Suraj Nair

Computer Science PhD student at @umdcs. Interested in cross-language IR and NLP

[Sumit @_reachsumit]

A Representation Sharpening Framework for Zero Shot Dense Retrieval

Amazon proposes a training-free framework that augments document representations with contrastive queries to improve zero-shot dense retrieval without retraining the model.

📝 https://t.co/vZM9cCIK5N

[Sumit @_reachsumit]

Token Pruning Optimization for Efficient Multi-Vector Dense Retrieval

Amazon proposes a learnable token pruning approach that reduces memory usage and inference time by up to 3.3x while maintaining competitive retrieval quality in zero-shot settings.

📝https://t.co/OxBcUimfsz

[Kenton Murray @kentonmurray]

This past semester, I had fun creating a new course at Johns Hopkins @JHUCompSci on Multilingual NLP. Basically, people frequently talk about NLP and AI within the context of English. This class was about the cool problems that exist when we look beyond English. 1/7

[Eugene Yang @EYangTW]

On Tuesday 3:30-5:00pm, I will be presenting a poster of "C3: Continued Pretraining with Contrastive Weak Supervision for Cross Language Ad-Hoc Retrieval" that I collaborated with @srnair04 @cramraj8 Rebecca Iglesias-Flores and @ApolloProf
https://t.co/62ql5knzDs

[Suraj Nair @srnair04]

@justice1996V Thanks @justice1996V. Your talk was great too.

[Craig Macdonald @craig_macdonald]

#ecir2022 paper (Transfer Learning Approaches for Building Cross-Language Dense Retrieval Models) by @srnair04 showing that our ColBERT-PRF approach (ICTIR 2021) works on a variety of languages

[Eugene Yang @EYangTW]

Happy to share that our paper entitled "C3: Continued Pretraining with Contrastive Weak Supervision
for Cross-Language Ad-Hoc Retrieval" (w/ @srnair04 @cramraj8 @tensorglow and @ApolloProf) at @SIGIRConf! Preprint coming shortly :)

[Suraj Nair @srnair04]

@thibault_formal @cadurosar @sclincha @bpiwowar Thanks @thibault_formal and co. Just so you know, there is a weird issue with the training data. Some of the mined negatives include relevant passage ids. Here’s the associated Github issue: https://t.co/3Qv3flsLug

[Eugene Yang @EYangTW]

I am part of the organizing team of NeuCLIR track at TREC next year with Dawn Lawrie, James Mayfield, @macavaney, @soldni, @ApolloProf and Paul McNamee! Join us at one of the planning sessions on Dec 14 at 900 EST and 2100 EST to learn more about it and provide feedback.

[Suraj Nair @srnair04]

Glad to share that "Transfer Learning Approaches for Building Cross-Language Dense Retrieval Models" was accepted as a full paper at #ecir2022. Joint work w/ @EYangTW, @lawrie_dawn, @kevinduh, Paul Mcnamee, @kentonmurray, @jcmayfield and @ApolloProf. Preprint and code to follow.

[Suraj Nair @srnair04]

@jobergum @luizhbonifacio @rodrigfnogueira Indexing sentencepiece tokens only might not be effective as shown in this paper: https://t.co/9rIvvwfW9s
However, in the past, character n-gram tokenization has been shown to be effective for retrieval in European languages.
https://t.co/l1sbMhM5yG

[Yanda Chen @yanda_chen_]

Check out our #ACL2021NLP paper "Cross-language Sentence Selection via Data Augmentation and Rationale Training" in Session 11D at 12-1pm (EDT) Tuesday Aug 3, 2021!

We present an effective approach for cross-language sentence retrieval in low-resource settings.

[Dr. Pedro Rodriguez @[email protected] @EntilZhaPR]

🚨#NLPaperAlert #ACL2021NLP #NLProc "Evaluation Examples Are Not Equally Informative: How Should That Change NLP Leaderboards?" w/@barrowjoseph @miserlis_ John P. Lalor @robinomial @boydgraber

https://t.co/YkX3kcOh50

Chat at #ACL2021NLP https://t.co/94v5U33Wo2 Summary thread:

[Suraj Nair @srnair04]

@aviaviavi__ @AkariAsai @aviaviavi__ Is there a paper associated with the submission? I'm interested in knowing the details of the approach.

[Behrooz Mansouri @behrouzmansoury]

Working on narrative extraction, storyline and timeline generation, temporal reasoning, detection, and removal of bias in generated stories? Consider submitting to the 4th Text2Story Workshop [https://t.co/H6YnpbGjAz]
Submission deadline: February 8th, 2021
#ECIR2021 @2021Ecir

[Suraj Nair @srnair04]

@TomKenter Maybe this paper: https://t.co/0xxRUMLrOO
The idea is to
- split the document into overlapping segments
- apply BERT on each segment
- concat the segment representation which is then fed to LSTM/Transformer

[Suraj Nair @srnair04]

@ian_soboroff I have a CLIR joke but I can't translate it.

[Joe Barrow @barrowjoseph]

Updated "AllenNLP The Hard Way" to use the newly released @ai2_allennlp 1.0!

Link: https://t.co/YVlroopSPw
Github: https://t.co/ouKOrK4GT9

[Suraj Nair @srnair04]

@HamedZamani @umasscs @UMassAmherst @wbc11 @jallan_umass Congratulations Hamed!