Sabitlenmiş Tweet
Who knew that writing a paper saying that the field of sentiment analysis in #NLProc is broken, can land you an OUTSTANDING PAPER AWARD!! 🎉🎉🎉 in the the sentiment analysis track at #EMNLP2023!!
English
Mukund Srinath @ EMNLP
93 posts
Mukund Srinath @ EMNLP
@MukundSrinath3
Machine Learning Scientist @ Expedia Group | NLP, IR and Trustworthy AI
Katılım Eylül 2020
351 Takip Edilen184 Takipçiler
Mukund Srinath @ EMNLP retweetledi
Wohhoo, our work got accepted at NAACL 2025! 🥳
TLDR: we built a methodology to improve reliability of any dataset by selecting only high-quality subsets!
Vipul Gupta@vipul_1011
🚨 New paper alert 🚨 Ever struggled with quick saturation or unreliability in benchmark datasets? Introducing SMART Filtering to select high-quality, reducing dataset size by 48% on avg (up to 68% for ARC!) and improving correlation with scores from ChatBot Arena! 📈✨ (1/N)
English
Mukund Srinath @ EMNLP retweetledi
For all Future PhDs @ #CSCW2024! 🎉 For anyone considering a PhD, check out my latest blog with tips on starting your application journey—from finding programs to crafting a compelling story. Hope it helps! 📚🚀 #AcademicTwitter #PhDLife
👉 @nlphci.phd/how-to-begin-with-phd-application-b13936d6b8ce" target="_blank" rel="nofollow noopener">medium.com/@nlphci.phd/ho…
English
Mukund Srinath @ EMNLP retweetledi
👋🏻 Excited for Day 1 at #CSCW2024! Can’t wait to connect with friends old and new. If you’re interested in #HumanCenteredAI or #AIforGood, let’s chat! I’ll be presenting my dissertation work on “Towards Dynamic Learning: A Framework for Simulating Adaptive Learning Systems.” #hci
English
Mukund Srinath @ EMNLP retweetledi
While at #CSCW2024 , I'm also happy to chat about fun things to do around! Costa Rica has been beautiful so far!
English
Mukund Srinath @ EMNLP retweetledi
Excited to share that my first first-author full paper with @guo_freeman has been accepted to #chiplay2024! In this paper, we examine the impacts of generative AI on indie game developers and discuss how it can be used to uplift developers while mitigating risks.
English
I had such a great time presenting at the #HCIplusNLPWorkshop at #NAACL2024! Huge thanks to @sulin_blodgett and @CurriedAmanda for putting together such a seamless and insightful workshop experience!
@Sanjana08395511 #NLProc #HCI
English
Mukund Srinath @ EMNLP retweetledi
Finally in 5th (maybe 6th?) attempt over past 16 months, my paper has been accepted!
It's been a marathon, but now I can peacefully focus on other projects. *sigh of relief*
Vipul Gupta@vipul_1011
Me: My paper has been rejected so many times, it has stopped bothering me anymore A friend (doing PostDoc): Congrats, you've mastered the ultimate PhD skill and are ready to graduate! Definitely need more friends like him lol 😂
English
Mukund Srinath @ EMNLP retweetledi
It is really motivating to see that the work O am doing is relevant and useful to the community. Although citations don't always capture the comprehensive significance of the work, it is certainly encouraging. #PhD #research #HumanCenteredAI #paper
Sanjana Gautam@Sanjana08395511
Thrilled to share that my research on human-AI interaction has reached a milestone of 150+ citations! 🎉 Grateful for the meaningful connections forged through research and excited to continue advancing the field. #humanAIinteraction #research #milestone #citations
English
TableLlama achieves comparable/better performance on in-domain tasks compared to other SOTA models.
TableLlama archives a significant gain in performance on six out-of-domain datasets.
English
Evaluation tasks
- Table Interpretation
- Table Augmentation
- Question Answering
- Fact Verification
- Dialogue Generation
- Data-to-Text
English
#NAACL2024
@naaclmeeting
TableLlama: Towards Open Large Generalist Models for Tables
Tianshu Zhang (@Tianshu_OSU)
arxiv.org/pdf/2311.09206
Català
Takeaways:
- LLMs possess some degree of reasoning, but a lot of it comes from instance-level and task-level overfitting
- In-context learning doesn’t fully bridge the gap either (Example on the addition task in a few-shot setting in different bases shown in the image)
English
However, it is first necessary to know if the LLM is able to understand the task.
The authors therefore additionally introduce a counterfactual comprehension task.
Example: Prompting the model to output the next number in a sequence in base-9
English
The approach is best explained using an example:
- GPT-3 achieves perfect performance on two-digit addition
- However, if the performance is not comparable in a less popular counterfactual setting (base-9 vs base-10), it might indicate memorization of the popular setting.
English
#NAACL2024
@naaclmeeting
Reasoning or Reciting? Exploring the Capabilities and Limitations of Language Models Through Counterfactual Tasks
Zhaofeng Wu (@zhaofeng_wu)
arxiv.org/pdf/2307.02477
English
LLMs seem to achieve super-human performance on several tasks but also fail on a number of basic problems. This paper probes whether LLM performance is simply a function of instance/task overfitting or if LLMs are reasoning over data.
English