Swabha Swayamdipta
@swabhz
Followers
7K
Following
3K
Media
42
Statuses
1K
Assistant Prof. @CSatUSC | Researcher in #NLProc | Previously @uwnlp @allenai
Los Angeles, CA
Joined February 2013
RT @murtazanazir: oh, and I am applying for PhD this time around. let me know if your lab is hiring! I like anything deep learning! recentl….
0
1
0
Last photo, not in the same cohort, but still important :)
0
0
1
RT @SimonsInstitute: On the horizon for March 31 – April 4, a workshop on The Future of Language Models and Transformers, organized by @sru….
0
6
0
RT @jaspreetranjit_: Feeling deeply grateful to have won an outstanding paper award at #EMNLP2024 for our work: OATH-Frames, congrats to my….
0
4
0
RT @roxanneelbaff: My favorite paper at #EMNLP2024:.„OATH-Frame: a novel framework to understand, synthesize and characterize large-scale p….
0
1
0
RT @robinson_n8: #NLProc can help analyze public attitudes towards homelessness on social media to inform public policy, @jaspreetranjit_ p….
0
2
0
RT @mdredze: Congrats @AlexanderSpangh and co-authors @sebgehr @VioletNPeng on our #EMNLP2024 Outstanding Paper Award @TechAtBloomberg. htt….
0
10
0
We won an outstanding paper award at #EMNLP2024! So proud of my PhD students @jaspreetranjit_ and @BrihiJ, the wonderful @rebeccadorn_ and our generous collaborators at USC's School of Social Work, especially @EricRicePhD. Thanks for making it happen, @CAIS_USC!.
#2 OATH-frames, a corpus of 2.4M X posts on homelessness, characterized as perceptions, critiques and responses and 9 finer-grained labels, in a collaboration between social workers and their LLM Assistants. Oral 10:30a, 11/14: .
6
8
99
#3 We present a fast out-of-distribution detection algorithm suite with soft clustering with Non-Negative Kernel Regression, in work led by my undergrad student, Aryan Gulati. Poster Session F, 11/14: Paper:
aclanthology.org
Aryan Gulati, Xingjian Dong, Carlos Hurtado, Sarath Shekkizhar, Swabha Swayamdipta, Antonio Ortega. Findings of the Association for Computational Linguistics: EMNLP 2024. 2024.
0
0
4
#2 OATH-frames, a corpus of 2.4M X posts on homelessness, characterized as perceptions, critiques and responses and 9 finer-grained labels, in a collaboration between social workers and their LLM Assistants. Oral 10:30a, 11/14: .
📢 Can LLMs 🤖 assist social workers 👩🏻💻 in characterizing discourse on social issues?. We introduce OATH-Frames: a resource for understanding Online Attitudes Towards Homelessness at scale, built with LLMs as assistants. 📄🔗 [1/n]
1
0
6
#1 We present SEPARABILITY, a meta-evaluation measure to improve the reliability of pairwise comparisons of generative outputs of language models. Compare without despair! .Poster Session A, 11/12: Paper:
aclanthology.org
Sayan Ghosh, Tejas Srinivasan, Swabha Swayamdipta. Findings of the Association for Computational Linguistics: EMNLP 2024. 2024.
Pairwise preference judgments are now standard for evaluating LLM generations and preference-tuning LLMs. In our new paper, we ask: are all test instances equally suitable for preference ratings?. We introduce a meta-evaluation measure called SEPARABILITY to estimate how likely a
1
0
5
Excited to be at #EMNLP2024 - come say hi! Also looking forward to meeting folks at these sessions, see 🧵below.
2
2
40