), reasoning, generation, and AI for science.
I'll be attending
#NeurIPS2023
and
#AAAI2024
. Happy to catch up there. ☕️
#AIResearch
#PhD
#UCSD
#LLMs
[Personal news]
📢So excited to share that I’ve just graduated
@uwcse
and will be joining
@UCSanDiego
@ucsd_cse
🌊☀️as an Assistant Professor in Fall 2024. Meanwhile I’m doing a postdoc
@allen_ai
. Look forward to working with students and colleagues on NLP, ML, etc.
53
35
823
7
90
496
[Personal news]
📢So excited to share that I’ve just graduated
@uwcse
and will be joining
@UCSanDiego
@ucsd_cse
🌊☀️as an Assistant Professor in Fall 2024. Meanwhile I’m doing a postdoc
@allen_ai
. Look forward to working with students and colleagues on NLP, ML, etc.
53
35
823
We're thrilled to introduce 🧊COLD decoding, a general constrained text generation approach via energy based modeling🌞.
We can plug in any differentiable constraints into an energy function, and apply Langevin Dynamics for efficient sampling. 🥳
paper
6
94
510
📢Introducing ❄️COLD-Attack⚔️, a unified framework for controllable jailbreaking of LLMs.
Thanks to the controllability, COLD-Attack enables new jailbreak scenarios that are hard to detect🧐:
1⃣revising a user query adversarially with minimal paraphrasing
2⃣inserting stealthy…
4
48
236
🤔How can a pre-trained left-to-right LM do nonmonotonic reasoning, that requires conditioning on a future constraint⏲️?
Our
#emnlp2020
paper introduces DELOREAN🚘:
an unsupervised backpropagation-based decoding strategy that considers both past context and future constraints.
2
40
218
🧑🔬LLMs for complex Chemistry reasoning!🧪
Interestingly, we found LLMs (GPT-4) have already encoded lots of ⚗️Chemistry knowledge.
🤔What is really missing is a structured process to elicit the right knowledge, and use the knowledge to perform grounded reasoning.
A very…
🚀Announcing StructChem: A simple yet effective prompting strategy, unlocking the power of LLMs for complex chemistry reasoning. This task requires:
- Extensive domain knowledge
- Precise scientific computing
- Compositional step-by-step reasoning
Paper: …
2
23
109
1
23
149
COLD decoding accepted by
#NeurIPS2022
It enables generating arbitrarily constrained text with pretrained LMs, through continuous text approximation, energy-based modeling, and Langevin dynamics
Checkout the latest version
Code
We're thrilled to introduce 🧊COLD decoding, a general constrained text generation approach via energy based modeling🌞.
We can plug in any differentiable constraints into an energy function, and apply Langevin Dynamics for efficient sampling. 🥳
paper
6
94
510
1
18
127
What if Harry Potter had been a Vampire?
Our
@emnlp2019
paper, “Counterfactual Story Reasoning and Generation”, presents the TimeTravel dataset that tests causal reasoning capabilities over natural language narratives.1/2
Paper:
from
@uwcse
and
@allen_ai
2
36
117
I’ll be at
#ICML
🏝️between 26th and 30th and give an invited talk about differentiable and structured text reasoning
at workshop Sampling and Optimization in Discrete Space (SODS) on 29th
☕️🍻Excited to meet old and new friends! Ping me if you’re around
[Personal news]
📢So excited to share that I’ve just graduated
@uwcse
and will be joining
@UCSanDiego
@ucsd_cse
🌊☀️as an Assistant Professor in Fall 2024. Meanwhile I’m doing a postdoc
@allen_ai
. Look forward to working with students and colleagues on NLP, ML, etc.
53
35
823
1
2
80
🚀 Join us at
#AAAI2024
for an enlightening workshop on LLMs & Causality! 🏹
🔗 website:
🗣️ Don't miss our stellar lineup of speakers!
@yudapearl
@MihaelaVDS
@emrek
@AndrewLampinen
@osazuwa
@guyvdb
#LLMs
#Vancouver
LLMs and Causality?
We teamed up to give you the most recent perspectives.
🧵 (1/n)
#causaltwitter
#causality
#llm
#machinelearning
Your truly,
@matej_zecevic
@amt_shrma
@Lianhuiq
@devendratweetin
@AleksanderMolak
@kerstingAIML
2
16
44
0
9
46
#emnlp2020
Check out 🚗DELOREAN in our Back to the Future paper🎥
Come say hi at Zoom Q&A S4: Nov 16 17:00-18:00 PST / Nov 17, 1:00-2:00 UTC⏳
Paper:
Video:
Code:
w
@YejinChoinka
@VeredShwartz
@ABosselut
👇
🤔How can a pre-trained left-to-right LM do nonmonotonic reasoning, that requires conditioning on a future constraint⏲️?
Our
#emnlp2020
paper introduces DELOREAN🚘:
an unsupervised backpropagation-based decoding strategy that considers both past context and future constraints.
2
40
218
1
7
47
Will drive to UCLA from UCSD today! So excited to meet old and new friends there! 🌊🏝️
#SoCalNLP2023
is this Friday!!! 🏝
Check out our schedule of invited speakers and accepted posters! 👉🏽
0
5
34
1
0
39
Given any off-the-shelf left-to-right 🤖language models (eg. GPT2), 🧊COLD can enable it to do diverse constrained generation and non-monotonic reasoning tasks 👀, by plugging in, e.g.,
1⃣lexical constraint, 2⃣coherence constraints, 3⃣minimal-edit constraint, ...
1
5
33
Super exciting news!!
Honored to receive a Best Paper award at NAACL 2022 for NeuroLogic A*esque Decoding, with an awesome team
@GXiming
@PeterWestTM
@liweijianglw
@wittgen_ball
@DanielKhashabi
@Ronan_LeBras
@Lianhuiq
@YoungjaeYu3
@rown
@nlpnoah
@YejinChoinka
!
2
23
99
2
1
26
So proud of
@YejinChoinka
! Also the BEST phd advisor!!!
Someone once chased down
@UW
#UWAllen
@allen_ai
#MacFellow
@YejinChoinka
at a conference to tell her studying commonsense
#AI
was a “fool’s errand.” Years later, they sought her advice re: teaching a class on that very topic.¯\_(ツ)_/¯
#NLProc
#visionary
1
12
83
0
0
26
We introduce a new large conversation dataset grounded in external web pages (2.8M turns, 7.4M sentences of grounding). Joint work w/ my MSR mentors
@JianfengGao0217
, Michel Galley, collaborators
@chris_brockett
,
@AllenLao
, Xiang Gao, Bill Dolan, and my advisor
@Yejin
2/2
1
5
24
🌟Check out all the other faculty in the UCSD NLP Group at [].
We welcome co-advised research projects, fostering a collaborative approach in advancing the field of natural language processing.
1
2
21
Welcome to stop by, say hi and take a look at our demon poster(
#07
)! See you at 13:50 - 15：30. (Location: Basilica)
@ACL2019_Italy
@ZhitingHu
1
7
19
Though flexible, energy-based modeling for discrete text is notoriously difficult for sampling. Inspired by our DeLorean work(), we develop an efficient gradient-based sampling procedure based on Langevin Dynamics and differentiable representations of text.
2
2
18
(Update Github link)🧐
📢
#emnlp2020
"Back to the Future: Unsupervised Backprop-based Decoding for Counterfactual and Abductive Commonsense Reasoning"
🚘DELOREAN(DEcoding for nonmonotonic LOgical REAsoNing)
Paper:
Github:
0
3
14
Once a soft text sample is obtained, we discretize it to get the desired text. 🤔To ensure fluency during discretization, we propose a simple "top-k filtering" method: at each step the LM (GPT2) provides top-k candidate tokens and we pick the best using the soft sample (logits)😀
1
1
12
Huge thanks to my advisor
@YejinChoinka
, and
@LukeZettlemoyer
@etzioni
@JianfengGao0217
@ABosselut
, Fei Xia, colleagues
@uwnlp
, collaborators
@Microsoft
@GoogleAI
@Meta
@allen_ai
, and my family and friends!
1
0
12
We use COLD on various challenging text generation problems: 1) Lexically Constrained Decoding, 2) Abductive Reasoning, 3) Counterfactual Story Rewriting.
Comparing w/ previous discrete search & differentiable reasoning approaches show the flexibility of COLD with strong perf.
2
0
10
Really Cool Work!! 👍
🤯Machine Learning has many paradigms: (un/self)supervised, reinforcement, adversarial, knowledge-driven, active, online learning, etc.
Is there an underlying 'Standard Model' that unifies & generalizes this bewildering zoo?
Our
@TheHDSR
paper presents an attempt toward it
1/
1
45
236
0
0
8
0
0
7
Code+Data:
Joint work w/
@ABosselut
@universeinanegg
@_csBhagav
@eaclark07
@YejinChoinka
2/2
#emnlp2019
0
1
6
Can Twitter be a toolken?
⚒️ToolkenGPT
🔥Now any LMs can use massive tools, no finetuning, no prompt length limit
💡Calling a tool is as natural as generating a word token--treat tools as token (“toolken”) embeddings. Expand toolset by plug more toolkens
🤔Can embed millions of tools for LMs in future?
2
21
138
0
0
6
Congrats Yejin!!!
We're thrilled to learn that
@YejinChoinka
has been selected as an ACL Fellow for 2022, a highly prestigious recognition of her extraordinary contributions to the field of computational linguistics. Congratulations Yejin!
2
18
221
0
0
6
@lilianweng
Nice post! FYI, our recent work on
#EMNLP2020
proposes a gradient-based decoding that controls GPT2 to generate explanations and counterfactuals.
0
0
6
Interesting work about language model and world model
Embodied Agent Experiences Enhance LLMs
-Deploy LLM agent in simulator of physical world to acquire diverse experience via goal-oriented planning
-Finetune LLM on that exp to teach acting in world
-Improves over base on 18 downstream tasks by 64% on avg
6
77
334
0
3
4
COLD-Attack includes three main steps:
1⃣ Energy function formulation: Specify energy functions properly to capture the attack constraints such as fluency, stealthiness, sentiment, and left-right-coherence.
2⃣Langevin dynamics sampling: Run Langevin dynamics recursively for…
1
1
4
0
0
4
So cute!
0
0
4
LIMA, such a cute name. ☺️
How do you turn a language model into a chatbot without any user interactions?
We introduce LIMA: a LLaMa-based model fine-tuned on only 1,000 curated prompts and responses, which produces shockingly good responses.
* No user data
* No mode distillation
* No RLHF
27
233
1K
0
0
4
This is joint work w/
@YejinChoinka
@VeredShwartz
@PeterWestTM
@_csBhagav
@Ronan_LeBras
@jdch00
@ABosselut
🙌👏👍
from
@allen_ai
@uwcse
@ai2_mosaic
0
0
3
@followML_
Thanks for checking it out! Yes, this is our V1 on arxiv. You may want to check the latest version here: .
1
1
3
While maintaining a great success rate to diverse LLMs, COLD-Attack achieves strong stealthiness with lower perplexity (PPL) compared to previous methods.
0
0
3
@ABosselut
@uwcse
@UCSanDiego
@ucsd_cse
@allen_ai
Thank you Antoine!! Hope to see you in person in the future!
0
0
2
1
0
2
0
0
1
0
0
1
@VeredShwartz
@faeze_brh
@uwcse
@UCSanDiego
@ucsd_cse
@allen_ai
Thank you Vered! Your materials were very helpful!
0
0
1
0
0
1
0
0
1
0
0
1
Happening now
@West
208+209
Come join the
#NeurIPS2019
workshop on Learning with Rich Experience. Note the location: West 208+209.
Look fwd to the super exciting talks by
@RaiaHadsell
@tommmitchell
JeffBilmes
@pabbeel
@YejinChoinka
& TomGriffiths, and the contributed presentations:
0
6
20
0
0
1
0
0
1
0
0
1
0
0
1
@universeinanegg
@uwcse
@UCSanDiego
@ucsd_cse
@allen_ai
Thank you Ari!!! We should catch up in person sometime at conferences!!
0
0
1
0
0
1
0
0
1
0
0
1
0
0
1
0
0
1
@_kumarde
@rajammanabrolu
@uwcse
@UCSanDiego
@ucsd_cse
@allen_ai
Thank you Deepak! Looking forward to working with you!!
0
0
1
0
0
1
0
0
1
0
0
1