🎉 Tomorrow is the day! 🎉
We are extremely excited to welcome everyone to the 2nd installment of the Deep Learning For Code Workshop!
Start: 8:30 am EST/ 2:30 PM CAT
ICLR: iclr.cc/virtual/2023/w
Webpage: dl4c.github.io/schedule/
#ICLR2023
Follow
Click to Follow zijianwang30
Zijian Wang
@zijianwang30
Zijian Wang’s Tweets
3
6
Most of us have been using coding assistants. Are they good or bad and what can we learn from user's interactions? Our first invited speaker Nadia Polikarpova from
will be talking about How Programmers Interact with AI Assistants! (1/N)
1
5
12
Show this thread
Thank you to all authors who submitted their works to the DL4C workshop @ ICLR 2023!
Instructions and Upload Form for Authors of Accepted Papers: forms.gle/LiTdK393NTz4VJ
Deadline: May 1st
4
5
1/ Multi-lingual HumanEval are now available on Huggingface!
+ Blog post on GPT-4 Code Generation Performance in 10+ languages
A 🧶
huggingface.co/datasets/mxeva
Collaborators including
1
6
12
Show this thread
📢We have extended the deadline to *February 10th 11:59 PM AoE*
Call for papers: dl4c.github.io/callforpapers/
Submission Link: openreview.net/group?id=ICLR.
Quote Tweet
The deadline to submit to the DL4C Workshop is less than a month away!
Link: openreview.net/group?id=ICLR.
Submit by February 3rdShow this thread
2
4
📢We are excited to announce our incredible lineup of speakers for the Second DL4C Workshop @ ICLR 2023!
Link: dl4c.github.io/speakers/
The speakers are:
1
19
76
Show this thread
📢 The deadline to submit to the DL4C Workshop is less than a month away!
Link: openreview.net/group?id=ICLR.
Submit by February 3rd
1
10
22
Show this thread
Code LMs need to know BOTH in- and cross-file context to write better code!
Check our new preprint arxiv.org/abs/2212.10007 to see how we improve Code LMs by jointly learning in- and cross-file context.
Joint work w/ and others at AWS AI Labs. (1/5)
3
8
15
Show this thread
Excited to share the 2nd workshop is back at ICLR'23. We look for contributions in all fields of DL4code, esp. HCI for Code, Evaluation for Code, Inference for Code, Responsible AI for Code, and related Open Sourcing efforts. Submit by Feb 3.
2
11
14
Show this thread
Excited to co-organize the 2nd workshop w/ . We are looking for contributions (due Feb. 3) and reviewers. See dl4c.github.io/callforpapers/ for details. Workshop sponsored by
Quote Tweet
Excited to share that the second installment of Deep Learning for Code workshop is back at ICLR 2023! Considering submitting your work by Feb 3rd, 2023. Details: dl4c.github.io/callforpapers/
@tscholak @GOrlanski @DL4Code
Show this thread
1
3
9
Thrilled to share MBXP, our new execution-based multilingual code generation benchmark ⬇️
Quote Tweet
We present multi-lingual code evaluation benchmarks MBXP and multi-lingual HumanEval, available in 10+ programming languages. With these datasets, we evaluate many facets of language model’s code generation abilities (arxiv.org/abs/2210.14868) @AmazonScience
Show this thread
read image description
ALT
1
10
Excited to announce a new product from AWS AI: Amazon CodeWhisperer
1
10
ACL would like to bring your attention to ACLRollingReview, a system that is intended to be used in the long term by all *ACL conferences for reviewing
as described in this post:
lnkd.in/gUDCMHU #NLProc
2
65
169
We're now accepting applications for the 2021 CSLI Undergraduate Summer Internship Program!
Interns will be paired with Stanford labs and participate in 8 weeks of mentored research. Stipend provided. Prior research experience not required: www-csli.stanford.edu/csli-summer-in
31
38
We hope that this work paves the way to large-scale studies of how readers form judgments of guilt in crime reporting, and encourages the development of systems that provide guidance on the presentation of these reports. 7/7
Show this thread
We showed that SuspectGuilt can form a basis for building tools that can predict guilt. We trained BERT models, and showed that these models were improved by genre-specific pretraining (w/ 470k+ unlabeled crime news) and token-level supervision from the highlighting. 6/7
1
Show this thread
However, for the number of *highlighted* hedges, it becomes a significant predictor for the author belief rating, suggesting that the less likely annotators thought the author believed the suspect was guilty, the more hedges they highlighted to justify the given rating. 5/7
1
Show this thread
SuspectGuilt provides a rich picture of how
linguistic choices affect subjective guilt judgments. For example, looking at the rating vs the number of hedges, we found that the number of hedges has no effect on the guilt assessments. 4/7
1
Show this thread
Annotators were asked to answer two questions & highlight in the story why they gave their response
1. Reader perception: How likely is it that
the main suspect(s) is(are) guilty?
2. Author belief: How much does the author
believe that the main suspect(s) is(are) are guilty? 3/7
1
Show this thread
Crime reporting has the power to shape public perceptions and social policies. How does the language of such reports act on readers? To address this question, we created SuspectGuilt, a corpus of 1800+ annotated crime stories from 1000+ local newspapers in the U.S. 2/7
1
Show this thread
Excited to share our work "Modeling Subjective Assessments of Guilt in Newspaper Crime Narratives" at w/ co-first-author and . #conll2020 #emnlp2020
Paper: arxiv.org/abs/2006.09589
Talk: slideslive.com/38939467
Thread👇 1/7
1
5
18
Show this thread
Just uploaded the GoldEn Retriever poster we presented at #emnlp2019 (exactly) two weeks ago: qipeng.me/research/slide
Lots of interesting discussions at EMNLP, looking forward to more advances in multi-hop QA in the wild!
📜: nlp.stanford.edu/pubs/qi2019ans
💾:
8
43
Come to see our posters at 10:30-12 (P20) and 15:30-16:18 (P40) #emnlp2019
Quote Tweet
At #emnlp2019 today:
Answering Complex Open-domain Questions Through Iterative Query Generation by @qi2peng2 et al. Poster 10:30–12:00
TalkDown: A Corpus for Condescension Detection in Context @zijianwang30 and @ChrisGPotts Poster 15:30–16:18
5
Fixing anti-social language use one problem at a time. TalkDown: A Corpus for Condescension Detection In Context by & will appear at #emnlp2019. Context helps; BERT doesn’t solve the problem. arxiv.org/abs/1909.11272 #NLProc
24
61
The NLP community has made great progress on open-domain question answering, but our systems still struggle to answer complex questions using lots of text. Read our latest blog post, by , about enabling multi-step reasoning in these systems!
20
49
Check out our #emnlp2019 paper “Answering Complex Open-domain Questions Through Iterative Query Generation” at arxiv.org/pdf/1910.07000
Quote Tweet
"What's the Aquaman actor's next movie?"
Complex questions are common in daily comms, but current open-domain QA systems struggle with finding all supporting facts needed. We present a system in #emnlp2019 paper that answers them efficiently & explainably: qipeng.me/blog/answering
Show this thread
4
2) a labeled Reddit dataset of condescending linguistic acts in context. We show the importance of context, motivate techniques to deal with low rates of condescension, and use our model to estimate condescension rates in various subreddits. Data & model:
Show this thread
1) Our new #emnlp19 paper (w/ ) on condescension modeling: arxiv.org/abs/1909.11272
Condescending language use is caustic. However, little work was done on modeling condescension due to its difficulty in detection without context. To address this, we present TalkDown,
1
7
Show this thread
Infer gender, age, and organization-status from image and text profile data. Identify over/underrepresented groups in your data. Great job presenting m3.euagendas.org #IC2S2
6
13
Quote Tweet
Check out our #TheWebConf poster on "Demographic Inference and Representative Population Estimates from Multilingual Social Media Data"! We're #1813. w/ @computermacgyve, @davlanade, @przemyslslaw, Timo Hartmann, @ffloeck, and @david__jurgens @TheWebConf
Show this thread
3
7
31
Show this thread
Check out our #TheWebConf poster on "Demographic Inference and Representative Population Estimates from Multilingual Social Media Data"! We're #1813. w/ , , , Timo Hartmann, , and
1
7
20
Show this thread
Check out our paper "Demographic Inference and Representative Population Estimates from Multilingual Social Media Data" at #TheWebConference. Online version coming soon!
2
3
Show this thread
Ever wondered how to get a representative sample of the general population from Twitter? We show how and provide a library to help in our newly accepted #www2019 paper . Work w/ , , , Timo Hartmann, , and
2
12
32
Show this thread