Shreya Havaldar
@shreyahavaldar
PhD student @cis_penn | student researcher @googledeepmind | multilingual NLP + cultural psychology | she/her 🌸
The struggle of making a personal website that’s pretty enough (so people know you can code) but also ugly enough (so people know you’re an academic)
🧠 Foundation models are reshaping reasoning. Do we still need specialized neuro-symbolic (NeSy) training, or can clever prompting now suffice? Our new position paper argues the road to generalizable NeSy should be paved with foundation models. 🔗 arxiv.org/abs/2505.24874 (🧵1/9)
Excited to share our new paper: "Instruction Following by Boosting Attention of Large Language Models"! We introduce Instruction Attention Boosting (InstABoost), a simple yet powerful method to steer LLM behavior by making them pay more attention to instructions. (🧵1/7)
Does #shame manifest differently across #cultures? Yes. Can LLMs identify #norms behind shame? Yes. Are women shamed more than men? Yes!!! Can #LLMs identify when someone is shamed? arxiv.org/abs/2402.11333 #NAACL2025
🚨 LLMs must grasp implied language to reason about emotions, social cues, etc. Our @GoogleDeepMind paper presents the Implied NLI dataset. Targeting social norms 🌎 and conversational dynamics 💬, we enhance LLM understanding of real-world implication! arxiv.org/abs/2501.07719
What makes features interpretable to create better explanations? We work with real-world experts to develop "The FIX Benchmark: Extracting Features Interpretable to eXperts"! 📝Blog: debugml.github.io/fix/ 📄Paper: arxiv.org/abs/2409.13684 💻Website: brachiolab.github.io/fix/
Excited to present our ICML 2024 paper: "Towards Compositionality in Concept Learning"! 🔗 Blog: debugml.github.io/compositional-… 📄 Paper: arxiv.org/abs/2406.18534 💻 Code: github.com/adaminsky/comp…
🚨Exciting news! #NLProc 📢The call for papers is now open for the Workshop on Human-Centered Large Language Modeling (#HuCLLM2024) at #ACL2024NLP. See CFP: hucllm-workshop.github.io/#Call%20for%20… Direct Submission Deadline: May 10, 2024 ARR-reviewed papers Commitment Deadline: May 17, 2024
We are pleased to announce that the first Conference on Language Modeling will be held at the University of Pennsylvania in Philadelphia at the Zellerbach Theatre. Thanks so much to UPenn CS as well as Mark Yatskar and Zachary Ives for facilitating the amazing venue.
‼️🚨 Postdoc Opportunity at @CIS_Penn 🚨‼️ Please RT 🙏 🎓 Are you interested in: -- NLP & ML x health behaviors -- studying heterogeneity -- translational research Apply here: docs.google.com/document/d/e/2… #DataScience #PostdocJobs #AcademicTwitter @WWBProject @PennMedCDH
I will be presenting our work on creating faithful groups of features for attribution at @XAI_in_Action workshop @NeurIPSConf today (Dec 16) at 4:30pm-5:30pm arxiv: arxiv.org/abs/2310.16316 blog: debugml.github.io/sum-of-parts Come to our poster for a chat!
Very cool work from a very cool person!!
How do CS and Medicine deal with Mental Health Conversational Agents (chatbots) differently? Our work in #EMNLP2023 systematically reviews related articles to gap the bridge between two fields. Please join us at West 3 on December 8th at 16:15 for my oral session!
The first @WWBProject reunion picture and @Diyi_Yang (sorry to miss you Jeffrey!) @JoaoSedoc @shreyahavaldar @danie
Anticipating how audiences may interpret the visual medium is critical for multimodal communication. However, can multimodal language models faithfully simulate human impressions of images? In our #EMNLP2023 paper, we present Impressions, a dataset of human perceptions of images
Super cool work from my lab 🥰
We develop Sum-of-Parts: faithful grouped attributions that overcome fundamental barriers of typical attributions. Paper: arxiv.org/abs/2310.16316 Blog: debugml.github.io/sum-of-parts/ Code: github.com/DebugML/sop Joint work with @_helenqu , Marco Gatti, Bhuvnesh Jain, @RICEric22.
If not for the caption, I would have thought this was GHC.
Selfied after talking with students in front of the Meta booth at ICCV - Paris
Is GPT psychologically WEIRD? Using the World Values Survey and other psych measures, we seat GPT within a global perspective. The culturally more distant a place is from the US, the lower the correlation with GPT @MohammadAtari90 @blasi_lang @DorsaAmir