397 points by EvgeniyZh 5 days ago
Really awesome, definitely bookmarking. Couple points:
- Shouldn't you link to the arxiv page instead of the PDFs in case there's a revision?
- What about a column for "area" as well? Would be nice to scroll through and see what areas have made progress relative to others lately.
- Some sort of very brief mini-about/summary at the top of the main page wouldn't hurt. A tagline of some sort.
- Now I notice there's an About under the hamburger menu -- why not just display HOME and ABOUT instead of hiding them in a menu? You don't have enough menu options to need to hide them in a hamburger menu.
Yeah, hamburger menu is weird. Also, it'd be nice to have a column for code (just links to github really).
We are an enthusiastic group of students trying to put in our two cents into the fast developing fields of Artificial Intelligence (AI) and Machine Learning (ML). ... One space we believe is currently not covered is that of an easily consultable repository for state of the art, quantifiable results across tasks. ... The main source of data is the community that can upload the relevant results of their field! We have further aggregated data from reading (a lot) of articles, from scraping the web, and from several existing datasets and websites. A portion of our dataset came from the Electronic Frontier Foundation’s "Measuring the Progress of AI Research."
Kudos for the effort. I wish more areas had something like this.
If they would put that at the top of the website page, it would help 1000%.
thanks, added a comment
Great to see another one of these resources. I'd highly recommend checking out NLP-progress  and Papers with Code  which I've found to be excellent resources.
Hey. One of the creators here. Yes we are still missing several tasks and results, the idea is that this could be maintained by the community by making it easy to upload results.
Definitely help is appreciated
We are collecting a lot of feedback
Hey thanks for this. Please add a finance category. Timeseries has some overlap with finance but there is plenty of finance AI research that is not restricted to time series. Financial data has many distinct characteristics and interesting challenges.
Commendable effort, but badly outdated already.
For example, the best result on CIFAR-10 for density models that I know of is 2.85 bits per dim (PixelSNAIL) rather than 3.35 (Glow), and this is a year old. PixelCNN++ is well known, and got 2.92.
For classification on CIFAR-10, current SOTA is much better than the 2015 Fractional-MP result (which also used non-standard augmentation, so perhaps not directly comparable to the rest)
For some reason they distinguish between accuracy and error rate for classification, and there is missing data for both. If you sort by error rate there are newer results for classification.
Yes. the idea is the community can help maintain it by adding results (and tasks or datasets)
I think this kind of resource is in a chicken-and-egg situation: for people to bother to contribute to it, it needs to be seen by others as an authoritative reference, and vice versa.
I agree that it's badly needed. There were other projects like this, that seem abandoned: http://rodrigob.github.io/are_we_there_yet/build/classificat...
This probably belongs on Wikipedia.
From the "I'm purposely playing dumb" perspective, I totally don't get that website on first glance. Even a 6 word sentence summary could make it a lot more interesting.
When I think of the things I would add to this, I realize how amazingly useful it would be to have more data on the tasks themselves. For starters, whether or not there is code and what license under which it is released, and for NLP what languages were used/are available for use. Also, sign language recognition fits somewhere between computer vision and NLP. If picked up by the community, this could become a huge knowledge base.
Looks super cool! There is a bit of data clean up to do. Just looking at speech recognition: "WER" and "Word Error Rate" should be the same thing, and sometimes it seems to be on a scale of 0 to 1 and other times a percentage. Also the Switchboard test set is duplicated. Finally, it really should be marked when data is augmented; many of these numbers are trained on outside data, which says more about how much data the researchers have access to as opposed to the ML system design.
I'm seeing duplicates on there. Maybe find a way to group them, then when you click the link you can then choose from a particular source?
I'm toying with the idea of making an Open Source repository of reproducible papers.
How did you get this started? How did you find people to collaborate?
I already bought reproduciblepapers.com
For the state of the art for segmentation on the Cityscapes dataset, PSPNet isnt there and to my knowledge at 2017, it was state of the art (SOTA). I dont see it on this site.
Yes. There are some missing results and tasks. The idea is the community will help maintain it by adding results, tasks and datasets
I can’t scroll the list on safari (iPhone) and instead have to use the small are left to the list scroll. Otherwise really cool site!
Very cool, will definitely keep this in my favorites.
However there is an issue with the pagination/navigation. I can't see the next items if there is more than 20.
I was just about to build this! Arghh you beat me to it..
I'm sure they'd be happy to accept help!
Ok! Was originally going to reach out to Sebastian Ruder and see if nlpprogress.com could be automated
what's your email??
Me too! Although it has been sitting in my list of potential projects for months so kudos to them for actually doing it...
Hey OP! This is really cool. Would there be a way to add a column that explains the uninitiated what each column means?
Very cool. I'd have liked a section concerning 3D Medical Image Segmentation/Classification.
Its easy to add datasets, results and tasks. the idea is the community can help maintain it :)
I really like that html table you are using! It's featureful and looks great. :)
can't connect, doesn't use TLSv1.2
For Atari they forgot Rainbow I think