Alissa O.

MLT student @LTIatCMU @tsvetshop | B.S. Computer Science & Mathematics @WPI

[Jon Soucy @jrsoucy]

Honored to be a part of this year’s cohort Rising Stars! Thank you @BostonU_BME @ColumbiaBME @JHUBME @CornellBME for organizing such an amazing event!

[Jon Soucy @jrsoucy]

Very excited to share our recent work to improve cell replacement therapies for restoring vision lost to glaucoma and other optic neuropathies! @BaranovLab @Levi_J_Todd @HMSeye @MassEyeAndEar

https://t.co/LXu9PsZom6

[Vijay V. @vijaytarian]

LLM APIs let you build NLP systems in seconds; just write a prompt and use it as you wish. But APIs cost money and have privacy concerns.
Our new library Prompt2Model turns a prompt into a small expert model that can match LLM performance but runs locally!
https://t.co/bdohmjpZMV

[Alissa O. @alissa_OS]

@shocheen @OhioState @OhioStateCSE @ai2_allennlp @allen_ai Congratulations!!!! That is amazing!!

[Sachin Kumar @shocheen]

📢 I am excited to share that I will be starting as an Assistant Professor @OhioState @OhioStateCSE in Fall 2024 🥳. Before that, I will spend a year as a postdoc in the @ai2_allennlp team @allen_ai!

[Vijay V. @vijaytarian]

Text clustering is widely used but unpredictable, because there are many "right" ways to cluster data. Semi-supervised clustering tries to solve this, but too often it requires a lot of user feedback to be useful. We explore the problem of “few-shot clustering” using LLMs.

[Clara Na @claranahhh]

Excited to share our paper (w/ @sanketvmehta & @strubell), "Train Flat, Then Compress: Sharpness-Aware Minimization Learns More Compressible Models," to appear at #EMNLP2022 Findings (will be presented at SustaiNLP 2022)!
Paper here 👉 https://t.co/DnpT0l0xml
🧵👇 (1/n)

[thamar | @thamar_solorio]

It really takes a village of wonderful, responsible, and creative volunteers to pull together such an amazing 60th edition of the ACL. THANK YOU! i enjoyed meeting people and hearing about all the exciting research happening #acl2022 @AlineVillav @preslav_nakov @SmaraMuresanNLP

[Alissa O. @alissa_OS]

Why dabble in NLP tasks like “joke explanation” knowing that a model has “potential… to falsely affirm stereotypes”? (PaLM 👀)

[Alissa O. @alissa_OS]

@Vit_Bhandari Thank you! :) Yes, for sure, that is one limitation. Ideally you’d want a model to predict these preferences, so you won’t have to rely on labeled data every time

[Alissa O. @alissa_OS]

Do you speak multiple languages and alternate between them in conversations? Do you *code-switch* depending on who you talk to and how comfortable you are with each language? These are some questions we explore in our work, to appear in #ACL2022 🍀🇮🇪 (https://t.co/uzpAS25IN9)

[Alissa O. @alissa_OS]

We believe our approach can apply to other speaker-driven tasks, too! For details please see our preprint (https://t.co/uzpAS25IN9).
Hope to see you in Dublin!

[Alissa O. @alissa_OS]

Using the SelfExplain framework from @tsvetshop, we thoroughly analyze key features driving code-switch predictions. For example, we find that speakers tend to accommodate the preferences of their conversational partners, which is in line with prior sociolinguistic studies!

[Alissa O. @alissa_OS]

@jrsoucy @ARVOinfo @MassEyeAndEar @HMSeye @BaranovLab 🎉🎉 Well deserved!!!

[Alissa O. @alissa_OS]

@jrsoucy @NIH @NatEyeInstitute @BaranovLab @HMSeye @MassEyeAndEar Congratulations!!! An amazing achievement 😀

[Vijay V. @vijaytarian]

This project was joint work with @Sid_Arora_18, Alissa Ostapenko and @siddalmia05, and advised by Florian Metze, Shinji Watanabe, and Alan W Black, all from @LTIatCMU. (9/9)