Barid Xi Ai

[Barid Xi Ai @AiBarid]

@cryptexcode Hi, I want to help. But I’m not at NAACL…Can we discuss it online?

[Barid Xi Ai @AiBarid]

@kearneymw Agree, how can I trust a company/university that can’t parse my cv with simple NLP :D

[Barid Xi Ai @AiBarid]

@alan_ritter @mlatgt Dear Prof. Alan Ritter, is this position still available?

[Barid Xi Ai @AiBarid]

@zouharvi How about Reproducibility? Every conference requires it, but it is not a prerequisite in the decision phase. 🤣

[Barid Xi Ai @AiBarid]

@MarroSamuele @ThomasMiconi Same

[Barid Xi Ai @AiBarid]

@chriswolfvision Our paper did not get any feedback. Sad like pic 1.

[Barid Xi Ai @AiBarid]

@pcastr I spent one week on response and revision. But no feedback. Sad.

[Barid Xi Ai @AiBarid]

@jkkummerfeld How about post-doc?

[Barid Xi Ai @AiBarid]

@BramVanroy Correction. 1M is enough in many cases. Wiki dumps in sw has 400k sentences.

[Barid Xi Ai @AiBarid]

@BramVanroy For corpus size, I think 100K sentences is enough in many cases. However, looking at the wiki dumps in sw, it only has around 40k sentences. It is also can form a good vocabulary for XNLI task.

[Barid Xi Ai @AiBarid]

@BramVanroy Another route is also practical. You can first get BPE codes and then get the vocabulary. Now you can adjust the BPE size according the minimum counts of tokens. Look at the original BPE GitHub. Rico may suggest this way.

[Barid Xi Ai @AiBarid]

@BramVanroy I swear I’ve read a title about BPE size, but I just cannot recall the name. However, instead of BPE size, I think considering the minimum counts of BPE codes is a general solution. (1/2)

[Barid Xi Ai @AiBarid]

@deliprao Awesome!

[Barid Xi Ai @AiBarid]

@delliott Neurips and ICLR allow rebuttal revisions. However, not every reviewer and AC buy it and may comment like “should be included in the initial revision”. So I think it does not change the whole thing. Actually, I’m interested in the email saying “don’t explicitly disallow”.

[Barid Xi Ai @AiBarid]

@zhuoran_yang @NeurIPSConf Really? I think authors are excluded in this phase unless a reviewer manually set the reader of the comment. But I’m not sure it is possible.

[Barid Xi Ai @AiBarid]

@AlexFengLiu1 @Ramy_E_Ali @NeurIPSConf Sad. A reviewer posted a very short comment titled with "Most issues clear" and promised to raise the score before ddl once back home from a trip. But I can't see it now. I don't know if the reviewer keeps the words.

[Barid Xi Ai @AiBarid]

@celticsforum Same here. I can’t help to check tweeter

[Barid Xi Ai @AiBarid]

@KingJames Are you in today?

[Barid Xi Ai @AiBarid]

@svpino 4