Stefan Neumann
@neumannstefan.com
Assistant professor @ TU Wien, associate faculty @ Complexity Science Hub.
Previously: KTH, Brown, Uni Wien.
I study algorithms for data science and social-network analysis. Connecting theory 🤝 practice.
More info: https://neumannstefan.com.
Previously: KTH, Brown, Uni Wien.
I study algorithms for data science and social-network analysis. Connecting theory 🤝 practice.
More info: https://neumannstefan.com.
The new EATCS Bulletin #147 is available!
eatcs.org/images/bulle...
In the TCS on the Web Column, I talked to the maintainers of the TCS Blog Aggregator: Nima Anari, Arnab Bhattacharyya and
Gautam Kamath.
It was a very fun interview!
@gautamkamath.com @schmiste-ch.bsky.social
eatcs.org/images/bulle...
In the TCS on the Web Column, I talked to the maintainers of the TCS Blog Aggregator: Nima Anari, Arnab Bhattacharyya and
Gautam Kamath.
It was a very fun interview!
@gautamkamath.com @schmiste-ch.bsky.social
October 30, 2025 at 12:44 PM
The new EATCS Bulletin #147 is available!
eatcs.org/images/bulle...
In the TCS on the Web Column, I talked to the maintainers of the TCS Blog Aggregator: Nima Anari, Arnab Bhattacharyya and
Gautam Kamath.
It was a very fun interview!
@gautamkamath.com @schmiste-ch.bsky.social
eatcs.org/images/bulle...
In the TCS on the Web Column, I talked to the maintainers of the TCS Blog Aggregator: Nima Anari, Arnab Bhattacharyya and
Gautam Kamath.
It was a very fun interview!
@gautamkamath.com @schmiste-ch.bsky.social
STOC'26 will provide an *optional* pre-submission feedback based on Gemini. The model is optimized for checking mathematical rigor.
Looks interesting. They also post example outputs and comments on them.
Announcement: acm-stoc.org/stoc2026/sto...
Example reviews: www.cs.cmu.edu/~dwoodruf/st...
Looks interesting. They also post example outputs and comments on them.
Announcement: acm-stoc.org/stoc2026/sto...
Example reviews: www.cs.cmu.edu/~dwoodruf/st...
October 25, 2025 at 2:20 PM
STOC'26 will provide an *optional* pre-submission feedback based on Gemini. The model is optimized for checking mathematical rigor.
Looks interesting. They also post example outputs and comments on them.
Announcement: acm-stoc.org/stoc2026/sto...
Example reviews: www.cs.cmu.edu/~dwoodruf/st...
Looks interesting. They also post example outputs and comments on them.
Announcement: acm-stoc.org/stoc2026/sto...
Example reviews: www.cs.cmu.edu/~dwoodruf/st...
Such a strong growth in submissions over the past few years (probably because of LLMs?). 38% more than two years ago. Wow.
An acceptance rate of less than 10% is really tough.
Good luck to everyone who submitted!
An acceptance rate of less than 10% is really tough.
Good luck to everyone who submitted!
October 16, 2025 at 1:57 PM
Such a strong growth in submissions over the past few years (probably because of LLMs?). 38% more than two years ago. Wow.
An acceptance rate of less than 10% is really tough.
Good luck to everyone who submitted!
An acceptance rate of less than 10% is really tough.
Good luck to everyone who submitted!
I was just briefly scrolling over the latest ACM Career News newsletter and I find it quite funny that it contains a picture of the German Federal Minister of Defense, Boris Pistorius.
October 15, 2025 at 7:27 AM
I was just briefly scrolling over the latest ACM Career News newsletter and I find it quite funny that it contains a picture of the German Federal Minister of Defense, Boris Pistorius.
Check out the latest edition of the EATCS Bulletin, featuring @timroughgarden.bsky.social on how he built one of the most successful TCS YouTube channels—with over 1 million views.
The other articles are also very exciting. Have a look!
www.eatcs.org/images/bulle...
The other articles are also very exciting. Have a look!
www.eatcs.org/images/bulle...
June 30, 2025 at 6:39 AM
Check out the latest edition of the EATCS Bulletin, featuring @timroughgarden.bsky.social on how he built one of the most successful TCS YouTube channels—with over 1 million views.
The other articles are also very exciting. Have a look!
www.eatcs.org/images/bulle...
The other articles are also very exciting. Have a look!
www.eatcs.org/images/bulle...
To remove noise, one can use the approach of "Denser than the densest subgraph" by Tsourakakis, Bonchi, Gionis, Gullo, Tsiarli. They change the objective function from density |E[S]| / |S| to measure how clique-like the subgraph is: |E[S]| - (|S| choose 2).
This gives much more coherent clusters.
This gives much more coherent clusters.
![Screenshot of the following text:
Therefore, in practice the densest subgraph tends to return medium- to large-sized subgraphs (w.r.t. the number of nodes) but they are not very clique-like. Some works have therefore tried to modify the densest subgraph objective function to return smaller, more clique-like structures. Tsourakakis et al. [8] did this by considering the density function |E[S]|−α (|S| choose 2) which is non-negative if G[S] contains more than an α-fraction of all its possible edges. This penalizes subgraphs with a lot
of vertices and empirically we can observe that it tends to return smaller, more coherent subgraphs.](https://cdn.bsky.app/img/feed_thumbnail/plain/did:plc:tnkvpa7fzk6rdtuqjcrn2pdu/bafkreih35lk6f4khg2lixzqncfeqtrwtex4k47lpibr3hfk6bxbhifntcy@jpeg)
March 11, 2025 at 9:47 AM
To remove noise, one can use the approach of "Denser than the densest subgraph" by Tsourakakis, Bonchi, Gionis, Gullo, Tsiarli. They change the objective function from density |E[S]| / |S| to measure how clique-like the subgraph is: |E[S]| - (|S| choose 2).
This gives much more coherent clusters.
This gives much more coherent clusters.
Then the next clusters are a bit more noisy (arguably the first two are also noisy, but ok it's vanilla densest subgraph) also make some sense after eyeballing:
Cluster 3: Middle East + India
Cluster 4: Russia + countries in the vicinity
Cluster 5: South America + Australia
Cluster 3: Middle East + India
Cluster 4: Russia + countries in the vicinity
Cluster 5: South America + Australia
March 11, 2025 at 9:39 AM
Then the next clusters are a bit more noisy (arguably the first two are also noisy, but ok it's vanilla densest subgraph) also make some sense after eyeballing:
Cluster 3: Middle East + India
Cluster 4: Russia + countries in the vicinity
Cluster 5: South America + Australia
Cluster 3: Middle East + India
Cluster 4: Russia + countries in the vicinity
Cluster 5: South America + Australia
Today I will be teaching densest subgraph in my course. So, I implemented the greedy peeling algorithm ("Charikar's algorithm") and ran it on some flight dataset from @netzschleuder.skewed.de.
The first two clusters nicely correspond to clusters in (1) Europe+US and (2) China.
The first two clusters nicely correspond to clusters in (1) Europe+US and (2) China.
March 11, 2025 at 9:31 AM
Today I will be teaching densest subgraph in my course. So, I implemented the greedy peeling algorithm ("Charikar's algorithm") and ran it on some flight dataset from @netzschleuder.skewed.de.
The first two clusters nicely correspond to clusters in (1) Europe+US and (2) China.
The first two clusters nicely correspond to clusters in (1) Europe+US and (2) China.
The (soft) deadline for this position is on Friday.
February 17, 2025 at 7:16 AM
The (soft) deadline for this position is on Friday.
The openness of Bluesky is great, and it is nice to see the creativity it encourages.
blueskyroast.com/roast/stefan...
blueskyroast.com/roast/stefan...
December 4, 2024 at 5:15 PM
The openness of Bluesky is great, and it is nice to see the creativity it encourages.
blueskyroast.com/roast/stefan...
blueskyroast.com/roast/stefan...
How do you make a theoretical computer scientist happy? By giving them a large whiteboard. 😍
April 17, 2024 at 12:21 PM
How do you make a theoretical computer scientist happy? By giving them a large whiteboard. 😍
Intriguing paper by Jon Kleinberg and Sendhil Mullainathan on arxiv today.
They look justifying the effectiveness of LLMs theoretically. In a nutshell, they show that, even though *detecting* languages is difficult, *generating* from languages can be done asymptotically.
(link below)
They look justifying the effectiveness of LLMs theoretically. In a nutshell, they show that, even though *detecting* languages is difficult, *generating* from languages can be done asymptotically.
(link below)
April 11, 2024 at 9:25 AM
Intriguing paper by Jon Kleinberg and Sendhil Mullainathan on arxiv today.
They look justifying the effectiveness of LLMs theoretically. In a nutshell, they show that, even though *detecting* languages is difficult, *generating* from languages can be done asymptotically.
(link below)
They look justifying the effectiveness of LLMs theoretically. In a nutshell, they show that, even though *detecting* languages is difficult, *generating* from languages can be done asymptotically.
(link below)
Very interesting talk today by Jingren Zhou on building large language models at Alibaba.
@caimlTUwien @tu_wien
@caimlTUwien @tu_wien
March 14, 2024 at 11:05 AM
Very interesting talk today by Jingren Zhou on building large language models at Alibaba.
@caimlTUwien @tu_wien
@caimlTUwien @tu_wien
Today is my final day at KTH. It was a great time and I will definitely miss it.
I am most thankful to Aris Gionis for all of his support during this time, the people in his group, and the colleagues in the TCS division at large.
Now I am heading back to Vienna (more info soon).
I am most thankful to Aris Gionis for all of his support during this time, the people in his group, and the colleagues in the TCS division at large.
Now I am heading back to Vienna (more info soon).
January 26, 2024 at 3:11 PM
Today is my final day at KTH. It was a great time and I will definitely miss it.
I am most thankful to Aris Gionis for all of his support during this time, the people in his group, and the colleagues in the TCS division at large.
Now I am heading back to Vienna (more info soon).
I am most thankful to Aris Gionis for all of his support during this time, the people in his group, and the colleagues in the TCS division at large.
Now I am heading back to Vienna (more info soon).
🚨🚨🚨
Very important iPadOS update for those of us living in Scandinavia.
Very important iPadOS update for those of us living in Scandinavia.
November 22, 2023 at 5:22 PM
🚨🚨🚨
Very important iPadOS update for those of us living in Scandinavia.
Very important iPadOS update for those of us living in Scandinavia.