ICML 2018 Accepted Papers Stats

ICML 2018 is being held in Stockholmsmässan, Stockholm Sweden

The Thirty-fifth International Conference on Machine Learning (ICML) is being held this week in Stockholm Sweden. ICML is one of the most prestigious AI-related academic conferences and is a highly coveted place to publish research papers.

Back in December, I wrote a post detailing some stats I compiled about the papers that were accepted at NIPS 2017. This time, with the help of Ryan Pilgrim, a data scientist at Infinia ML, I’m doing the same for ICML. My interest in this topic is at least partly related to the fact that I’m the CEO of one of the companies with published work at ICML. Our chief scientist, Dr. Lawrence Carin, is one of the most prolific AI researchers in the world so it’s interesting to see the trends of other researchers and companies that are active in the field.

This year, 621 papers were accepted out of 2,473 submissions for a 25.1% acceptance rate. That’s 47% more papers submitted than last year’s 1,676 (with a similar 25% acceptance rate).

It’s worth noting that analysis such as what you’ll find below is directionally accurate, but can vary from other sources. That’s because the ICML website does not make it easy to do the analysis and requires some NLP hacks and workarounds. Just even consolidating the variations on institution names will lead to variability. There are a number of different ways companies like “Google” are listed as the institution for a researcher. Some human discretion is needed when trying to consolidate the various spellings.

Author Stats

This year, Tong Zhang of Tencent Labs was named on the most papers, eight in total. Lawrence Carin (Vice Provost of Research at Duke and Chief Scientist at Infinia ML) shows up in second place with seven papers. As I talked about in the NIPS post, Lawrence Carin oversees a large ML lab at Duke that is responsible for many papers.

Most papers by author:
1. tong zhang: 8 (tencent labs)
2. lawrence carin: 7 (duke / infinia ml)
3. jun zhu: 6 (tsinghua university)
3. quanquan gu: 6 (uva)
3. le song: 6 (ga tech)
3. remi munos: 6 (nyu / courant institute)
3. pieter abbeel: 6 (uc berkeley)
3. sergey levine: 6 (uc berkeley)
4. bernhard schölkopf: 5 (max planck institute)
4. eric xing: 5 (carnegie mellon)
4. amin karbasi: 5 (yale)
4. shimon whiteson: 5 (oxford)

Besides Lawrence Carin, the only other person that had five or more papers at NIPS is Le Song of Georgia Tech.

Institution Stats

Below are the top 58 institutions that had six or more papers. Obviously, Google stands out. We kept the counts separate across Google, Google AI, and DeepMind, but if you combined them all, Google has an author on 82 papers, over 13% of the total. No other group is even close. Facebook has made a bigger effort at ICML with 21 papers versus only 11 at NIPS.

UC Berkeley has the most papers from a university, but two fewer than they had at NIPS. CMU had a whopping 48 papers at NIPS, but “only” 32 at ICML.

Most papers by institution:
google: 43
berkeley: 33
deepmind: 33
cmu: 32
stanford: 29
mit: 28
microsoft: 28
google ai: 24
facebook: 21
oxford: 20
princeton: 20
inria: 18
epfl: 17
eth zurich: 16
cornell: 16
max planck institute: 15
ut austin: 15
cambridge: 15
ucla: 15
university of toronto: 15
columbia: 13
tsinghua university: 12
amazon: 11
harvard: 11
tencent: 11
usc: 11
georgia tech: 11
duke: 11
peking university: 10
purdue: 10
riken: 10
uber: 10
yale: 9
nyu: 9
university of wisconsin: 9
mila: 8
paristech: 8
imperial college london: 8
johns hopkins university: 8
uiuc: 8
university of virginia: 8
rice: 8
ibm: 7
technion: 7
openai: 7
vector institute: 7
university of washington: 7
cnrs: 7
caltech: 7
google research: 7
university of chicago: 6
university of tubingen: 6
alan turing institute: 6
northwestern: 6
university of montreal: 6
uc davis: 6
university of pennsylvania: 6
university of edinburgh: 6

Here are top 24 universities by papers published.

Most papers by universities:
berkeley: 33
cmu: 32
stanford: 29
mit: 28
oxford: 20
princeton: 20
epfl: 17
eth zurich: 16
cornell: 16
ut austin: 15
cambridge: 15
ucla: 15
university of toronto: 15
columbia: 13
tsinghua university: 12
harvard: 11
usc: 11
georgia tech: 11
duke: 11
peking university: 10
purdue: 10
yale: 9
nyu: 9
university of wisconsin: 9

Below is a list of paper totals by commercial entities. If you count the papers from all the Infinia ML employees, including Lawrence Carin, Ricardo Henao, and Hongteng Xu, my company is tied for 9th overall with IBM and ahead of Apple, Disney, and Element AI.

Total papers by commercial institution:
google: 43
deepmind: 33
microsoft: 28
google ai: 24
facebook: 21
tencent: 11
amazon: 11
uber: 10
infinia ml: 7
google research: 7
ibm: 7
disney research: 4
apple: 3
yahoo research: 3
data61: 3
ant financial: 2
ntt: 2
criteo: 2
yandex: 2
toyota technological institute: 2
prowler.io: 2
element ai: 2
rj research consulting: 1
qihoo/360: 1
alibaba group: 1
mass general hospital: 1
hellebore capital limited: 1
preferred networks, inc.: 1
dolores technologies: 1
hikvision research institue: 1
mitsubishi electric research labs: 1
labsix: 1
covariant.ai: 1
nvidia: 1
bosch: 1
zalando research: 1
d. e. shaw \& co.: 1
intel: 1
twitter: 1
curai: 1
two sigma investments: 1
bae systems fast labs: 1
aitrics: 1
recruit technologies co. ltd.: 1
siemens ag: 1
ea seed: 1
autodesk: 1
4paradigm: 1
petuum inc: 1
philips research: 1
sentient technologies: 1
snap: 1
orange labs: 1
qualcomm india private limited: 1
rwth: 1
ign: 1
axa im chorus: 1
baidu: 1
afresh technologies: 1
d-wave: 1
telefonica: 1
huawei: 1
saudi aramco: 1
rolls-royce: 1
bytedance inc.: 1
quadrant.ai: 1
magic leap: 1
guo: 1

ICML + NIPS

Since I had the stats for both NIPS 2017 and ICML 2018, I wondered which authors published the most between the two conferences. Lawrence Carin has the most by a long-shot. It’s uncommon to publish multiple papers at one of these conferences — those who publish multiple at both conferences make up a very small group. Here are the authors with at least eight papers between the two conferences:

Most NIPS + ICML papers:
lawrence carin (duke university): 17
tong zhang (tencent labs): 12
le song (georgia institute of technology): 11
michael jordan (university of california, berkeley): 9
nicolas heess (deepmind): 8
razvan pascanu (deepmind): 8