J. Shane Culpepper

476 posts

J. Shane Culpepper

J. Shane Culpepper

@sculpepper

Computer Science Research

Australia Katılım Nisan 2008
269 Takip Edilen383 Takipçiler
J. Shane Culpepper retweetledi
Ratan Sebastian
Ratan Sebastian@ratansebastian·
At #SIGIR2022 and don't know who to connect with? I tried to implement some of @sculpepper 's advise on this and created SIGIR-connect.Find conference presenters who have collaborated with researchers you're interested in talking to: fierce-plains-57959.herokuapp.com .Happy connecting!
English
0
1
1
0
J. Shane Culpepper
J. Shane Culpepper@sculpepper·
Boy some folks are attached to their paper appendix.
English
1
0
1
0
J. Shane Culpepper
J. Shane Culpepper@sculpepper·
@srchvrs There are few things in life more rewarding than spending all your time fixing other people’s mistakes. There. I’m practicing my Twitter sarcasm too!
English
0
0
1
0
Leo Boytsov
Leo Boytsov@srchvrs·
@sculpepper PS: professional devs routinely deal with codebases having hundreds of Ks and/or Ms lines of code. That is, given a 1M LOC poorly documented codebase, devs often have to pinpoint and fix bugs and inefficiencies.
English
1
0
0
0
J. Shane Culpepper
J. Shane Culpepper@sculpepper·
Nothing humbles me quicker than writing code. I find that I spend 90% of the time staring at a snippet I just wrote and repeatedly asking myself “What did I do wrong here? It sure looks right.”
English
1
0
5
0
J. Shane Culpepper
J. Shane Culpepper@sculpepper·
@srchvrs Under appreciated for sure. But it’s pretty hard to do empirical science using data without it and that’s what most of us seem to be claiming we do 🙂.
English
1
0
1
0
J. Shane Culpepper retweetledi
Mark Sanderson
Mark Sanderson@IR_oldie·
We are looking for a research fellow to join us to research new approaches to handling complex information needs. Applications close on Wednesday, 26th of May 2021. rmit.edu.au/careers/curren…
English
1
11
18
0
J. Shane Culpepper
J. Shane Culpepper@sculpepper·
@srchvrs @jobergum @jpountz @JamieCallan In or WSDM experiments jass was never stat sig worse, but we configured it that way and used the same ranking model. Change the model and maybe it is all the time. If you really care, know your data and choose whatever consistently gives you the results you want to achieve.
English
1
0
0
0
J. Shane Culpepper
J. Shane Culpepper@sculpepper·
@srchvrs @jobergum @jpountz @JamieCallan I would argue 10-15% degradation is also a mischaracterization of what jass is capable of me. You are carrying the torch for ANN and that’s fine. I am not arguing superiority for any of them. They all filll a gap and none of them are clearly superior all the time.
English
1
0
0
0
Leo Boytsov
Leo Boytsov@srchvrs·
We just had a lovely discussion of how fast or slow Lucene actually is. And one outcome is that BM-WAND seems to be a mixed success. I get 20 ms vs 15 ms on Lucene 8 vs Lucene 7 (@MSMarcoAI passages) for top-1K and 8 ms vs 11 ms for top-10 retrieval. @sculpepper thoughts?
Jimmy Lin@lintool

From the "ivory tower" to the "real-world": the story of how block-max WAND made its way into Lucene 8 by @jpountz et al. is told in a #ecir2020 preprint: cs.uwaterloo.ca/~jimmylin/publ… Lessons for academics seeking to achieve research impact?

English
1
1
8
0
J. Shane Culpepper
J. Shane Culpepper@sculpepper·
@srchvrs @jobergum @jpountz @JamieCallan JASS on the other hand gives exact control. You can say exactly how many postings to score and have and exact SLA — at the cost of exactness. But in practice you lose very little. All IR BOW ranking score are heuristic anyway. See our 2018 WSDM paper for more.
English
1
0
1
0
J. Shane Culpepper
J. Shane Culpepper@sculpepper·
@srchvrs @jobergum @jpountz @JamieCallan I’m not sure that was exactly what we said but generally it works very very well — if there are one or two pivot terms driving the traversal. There are paranoid cases where bmw and max score degrade badly — all common terms with close max impact scores.
English
1
0
1
0
J. Shane Culpepper
J. Shane Culpepper@sculpepper·
@jobergum @srchvrs @jpountz @MSMarcoAI Indeed. There are still tail queries with BMW that can be slower than exhaustive. They are rare but exist. It also depends how many you return. The smaller you candidate set, the faster. Still good to see such things make it mainstream — definitely a net positive to users.
English
1
0
0
0
J. Shane Culpepper
J. Shane Culpepper@sculpepper·
@CharlotteHase Maybe a FAANG should use some of that mountain of cash they’ve accumulated off of us to buy all the journal and conference publishers out, and make it all free and open access like it should be instead of asking us to do it?
English
0
0
5
0
Tetsuya Sakai (酒井哲也)
Tetsuya Sakai (酒井哲也)@tetsuyasakai·
There will be an open table discussion at #wsdm2021 which I won't be able to join. But here is a summary of my tweets regarding the interesting questions raised by Ferrante/Ferro/Fuhr in their arxiv paper. @frrncl @julian_urbano waseda.box.com/wsdm2021openta…
Tetsuya Sakai (酒井哲也) tweet media
Tetsuya Sakai (酒井哲也)@tetsuyasakai

@pcastells @ian_soboroff @IR_oldie @julian_urbano @frrncl @iadh @srchvrs Possible values RR can take (prec@1st reldoc found) are clearly not equispaced, but maybe they could be a subset of an underlying interval scale representing SERP utility? (Unlike your discrete unit example, the scale is hidden.) A related discussion in sigir.org/wp-content/upl…

English
1
1
12
0
J. Shane Culpepper retweetledi
Santiago
Santiago@svpino·
25 popular libraries and frameworks for building machine and deep learning applications. Covering: ▫️ Data analysis and processing ▫️ Visualizations ▫️ Computer Vision ▫️ Natural Language Processing ▫️ Reinforcement Learning ▫️ Optimization A mega-thread. 🐍 🧵👇
English
25
320
1.3K
0
J. Shane Culpepper retweetledi
Computer Science
Computer Science@CompSciFact·
'Just because you've implemented something doesn't mean you understand it.' -- Brian Cantwell Smit
English
10
108
588
0
J. Shane Culpepper
J. Shane Culpepper@sculpepper·
@srchvrs @lintool I think history shows otherwise. In many settings opportunities are limited. More people don’t just make more. There will always be people who are willing to outdo you when they are motivated to achieve their goals in every scarcity scenario. Nothing fair about it.
English
1
0
0
0
Leo Boytsov
Leo Boytsov@srchvrs·
@sculpepper @lintool This is based on the assumption of a zero-sum game. But history shows it is a false assumption. Short term, sum games zero sum, but new people create new opportunities. This is likely a more efficient way to make progress.
English
2
0
0
0
Jimmy Lin
Jimmy Lin@lintool·
Work-life balance fundamentally means choice, including the choice to not have a life outside work. It's not the life I would choose right now because I have a family and greatly value time with my family. But what's the issue with others pursuing that path?
English
2
0
25
0
J. Shane Culpepper
J. Shane Culpepper@sculpepper·
@srchvrs @lintool We might all be fools for wanting it, but it won’t change the fact that we do or that many others do too and only one is available.
English
0
0
0
0
J. Shane Culpepper
J. Shane Culpepper@sculpepper·
@srchvrs @lintool The pigeon hole process applies in life. It’s just a reality. If 200 people apply for something and really need it, and there is only one of them (a high level job for example, or a budget amount of money), it can’t be anything but competitive.
English
2
0
0
0

Keşfet

@srchvrs @jobergum @jpountz @JamieCallan @MSMarcoAI @CharlotteHase @IR_oldie @ian_soboroff