Gopal Sharma

Would you quit your job right now if someone walked up and offered you $100,000 cash to do so right now?

@CVPR When in doubt about what session to attend with a conflicting schedule, select the session with a great speaker even if the topic is not perfectly aligned with your interest. You will end up learning more from a good representation

Any advice for a CVPR first-time attendee?

@CharaspowerAI Why is the girl in the background annoyed?

I've been wanting to get to the bottom of this story for so long. #DreamMachine #LumaAI 😂

Lab members when you ask them to volunteer

Professors when you need their signature

@CVPR a lot of students won't be able to make it due to not being able to get a US visa. Wouldn't it be rather fair to allow them to reduce their registration from in-person to online?

@rrhoover 1-click job applications/form-filling

What's a problem you would pay to have solved?

@CSProfKGD @dimadamen @jon_barron @eric_brachmann @david_picard @taiyasaki Onboarding doc should have example scenarios

@dimadamen @jon_barron @eric_brachmann @david_picard @taiyasaki Given the number of new people coming into the area, I think we shouldn’t take the obvious to us senior folk for granted. Documenting expectations this is not a bad thing.

#CVPR2024 motion #1

@jon_barron @david_picard @CSProfKGD @dimadamen @eric_brachmann @taiyasaki How about making an instructional video with examples for reviewers and ACs and putting the link on the review portal? People will likely go through a short video than a dense text.

@david_picard @CSProfKGD @dimadamen @eric_brachmann @taiyasaki yeah but you gotta figure that future CVPR 2025 new reviewers are more likely to read the CVPR 2025 onboarding docs than they are to read the CVPR 2024 TPAMI propositions.

@CSProfKGD Not sure how will it prevent reviewers from asking and ACs from accepting. I can easily report quantitative numbers in the text. Plus, adding qualitative results sometimes helps a lot to convince reviewers

#CVPR2024 motion #2

@jon_barron Agreed. A possible name for this AI-> "Border Collie". It should be possible to fine-tune existing LLMs with reviews written by outstanding reviewers of past conferences.

@beakywings In the limit I feel like this idea is basically a high-involvement AI AC/paper shepherd, which is also a good idea.

@tkasarla_ @ricucch @UNIMORE_univ @ELLISforEurope @Ellis_Amsterdam Congratulations!!

I will be spending this summer doing my #ELLISPhD research with @ricucch @UNIMORE_univ until September! Great time to be in Italy! 🎉 @ELLISforEurope @Ellis_Amsterdam

@frncswllms @gadelha_m @ylecun Great idea! I think openreview can be converted into this as researchers are already associated with this. Though, there might be a constant need of moderators

I've given this rant in person but never on twitter: - Instead of peer review, I'd love a system that is like Reddit. - Different subtopics are subreddits (e.g. r/cs.cv for vision). - Users post papers they want to publish to different subreddits. - Your account is associated with your real name so you're accountable for what you write (verified with an institutional email address or ID). - When you post a new paper, authors are anonymized for a period of time. - There are no explicit reviews, but there is a comments section where people can ask questions. - Replies are public, you can link experiments, etc. The thread becomes a living document of your work and questions and answers. You can also post revisions. Every year, the top X papers in terms of upvotes are presented at whatever conference. Interested readers will find the papers they want, we won't burden researchers with piles of papers to review.

This is a weird take (esp. about peer review). The tone of the message feels like @ylecun is gatekeeping science to lie wihtin the modern academic framework for paper submissions. Modern peer review as we know it didn't start until the 1950s [1] and is (IMO) a deeply flawed system (better than random [2] in some cases but not massively so). Any nontrivial piece of research can't realistically be fully grokked in the short amount of time reviewers have to submit a review. Computer Science is particularly bad where there are literally thousands of papers per year and not enough reviewers to handle the volume. Many authors treat paper submission as a game to squeeze their work into a conference for prestige/citations -- results are cherry picked, methods are optimized for particular benchmarks not represenative of the real world, etc.. This is NOT science IMO. Science is the process of applying the scientific method to answer questions. How these results are documented (code, a paper, a blog, a physical object that proves something is possible, a successful business, etc.) Reproducibility helps everyone else confirm the answer to a question is plausible. Let's not gatekeep science to a particular framework (especially one that has so many fundamental issues)! Anyone can do science with the right attitude! [1] blog.f1000.com/2020/01/31/a-b… [2] blog.neurips.cc/2021/12/08/the…

Shout out to IITR alum Bandhav for this amazing work!

@CSProfKGD We should also have a list of outstanding ACs. The quality of reviews can be improved by active involvement of ACs

As an area chair, over time you develop positive attitudes towards certain people because of the quality of their reviews. Glad to see so many familiar names 💪 YES, ACs know the names of their reviewers. There is a flip side too 😡

This "withdraw submission" button on CMT is so scary. It should not be on the front panel.

@IAmEricHedlin This will violate doubly blinded system. But we can have a universal reviewer rating (score assigned to a reviewer by ACs) that is publicly available

@beakywings What about making the names of reviewers public after the rebuttal? So peoples reputation is affected by the quality of their reviews

Maybe we should have an LLM for reviewers that guides them to write a better review. Perhaps a condescending LLM that berates reviewers for writing crappy reviews e.g. "Is this the best you can write as weakness?" or "Do you even research bro?"

Something that came up in today's discussion of a rebuttal