In the "symbolic AI" area it is so common for results to just be forgotten for 20-30 years and then come back.
For instance Datalog was developed in 1986 but when I got interested it in 2006 I found it was really obscure, by 2016 there was some literature and a few good implementations and it is rolling on.
There were those old "rules engines": many of the classic expert systems were developed before the famous RETE algorithm and even then there were RETE implementations that were terribly slow because they didn't use hashtables! When that stuff was fashionable people thought 10,000 rules were a lot, today it is more like 10,000,000 rules are a lot and I'd credit that more to algorithms being better than the computers being more powerful.
Similarly since the mid 2000's there has been a revolution in SAT and SMT solving which goes back to the very early (1960s) work on trying to solve problems w/ logic it is just now we succeed instead of fail. (It's a lot like the neural network revolution and used the same method of competition as seen in TREC and visual recognition.)
there is probably difference if the paper claims that it resurrected some old concept and implemented it with better results compared to "this paper introduces X" as in case of this post.
Pardon me for being a currently pissed off PhD student dealing with a rejected paper but: who gives a fuck? Like honestly what is the harm of someone reinventing something, especially when the previous instance is long dead and buried/forgotten (like in this case). Oh it's a waste of resources you say? As if the constant demand for novelty isn't also.
My paper got rejected because a reviewer dug up some vaguely similar thing from 3 years ago and claimed it is "well used and well supported" even though the repo has had 3 commits in 3 years. Oh that's an extreme case you say? The presumptive necessity of "novelty" is what enables such extremes.
Like the schizophrenia of modern society is hilarious: 15 brands of potato chips is good (competition in the market is healthy) but one molecule in common (in science) and you're a failure.
nope - welcome to peer review where everything's made up and the points don't matter.
note: some venues have a rebuttal phase which 99% doesn't make a difference anyway but this one didn't.
note2: if you're thinking i got rejected for some other reason: i got (weak accept, accept, reject) and the lone reject came from the wackadoodle. btw how does (weak accept+accept+reject) == reject? see my earlier comment about how the scores don't matter - i emailed the PC and they said they use the scores to resolve a failure to come to a consensus by the reviewers (which is 180deg the opposite of how a scoring system is used - scoring systems "objectively" measure and then humans break ties).
note3: this isn't a podunk conference - it's a big name, prestigious conference.
Let's assume that the reviewer is correct and the key idea of the paper is not new.
(edit: parent subsequently reports they mention the precedent work in their paper. This is a different scenario. I leave the rest of my comment which was about the scenario as originally described)
What's the alternative? You want to present an idea that isn't novel?
> how does (weak accept+accept+reject) == reject
The editor or program chair that makes the decision knows from Reviewer 3 that the paper is not novel. If the paper's value is based on novelty, it has to be rejected.
If the paper has some value other than novelty, edit it to include the key citation and very carefully explain the contribution of the paper, and submit somewhere else. If there is no such contribution, tough. Do a better literature survey next time.
> it's a big name, prestigious conference
Imagine how you would feel if you were the author of the obscure paper and someone who didn't bother to read it got your idea published in a prestigious conference.
> The presumptive necessity of "novelty" is what enables such extremes.
Did your paper claim novelty? That claim obviously would need to be removed. Without novelty, what is the contribution? Your paper needs to explain.
If the world needs to know about the previous paper, so that the idea is recognized, write a review or case study that highlights it.
people will bend over backwards to accomodate/justify the stupid/biased/rotten practices in peer review.
> Let's assume that the reviewer is correct and the key idea of the paper is not new.
umm why are we assuming this? i mentioned the other project in my paper and pointed out the distinctions. even so:
> You want to present an idea that isn't novel?
yes. explain to me in very specific detail what the issue is with that in 2023 where page limits are an anachronism? don't have room at the conference venue itself - i'm fine recording the presentation and putting it up on youtube.
> The editor or program chair that makes the decision knows from Reviewer 3 that the paper is not novel.
reviewer 3 is demonstrably wrong/lying - my project is distinct and gets more use/support (it has been merged into the upstream core ecoystem project).
> If the paper has some value other than novelty, edit it to include the key citation and very carefully explain the contribution of the paper
done and done but there's a crucial component you're eliding over - reviewer has to actually read the paper.
> Do a better literature survey next time.
lol
> Imagine how you would feel if you were the author of the obscure paper and someone who didn't bother to read it got your idea published in a prestigious conference.
there's no paper. it's a github repo.
you see the absurdity grows and grows until one can't even believe that it's possible such a horrible miscarriage of academic justice/honesty was committed. but i have all the receipts as the kids say and the real miscarriage is it's worthless to complain because the entire practice is complete and utter bs.
i also didn't tell you what my favorite color is nor my wife's maiden name - maybe you shouldn't be such a presumptuous condescending <word>.
again if you don't believe me, it doesn't matter. the paper exists and the reviews exist and the email exchange with the PC exists and i have shown all of these things to my own committee (and if i knew who you were and trusted you, i would show them to you as well).
> I've edited my comment to reflect the new information.
your edits further cements your position as presumptuous and obliging of a process that warrants/merits no such thing.
They dug it up as a cause for rejection, like one digs up skeletons in someone's closet to discredit them.
I'm pretty exhausted from dealing with this and now explaining it to you but I have an update: I've just now gotten an email from the PC (after emailing them yesterday pointing out the flaws in reviewer 3's reasoning) saying I'm welcome to come present (and highlight the differences) in a "lightning talk" that won't be published in the proceedings. Looks like they're going with the "do it for the exposure" influencer model. I've declined the "opportunity".
Interesting concept to handle missing data using Trinary decision trees. At a high-level, it seems reminiscent of Multiple Imputation in randomForests which could address missingness. Though the Trinary tree takes a different approach by not presuming the missing values harbor any significant information about the response. It's intriguing that it shines in MCAR settings, but falls short with Informative Missingness.
> "Notably, the Trinary tree outperforms its peers in MCAR settings, especially when data is only missing out-of-sample, while lacking behind in IM settings."
This somewhat mirrors the behavior of early imputation strategies. One must ponder, however, how the Trinary tree would perform vis-a-vis older methods like CART's surrogate splits or C4.5's probabilistic splits for handling missing values. These older methods were crafted with an intuition somewhat similar to the Trinary tree.
It's also great to see the amalgamation of Trinary tree with the Missing In Attributes approach into the TrinaryMIA tree. But the efficacy of this hybrid model isn't completely surprising. MIA has historically shown resilience in diverse missing data scenarios, and combining that with the Trinary's approach could harmonize their strengths.
What would be really enticing is to see if the essence of the Trinary decision tree can be injected into boosting models like XGBoost or LightGBM. Since these models are notorious for their treatment of missing values, maybe there's some potential symbiosis there?
It isn't even that much of a speed hit using the classical sorting CART implementation. However xgboost and ligthgbm use histogram based approximate sorting which might be harder to adapt in a performant way. And certainly the code will be a lot messier.
Came here to cite your work, I even mention "CloudForest" in my slides still as "an interesting implementation that is also capable of handling NANs in DTs in a slightly different way." Crazy this has already been 10 years.
Yes, but trinary honestly makes more sense in contemporary English. I'm not sure whether preserving Greco-Latin roots or selecting for usability is more important here.
Update: there's also the "Ternary tree", the "Ternary search tree", the "Ternary heap", and the "Ternary numeral system". Most have even Wikipedia articles. There is no Wikipedia article for anything "Trinary" related to computer science.
Either encode with nil or if space is not a problem, use an aditional field that stores a bitmap.
Which allows to use structures with pointers without making nil semantics fuzzy.
Can even represent that bitmap as a vector and create a presence operator that is essentially a kind of intersection operation.
25 comments
[ 319 ms ] story [ 439 ms ] threadFor instance Datalog was developed in 1986 but when I got interested it in 2006 I found it was really obscure, by 2016 there was some literature and a few good implementations and it is rolling on.
There were those old "rules engines": many of the classic expert systems were developed before the famous RETE algorithm and even then there were RETE implementations that were terribly slow because they didn't use hashtables! When that stuff was fashionable people thought 10,000 rules were a lot, today it is more like 10,000,000 rules are a lot and I'd credit that more to algorithms being better than the computers being more powerful.
Similarly since the mid 2000's there has been a revolution in SAT and SMT solving which goes back to the very early (1960s) work on trying to solve problems w/ logic it is just now we succeed instead of fail. (It's a lot like the neural network revolution and used the same method of competition as seen in TREC and visual recognition.)
And i got the idea from a lab mate, Timo Erkkila's RF-ACE project though neither of us thought it was a particularly novel idea.
My paper got rejected because a reviewer dug up some vaguely similar thing from 3 years ago and claimed it is "well used and well supported" even though the repo has had 3 commits in 3 years. Oh that's an extreme case you say? The presumptive necessity of "novelty" is what enables such extremes.
Like the schizophrenia of modern society is hilarious: 15 brands of potato chips is good (competition in the market is healthy) but one molecule in common (in science) and you're a failure.
is there any way to appeal such decision? it is not healthy when single person can block lots of hard work..
note: some venues have a rebuttal phase which 99% doesn't make a difference anyway but this one didn't.
note2: if you're thinking i got rejected for some other reason: i got (weak accept, accept, reject) and the lone reject came from the wackadoodle. btw how does (weak accept+accept+reject) == reject? see my earlier comment about how the scores don't matter - i emailed the PC and they said they use the scores to resolve a failure to come to a consensus by the reviewers (which is 180deg the opposite of how a scoring system is used - scoring systems "objectively" measure and then humans break ties).
note3: this isn't a podunk conference - it's a big name, prestigious conference.
(edit: parent subsequently reports they mention the precedent work in their paper. This is a different scenario. I leave the rest of my comment which was about the scenario as originally described)
What's the alternative? You want to present an idea that isn't novel?
> how does (weak accept+accept+reject) == reject
The editor or program chair that makes the decision knows from Reviewer 3 that the paper is not novel. If the paper's value is based on novelty, it has to be rejected.
If the paper has some value other than novelty, edit it to include the key citation and very carefully explain the contribution of the paper, and submit somewhere else. If there is no such contribution, tough. Do a better literature survey next time.
> it's a big name, prestigious conference
Imagine how you would feel if you were the author of the obscure paper and someone who didn't bother to read it got your idea published in a prestigious conference.
> The presumptive necessity of "novelty" is what enables such extremes.
Did your paper claim novelty? That claim obviously would need to be removed. Without novelty, what is the contribution? Your paper needs to explain.
If the world needs to know about the previous paper, so that the idea is recognized, write a review or case study that highlights it.
> Let's assume that the reviewer is correct and the key idea of the paper is not new.
umm why are we assuming this? i mentioned the other project in my paper and pointed out the distinctions. even so:
> You want to present an idea that isn't novel?
yes. explain to me in very specific detail what the issue is with that in 2023 where page limits are an anachronism? don't have room at the conference venue itself - i'm fine recording the presentation and putting it up on youtube.
> The editor or program chair that makes the decision knows from Reviewer 3 that the paper is not novel.
reviewer 3 is demonstrably wrong/lying - my project is distinct and gets more use/support (it has been merged into the upstream core ecoystem project).
> If the paper has some value other than novelty, edit it to include the key citation and very carefully explain the contribution of the paper
done and done but there's a crucial component you're eliding over - reviewer has to actually read the paper.
> Do a better literature survey next time.
lol
> Imagine how you would feel if you were the author of the obscure paper and someone who didn't bother to read it got your idea published in a prestigious conference.
there's no paper. it's a github repo.
you see the absurdity grows and grows until one can't even believe that it's possible such a horrible miscarriage of academic justice/honesty was committed. but i have all the receipts as the kids say and the real miscarriage is it's worthless to complain because the entire practice is complete and utter bs.
You said the reviewer dug it up. You didn't say you cited it. That's a very different situation!
I've edited my comment to reflect the new information.
again if you don't believe me, it doesn't matter. the paper exists and the reviews exist and the email exchange with the PC exists and i have shown all of these things to my own committee (and if i knew who you were and trusted you, i would show them to you as well).
> I've edited my comment to reflect the new information.
your edits further cements your position as presumptuous and obliging of a process that warrants/merits no such thing.
The difference between
"My paper got rejected because a reviewer dug up some vaguely similar thing from 3 years ago"
and
"I mentioned the other project in my paper and pointed out the distinctions."
seems to me to be quite important. After hearing the first by itself, the second came as a surprise.
I'm pretty exhausted from dealing with this and now explaining it to you but I have an update: I've just now gotten an email from the PC (after emailing them yesterday pointing out the flaws in reviewer 3's reasoning) saying I'm welcome to come present (and highlight the differences) in a "lightning talk" that won't be published in the proceedings. Looks like they're going with the "do it for the exposure" influencer model. I've declined the "opportunity".
> "Notably, the Trinary tree outperforms its peers in MCAR settings, especially when data is only missing out-of-sample, while lacking behind in IM settings."
This somewhat mirrors the behavior of early imputation strategies. One must ponder, however, how the Trinary tree would perform vis-a-vis older methods like CART's surrogate splits or C4.5's probabilistic splits for handling missing values. These older methods were crafted with an intuition somewhat similar to the Trinary tree.
It's also great to see the amalgamation of Trinary tree with the Missing In Attributes approach into the TrinaryMIA tree. But the efficacy of this hybrid model isn't completely surprising. MIA has historically shown resilience in diverse missing data scenarios, and combining that with the Trinary's approach could harmonize their strengths.
What would be really enticing is to see if the essence of the Trinary decision tree can be injected into boosting models like XGBoost or LightGBM. Since these models are notorious for their treatment of missing values, maybe there's some potential symbiosis there?
It isn't even that much of a speed hit using the classical sorting CART implementation. However xgboost and ligthgbm use histogram based approximate sorting which might be harder to adapt in a performant way. And certainly the code will be a lot messier.
Guess that developer was right all along...
Update: there's also the "Ternary tree", the "Ternary search tree", the "Ternary heap", and the "Ternary numeral system". Most have even Wikipedia articles. There is no Wikipedia article for anything "Trinary" related to computer science.
Can even represent that bitmap as a vector and create a presence operator that is essentially a kind of intersection operation.