At
#CHI2024
, we try to do justice to
#crowdsourcing
by answering this question:
How does GPT-4's labeling ability compare to that of a realistic, well-executed MTurk *pipeline*?
Led by
@StevenHe918
, we publish our love letter to crowdsourcing:
#LLM
1/🧵
I got tenure! Our dean,
@ISTtapia
, met me in person and handed me the magic paper. I'll be an Associate Professor with tenure, effective July 1.
Thanks to everyone in
@PSUCrowdAILab
, my collaborators, mentors, friends, family & my caring, loving partner.
This is truly for you.
This is probably timely for the ChatGPT era-- Our
#CHI2023
Late-Breaking-Work paper raises an interesting question:
Are conversations truly desired by users for every kind of question?
arXiv:
Doing a PhD is not easy. Sometimes it's uncertain, sometimes it's stressful. -- But today is not one of those days.
Today I successfully defended my PhD thesis.
Super happy that I decided to come to
#UIST2023
to witness VizWiz
@jeffbigham
won the Lasting Impact Award! VizWiz is really an amazing project that inspired me and many of my peers. Congratulations!!
People say, in academia, we should celebrate even small victories.
I've recently been acknowledged for my teaching efforts. I'm part of the 2023 Dean's Circle of Teaching Excellence at my college :)
We've been together for 12 years, among which 8 years were long-distance between the U.S. and Taiwan.
Some personal news: We're married. We live in State College now.
When I was a student, I never had anything accepted at ACL.
10 months ago I started my faculty job.
@tinyaohsu
is my first student. Today, a follow-up work of our
#CHI2019
LBW, entitled "Visual Story Post-Editing", is accepted by
@ACL2019_Italy
.
Thanks for making it possible.
It's hard to read all
#COVID19
papers, so we highlight them for you. We had 248 mturk workers label the *Background, Purpose, Method, Finding, and Other* for 10,966 English abstracts in CORD-19. The kappa(crowd, expert)=.74, when kappa(expert, expert)=.79.
The amazing
#In2Writing
workshop will be again at
#CHI2024
! This year we have an very intriguing theme:
🔥Dark Sides: Envisioning, Understanding, and Preventing Harmful Effects of Writing Assistants🔥
Take a look at our website!
An international student once asked me what "leverage" means in English. It was probably during a job hunting season so I explained that in American English, "leverage" often means to gain advantage in negotiation.
He then said: In papers?
Me: Oh. In papers. It just means "use."
"Ah. That was 2020, one of the top AI conferences decided to desk reject 40% of submissions without giving any reasons."
"40%?!"
"Out of deep despair, some frustrated authors decided to give CSCW a try."
"CSCW?! The HCI one?"
"And THAT, was how we survived the AI Winter."
🔥New Dataset Alert🔥Our
#EMNLP2022
Finding paper, led by
@tingyaohsu
, introduces SciCap, a dataset for *Scientific Figure Captioning* that contains 416k+ line charts with captions extracted from 290k+ arXiv papers.
paper:
dataset:
🚨New Preprint Alert🚨
In our new preprint, we argue that generating captions for figures (line chart, bar chart, etc.) in scholarly papers is more of a *text summarization task* than a vision-to-language task.
Paper:
🧵1/n
#NLProc
#SciDoc
I'm thrilled to finally answer *that* question: I will join the College of Information Sciences and Technology (IST) at Pennsylvania State University (PSU) as a tenure-track assistant professor, starting Fall 2018.
I will fly in Montreal for
#CHI2018
today. Come say hi : )
You are overly social.
You walk too much, talk too much, maybe drink too much.
You sleep too little.
You're tired. You're emotional.
You have a wonderful time.
Thank you,
#CHI2022
. I'm glad we decided to come to see you.
Let's do this again next year.
Hey our paper on story ideation with the crowd to support creative writing-- led by my amazing PhD student
@appleternity
-- got (conditionally) accepted by
#CHI2020
.
THE FIRST CHI PAPER FROM
@PSUCrowdAILab
!!!
A group of students from Taiwan attending
#AAAI2023
missed Taiwanese food, so we took a short visit to Rockville and had an amazing Taiwanese dinner. :)
#CHI2020
paper alert! We introduce Heteroglossia, a Google Doc add-on that allows creative writers to request story ideas from crowd workers. Each worker is assigned a fictional character and come up with plot ideas from the character's perspective. 1/
The 2nd
#In2Writing
workshop will be at
#CHI2023
!
This year we invite 🔥2-page position papers🔥 that portray thoughts on writing assistants (see CFP). Submit a paper to join the in-person event!
🗓️Submission Deadline: 2/23
🗓️Workshop Date: 4/23 (Sun)
🔗
One thing about being a faculty is that paper rejection hits differently. It's sad, but you need to take care of the team and make sure they're okay and move towards the next goal. Also, you just don't have much time/energy to be sad. CSCW+CHI LBW+ACL deadlines are within a week.
My students and I will be at
#EMNLP2023
in person. We have a main paper on location-based VQG () and a Findings paper that *evaluates* (instead of generates, as our INLG paper) figure captions as summaries ().
Come say hi!
#NLProc
A student asked me what's more important in academia, working hard or being intelligent?
I was totally not prepared for this. So, I opened a big, fat, delicious mooncake I bought in Pittsburgh and shared half with the student.
That's pretty much the only answer I know.
I'm not the best writer in the world. But everytime I submit a big grant proposal, I feel I have the best students, collaborators, and partner in the world.
Thank you for all the support.
Today is probably a good day. My partner got her 2nd dose, no symptoms so far except muscle pain. And we got 1 Long and 1 Findings at
#ACL2021NLP
.
Time to go to bed :)
Today a student asked me if we could change our meeting time because his roommate, whose laptop is broken, is using his laptop to take an exam.
I can't stop thinking about this the whole day.
Unpopular opinions:
1. We should cancel
#CHI2020
@sig_chi
b/c an outbreak in the US is likely coming:
2. Rolling deadline + quick turnaround time + unlimited pages = everyone gets tired. I reviewed for the past 2
@ACM_CSCW
deadlines and I'm concerned.
We're launching the 1st Scientific Figure Captioning (SciCap) Challenge!
We invite AI/NLP/CV researchers to build systems that caption all types of figures in arXiv papers. The challenge will be hosted at the CLVL workshop at
#ICCV2023
.
Join us here:
Talked to an NLP friend about their
#EMNLP2023
Findings paper
ME: so which workshop are you going to present this?
FRIEND: We... kinda give up
This is really sad.
#NLProc
Chorus Is Going To CHI!
Our paper, "Evorus: A Crowd-powered Conversational Assistant Built to Automate Itself Over Time", was accepted by
#CHI2018
. Thanks to cool coauthors
@jeffbigham
and
@josephcc
, and friends who gave feedback and proofread. I can't do this without your help.
One question I really don't like when I submit my letters is "Respect for authority". What does it mean by suggesting a prospective PhD student is "top 1% at having respect for authority"?
It's that time of the year I tell this story: In 2011, I applied to 18 U.S. CS PhD programs and got 16 rejections + 2 offers, both with no funding. CMU took me into its master program, not PhD.
I was frustrated and scared, too. But don't let rejections stop you.
Students, not getting accepted to PhD programs doesn’t say anything about you: I got rejected from the university where I’m now a professor and I have plenty of prof friends who all got many rejections before something worked out. Have faith in yourselves!
Yesterday was my birthday. I didn't go to school as I don't teach on Thu. To my surprise,
@appleternity
& Alan showed up at my door with a cake--and brought all the
@PSUCrowdAILab
and friends on a Zoom call!
This year wasn't easy. I'm really lucky to have your company & support.
A student is learning PyTorch and said she has been asking people so many stupid questions. I said, no worries, you will pass that on. Someone someday will ask you those stupid questions, and you will answer them.
After coming back from
#CHI2022
on Thu, I tested negative (in-home antigen rapid test) 3 days in a row and don't have any symptoms so far. Will keep monitoring my situation, as I was quite social at the conference.
#COVID
Visual storytelling models are often end2end (img->story). Our
#AAAI2020
paper () uses an old-school modular pipeline, allowing the use of external data and KG:
img2word->*KG comes to help*->word2story
Come to poster
#6878
on Sunday (Session3, 3:45-5:15)!
(This is my building's only elevator today. To all the
#CHI2024
rejections: You can always revise and resubmit, just not to this particular CHI and not via this particular elevator.
We committed to ACL before we saw the reviews because that's what a real commitment looks like.
(OR everything is confusing and we didn't understand the process correctly. Choose your narrative.
Our scores aren't great, but the comments by
#UIST2019
reviewers are helpful. I guess now we have more things to do for the summer.
AND I WILL TAKE MY LAB TO WATCH DETECTIVE PIKACHU.
Users might edit the stories that machines generated for them. If we have a set of pre-/post-edited stories, can a model be trained to post-edit machine-gen stories automatically? Our
#acl2019nlp
paper said yes!
Paper:
Data:
(In 2013 if an NLP person told me they're working on alignment I'd assume they're working on word alignment, probably for machine translation or paraphrasing.
Super exciting for this new collaboration! Thank
@penn_state
Center for Socially Responsible Artificial Intelligence (CSRAI) and
@ISTatPENNSTATE
for the support!
Happy to share the news that Kenneth Huang (
@windx0303
) and I got a grant! We will build AI systems to foster and evaluate creative thinking in science education, focusing on algorithmic fairness for underrepresented minority students in STEM.
Our
#EACL2023
paper is a simple yet interesting exploration of the *nationality bias* of GPT-2.
We prompted GPT-2 to talk about people from 193 different countries and measured the generated texts' sentiment scores (VADER). 1/3
arXiv:
Exactly 1 year ago--Mar 10, 2020--I met w/ Chieh-Yang
@appleternity
at 3pm in the lab to talk abt an idea of predicting story arcs for long novels.
That was the last in-person meeting I had.
Today, our work, "Semantic Frame Forecast", is accepted by
#NAACL2021
as a long paper.
Can we predict what will happen in the next 10/100/1000 sentences in a fiction book?
Our
#NAACL2021
paper by
@appleternity
treats a fiction as a sequence of text blocks & predicts the *TF-IDF vector of semantic frames* of block n+1, given prev blocks: 1/n
If you can't make it to
#AAAI2020
, and your paper is on image captioning, visual storytelling, or vision-to-language tasks, we can probably present your work for you.
Students from
@PSUCrowdAILab
kindly offer to do this. DM me! (We can only afford to help one extra paper.)
@windx0303
This is the most updated information we have for now. If we have more information to share, we'll update it on the homepage for the conference and post it to social media.