Yarden Tal

What is the connection between model size and gender bias? 📢Excited to introduce my first paper “Fewer Errors, but More Stereotypes? The Effect of Model Size on Gender Bias” with @inbalmagar and @royschwartzNLP at the #GeBNLP workshop at #NAACL2022 arxiv.org/abs/2206.09860 1/n

Ever skimmed an article, pinpointing key info, and wished for a tailor-made summary without crafting it yourself?🤔 Introducing SummHelper: your go-to for personalized summarization. 📜✏️ w/ Niv Nachum @pyshmulik @obspp18 Ido Dagan 1/n

Is dataset debiasing the right path to robust models? In our work, “Fighting Bias with Bias”, we argue that in order to promote model robustness, we should in fact amplify biases in training sets. w/ @royschwartzNLP In #ACL2023NLP Findings Paper: arxiv.org/abs/2305.18917 🧵👇

Excited to present our new work on Adaptive Inference methods at #ACL2023NLP! Our paper uncovers fascinating insights about the Multi-Model and Early-Exit approaches. Work done with @MichaelHassid, @Jonatha25240734 and @royschwartzNLP. 🧵1/9 arxiv.org/abs/2306.02307

Looking for a new SOTA few-shot classifier? We present QAID! QAID uses batch contrastive learning with BERTScore & classifies by retrieving the intent name. SOTA on few-shot Intent Detection! 🎉 arxiv.org/abs/2303.01593 Accepted to @iclr_conf 2023 🥳 #NLProc #NLP #ML

Happy to share our #AAAI23 paper VASR: Visual Analogies of Situation Recognition. arxiv.org/abs/2212.04542 Man:king :: man:woman. Now, with images. Which candidate best solves the analogy? w. @ron_yosef @StrugoEli @HyadataLab @royschwartzNLP @GabiStanovsky 🧵1/8

We want to pretrain🤞 Instead we finetune🚮😔 Could we collaborate?🤗 ColD Fusion: 🔄Recycle finetuning to multitask ➡️evolve pretrained models forever On 35 datasets +2% improvement over RoBERTa +7% in few shot settings 🧵 #NLProc #MachinLearning #NLP #ML #modelRecyclying

Attending #NeurIPS2022 in New Orleans 😀 We will present our paper WinoGAViL (arxiv.org/abs/2207.12576) in Poster Session 1 (nips.cc/virtual/2022/p…). The paper was selected as a featured presentation, on Dec 7, 5-7 p.m (similar to our oral presentation in previous years).

How much does Attention actually attend? Apparently, not as much as you might have thought. New paper in Findings of EMNLP with Michael Hassid, @haopeng_nlp @wittgen_ball @ivanspmontero @nlpnoah @royschwartzNLP arxiv.org/abs/2211.03495 #EMNLP_2022 (1/n)

I don’t train from scratch, I use RoBERTa🧐 Wait… Why not cross-encoder/stsb-roberta?facebook/muppet-roberta? We automatically identify the best models on 🤗(periodically) Just pick the best one and finetune on your task ibm.github.io/model-recyclin… arxiv.org/abs/2211.00107

Can an AI joke & win Cards Against Humanity? #emnlp2022 findings short paper (#3479): "Cards Against AI: Predicting Humor in a Fill-in-the-blank Party Game" w/ @HyadataLab Paper: arxiv.org/abs/2210.13016 Code: github.com/ddofer/CAH Data: @CAH labs 1/ #CAH #NLP

Analogies are everywhere! We Gotta Catch 'Em All! ◓😎 📣 #emnlp2022 (main conference) long paper 🎉: 🎪 Life is a Circus and We are the Clowns: Automatically Finding Analogies between Situations and Processes 🤡 w/ @HyadataLab paper: arxiv.org/abs/2210.12197 🧵 1/

Machine Translation often fails on syntax We trained the model to extract syntax and reencode the predicted syntax arxiv.org/abs/2101.12640 me @oabend #conll2020 #NLProc

Transformers are not Seq2Seq Given a context They predict a single token We use this to update representations between predicted tokens & feed a changing graph relevant to the current token arxiv.org/abs/2101.12640 me @oabend #conll2020 #NLProc

Happy to share that our paper was accepted to #NeurIPS2022, Datasets and Benchmark :) Camera-ready version is available here: openreview.net/pdf?id=aJtVdI2… Also, check out the Huggingface integration + Colab: huggingface.co/datasets/nlphu… We'd be there in person and would love to talk!

Every machine learner knows that the dataset quality is crucial. Active learning detects the most valuable samples for annotation. Our #NeurIPS2022 paper proposes a theoretically grounded (and effective!) approach to low-budget active learning. arxiv.org/abs/2205.11320

For my first course at @jhuclsp, I am leading a class on recent developments in "self-supervised models." Here is the list of the papers and slides we cover: self-supervised.cs.jhu.edu Would love to hear Twitter's suggestions for additional exciting developments to discuss!🤗

Excited to share that next week I will be giving an oral presentation of our recently published paper @ Interspeech 22. 🌟Presentation video🌟: youtube.com/watch?v=3EbaWb… 📜🐍🔊full paper, code and audio: lnkd.in/dMYpMEMX @MosheMandel1 @FelixKreuk @adiyossLC

Efficiency is becoming a central theme in #NLProc. But there are so many different aspects of efficiency! Our recent paper surveys the wealth of approaches on the topic, providing a taxonomy of ideas, and outlining important open problems. arxiv.org/abs/2209.00099 1/3

Which images best fit the cue werewolf? Did you know V&L AI models only get ~50% on our challenging WinoGAViL association task but humans get 90%? w. Nitzan Ron Yuval @mohitban47 @GabiStanovsky @royschwartzNLP Paper: arxiv.org/abs/2207.12576 Website: winogavil.github.io 🧵

Using Phonetic Features is becoming a commonly used methodology for improving model performance in various speech-related applications. 🌟Check out our work🌟@ISCAInterspeech 💡Project Page: pages.cs.huji.ac.il/adiyoss-lab/SC… 📜Paper: arxiv.org/pdf/2206.11000… 🐍Code: github.com/slp-rl/SC-PhASE