@julianpollmann.bsky.social
bike maniac & research software engineer at HSD, @zdd-hsd.bsky.social
Mostly #NLP, #AI, #SoftwareEngineering
Mostly #NLP, #AI, #SoftwareEngineering
Reposted
Our fingerprint analysis work has now been published 🎉
--> link.springer.com/article/10.1...
Also check out our small helper library #chemap
--> github.com/matchms/chemap
(work together with @julianpollmann.bsky.social)
#cheminformatics #openscience #Python
--> link.springer.com/article/10.1...
Also check out our small helper library #chemap
--> github.com/matchms/chemap
(work together with @julianpollmann.bsky.social)
#cheminformatics #openscience #Python
Count your bits: fingerprint benchmarking to assess broad chemical space representation - Journal of Cheminformatics
Quantifying molecular similarity is a cornerstone of cheminformatics, underpinning applications from virtual screening and nearest-neighbor search to chemical space visualization and the evaluation of...
link.springer.com
June 30, 2026 at 8:53 AM
Our fingerprint analysis work has now been published 🎉
--> link.springer.com/article/10.1...
Also check out our small helper library #chemap
--> github.com/matchms/chemap
(work together with @julianpollmann.bsky.social)
#cheminformatics #openscience #Python
--> link.springer.com/article/10.1...
Also check out our small helper library #chemap
--> github.com/matchms/chemap
(work together with @julianpollmann.bsky.social)
#cheminformatics #openscience #Python
Thanks Canyon, the only Domain where I could escape the AI Hype was riding my Bike: now AI ist integrated into a Road Bike.
www.rennrad-news.de/news/canyon-...
www.rennrad-news.de/news/canyon-...
Canyon stellt Rennrad der Zukunft mit KI und Rundum-Radar vor
Canyons neues KI-Rennrad der Zukunft erkennt Gefahren, bevor sie entstehen, und könnte schwere Unfälle verhindern.
www.rennrad-news.de
June 19, 2026 at 7:40 PM
Thanks Canyon, the only Domain where I could escape the AI Hype was riding my Bike: now AI ist integrated into a Road Bike.
www.rennrad-news.de/news/canyon-...
www.rennrad-news.de/news/canyon-...
Reposted
New post: Tagging my blog with BERTopic and LLMs.
I suspect as token costs increase, we'll see more blended traditional ML/LLM systems. And, in general, they work really well!
vickiboykis.com/2026/05/18/t...
I suspect as token costs increase, we'll see more blended traditional ML/LLM systems. And, in general, they work really well!
vickiboykis.com/2026/05/18/t...
Tagging my blog posts with BERTopic and LLMs
LLMs mean you still need a human in the loop, but in a different part
vickiboykis.com
May 19, 2026 at 10:51 AM
New post: Tagging my blog with BERTopic and LLMs.
I suspect as token costs increase, we'll see more blended traditional ML/LLM systems. And, in general, they work really well!
vickiboykis.com/2026/05/18/t...
I suspect as token costs increase, we'll see more blended traditional ML/LLM systems. And, in general, they work really well!
vickiboykis.com/2026/05/18/t...
Really looking forward to #PyConDe & #PyData 2026 edition!
Some open Missions:
☑️ Soak in everything regarding Knowledge Graphs
☑️ Solve open mysteries in Python Multiprocessing
☑️ Meet new people and old friends
☑️ get a feel for the hot topics (except obvious ones)
get there (partially) & back by bike
Some open Missions:
☑️ Soak in everything regarding Knowledge Graphs
☑️ Solve open mysteries in Python Multiprocessing
☑️ Meet new people and old friends
☑️ get a feel for the hot topics (except obvious ones)
get there (partially) & back by bike
April 13, 2026 at 8:28 AM
Reposted
Just updated our preprint on benchmarking molecular fingerprints!
--> www.biorxiv.org/content/10.1...
Some key points
- Count fingerprints should be the default (not binary!)
- Unfolded fingerprints are often worth it.
- Larger radius Morgan or FCFP fingerprints are good first bets.
--> www.biorxiv.org/content/10.1...
Some key points
- Count fingerprints should be the default (not binary!)
- Unfolded fingerprints are often worth it.
- Larger radius Morgan or FCFP fingerprints are good first bets.
February 27, 2026 at 3:14 PM
Just updated our preprint on benchmarking molecular fingerprints!
--> www.biorxiv.org/content/10.1...
Some key points
- Count fingerprints should be the default (not binary!)
- Unfolded fingerprints are often worth it.
- Larger radius Morgan or FCFP fingerprints are good first bets.
--> www.biorxiv.org/content/10.1...
Some key points
- Count fingerprints should be the default (not binary!)
- Unfolded fingerprints are often worth it.
- Larger radius Morgan or FCFP fingerprints are good first bets.
Reposted
Work done with @julianpollmann.bsky.social at @zdd-hsd.bsky.social
Code:
- Central functionalities are now pip installable --> github.com/matchms/chemap
- Notebooks for experiments --> github.com/florian-huber/molecular_fingerprint_comparisons
#openscience #opensource #cheminformatics #python
Code:
- Central functionalities are now pip installable --> github.com/matchms/chemap
- Notebooks for experiments --> github.com/florian-huber/molecular_fingerprint_comparisons
#openscience #opensource #cheminformatics #python
GitHub - matchms/chemap: Library for computing molecular fingerprint based similarities as well as dimensionality-reduction-based chemical space visualizations.
Library for computing molecular fingerprint based similarities as well as dimensionality-reduction-based chemical space visualizations. - matchms/chemap
github.com
February 27, 2026 at 3:17 PM
Work done with @julianpollmann.bsky.social at @zdd-hsd.bsky.social
Code:
- Central functionalities are now pip installable --> github.com/matchms/chemap
- Notebooks for experiments --> github.com/florian-huber/molecular_fingerprint_comparisons
#openscience #opensource #cheminformatics #python
Code:
- Central functionalities are now pip installable --> github.com/matchms/chemap
- Notebooks for experiments --> github.com/florian-huber/molecular_fingerprint_comparisons
#openscience #opensource #cheminformatics #python
Reposted
Haftbefehl shows that Germany loves art born from alienation – just not the people who create it
Haftbefehl shows that Germany loves art born from alienation – just not the people who create it
A hit Netflix documentary about Germany’s favourite rapper demonstrates how popular the aesthetics of migrant life are – just as politicians debate how to remove it from inner cities
www.theguardian.com
November 12, 2025 at 1:43 PM
Haftbefehl shows that Germany loves art born from alienation – just not the people who create it
Reposted
New #matchms release (0.31)🚀
With functionalities that were on our TODO list for a looooong time: Flash Entropy and BLINK scores! The new "FlashSimilarity" allows computing modified cosine, spectral entropy etc., about 100x faster (or more if you use Linux).
#Python #opensource #massspec
With functionalities that were on our TODO list for a looooong time: Flash Entropy and BLINK scores! The new "FlashSimilarity" allows computing modified cosine, spectral entropy etc., about 100x faster (or more if you use Linux).
#Python #opensource #massspec
October 6, 2025 at 4:00 PM
New #matchms release (0.31)🚀
With functionalities that were on our TODO list for a looooong time: Flash Entropy and BLINK scores! The new "FlashSimilarity" allows computing modified cosine, spectral entropy etc., about 100x faster (or more if you use Linux).
#Python #opensource #massspec
With functionalities that were on our TODO list for a looooong time: Flash Entropy and BLINK scores! The new "FlashSimilarity" allows computing modified cosine, spectral entropy etc., about 100x faster (or more if you use Linux).
#Python #opensource #massspec
Reposted
4/4
We also highlight options for count fingerprints, such as log-counts and IDF weighted counts. The latter can be used to adjust the bit importance to a dataset of your choice.
An example use-case are chemical space visualizations.
Preprint: www.biorxiv.org/content/10.1...
We also highlight options for count fingerprints, such as log-counts and IDF weighted counts. The latter can be used to adjust the bit importance to a dataset of your choice.
An example use-case are chemical space visualizations.
Preprint: www.biorxiv.org/content/10.1...
June 23, 2025 at 9:22 AM
4/4
We also highlight options for count fingerprints, such as log-counts and IDF weighted counts. The latter can be used to adjust the bit importance to a dataset of your choice.
An example use-case are chemical space visualizations.
Preprint: www.biorxiv.org/content/10.1...
We also highlight options for count fingerprints, such as log-counts and IDF weighted counts. The latter can be used to adjust the bit importance to a dataset of your choice.
An example use-case are chemical space visualizations.
Preprint: www.biorxiv.org/content/10.1...
Reposted
3/4
A huge issue is bit collisions.
Fingerprints with a high bit occupation (RDKit, MAP4) often lead to (1) arbitrary misinterpretations, (2) shifts to high Tanimoto scores, (3) very different handling of small and large molecules.
--> Consider using sparse fingerprints!
--> Morgan >> MAP4 / RDKit
A huge issue is bit collisions.
Fingerprints with a high bit occupation (RDKit, MAP4) often lead to (1) arbitrary misinterpretations, (2) shifts to high Tanimoto scores, (3) very different handling of small and large molecules.
--> Consider using sparse fingerprints!
--> Morgan >> MAP4 / RDKit
June 23, 2025 at 9:22 AM
3/4
A huge issue is bit collisions.
Fingerprints with a high bit occupation (RDKit, MAP4) often lead to (1) arbitrary misinterpretations, (2) shifts to high Tanimoto scores, (3) very different handling of small and large molecules.
--> Consider using sparse fingerprints!
--> Morgan >> MAP4 / RDKit
A huge issue is bit collisions.
Fingerprints with a high bit occupation (RDKit, MAP4) often lead to (1) arbitrary misinterpretations, (2) shifts to high Tanimoto scores, (3) very different handling of small and large molecules.
--> Consider using sparse fingerprints!
--> Morgan >> MAP4 / RDKit
Reposted
2/4
We focused on weaknesses of the fingerprints.
Many show frequent duplicates, so same fingerprint for different compounds. Most problematic: this can include *very* different compounds ending up with identical fingerprints.
- MAP4 >> Morgan-type >> daylight
- count >> binary
#cheminformatics
We focused on weaknesses of the fingerprints.
Many show frequent duplicates, so same fingerprint for different compounds. Most problematic: this can include *very* different compounds ending up with identical fingerprints.
- MAP4 >> Morgan-type >> daylight
- count >> binary
#cheminformatics
June 23, 2025 at 9:22 AM
2/4
We focused on weaknesses of the fingerprints.
Many show frequent duplicates, so same fingerprint for different compounds. Most problematic: this can include *very* different compounds ending up with identical fingerprints.
- MAP4 >> Morgan-type >> daylight
- count >> binary
#cheminformatics
We focused on weaknesses of the fingerprints.
Many show frequent duplicates, so same fingerprint for different compounds. Most problematic: this can include *very* different compounds ending up with identical fingerprints.
- MAP4 >> Morgan-type >> daylight
- count >> binary
#cheminformatics
Reposted
New preprint out!
1/4
@julianpollmann.bsky.social and I went down several rabbit holes to assess some commonly used molecular fingerprints.
Bottom line: For large datasets, make an effort to select suitable settings. "We used Tanimoto" is not good enough.
--> www.biorxiv.org/content/10.1...
1/4
@julianpollmann.bsky.social and I went down several rabbit holes to assess some commonly used molecular fingerprints.
Bottom line: For large datasets, make an effort to select suitable settings. "We used Tanimoto" is not good enough.
--> www.biorxiv.org/content/10.1...
June 23, 2025 at 9:22 AM
New preprint out!
1/4
@julianpollmann.bsky.social and I went down several rabbit holes to assess some commonly used molecular fingerprints.
Bottom line: For large datasets, make an effort to select suitable settings. "We used Tanimoto" is not good enough.
--> www.biorxiv.org/content/10.1...
1/4
@julianpollmann.bsky.social and I went down several rabbit holes to assess some commonly used molecular fingerprints.
Bottom line: For large datasets, make an effort to select suitable settings. "We used Tanimoto" is not good enough.
--> www.biorxiv.org/content/10.1...
Reposted
Reposted
Wir dachten, KI würde uns irgendwann durch Terminator-Roboter versklaven. Aber wir werden durch Bequemlichkeit versklavt.
May 12, 2025 at 10:05 PM
Wir dachten, KI würde uns irgendwann durch Terminator-Roboter versklaven. Aber wir werden durch Bequemlichkeit versklavt.
Hello @pyconde.bsky.social
Curious to listen to some interesting Talks and meet new people 😀
Curious to listen to some interesting Talks and meet new people 😀
April 23, 2025 at 7:30 AM
Hello @pyconde.bsky.social
Curious to listen to some interesting Talks and meet new people 😀
Curious to listen to some interesting Talks and meet new people 😀
Reposted
First draft online version of The RLHF Book is DONE. Recently I've been creating the advanced discussion chapters on everything from Constitutional AI to evaluation and character training, but I also sneak in consistent improvements to the RL specific chapter.
rlhfbook.com
rlhfbook.com
April 16, 2025 at 7:01 PM
First draft online version of The RLHF Book is DONE. Recently I've been creating the advanced discussion chapters on everything from Constitutional AI to evaluation and character training, but I also sneak in consistent improvements to the RL specific chapter.
rlhfbook.com
rlhfbook.com
Reposted
„Bei der #SPD haben noch die Mitglieder die Chance, die Notbremse zu ziehen und den Abbau wichtiger Grundrechte zu verhindern. Wir appellieren daher an die Sozialdemokraten: Stimmt dieser Überwachungsliste nicht zu!“ 🤝👏 @ccc.de
CCC | CCC fordert Notbremse für den Überwachungskatalog im Koalitionsvertrag
Der Chaos Computer Club ist eine galaktische Gemeinschaft von Lebewesen für Informationsfreiheit und Technikfolgenabschätzung.
www.ccc.de
April 10, 2025 at 9:55 AM
Reposted
I think this is an important point that we missed the last 10y. We need to call out fiction as fiction and fighting for reality (can be boring, but it’s real).
Reality is, there is no AGI, just language models. If you think they are giving you some truth, the stupidity is on you, not the LLM.
Reality is, there is no AGI, just language models. If you think they are giving you some truth, the stupidity is on you, not the LLM.
This isn’t a report, this is a fiction writing exercise www.nytimes.com/2025/04/03/t...
April 5, 2025 at 7:16 AM
I think this is an important point that we missed the last 10y. We need to call out fiction as fiction and fighting for reality (can be boring, but it’s real).
Reality is, there is no AGI, just language models. If you think they are giving you some truth, the stupidity is on you, not the LLM.
Reality is, there is no AGI, just language models. If you think they are giving you some truth, the stupidity is on you, not the LLM.
Reposted
👨🏫 Can you help spread the word? Our community supports future #AI professionals like engineers and data analysts. With backing from PySV and Pioneers Hub, we provide students free virtual passes to Germany's top AI conference. Learn more at 2025.pycon.de/blog/free-re....
Enabling the Next Generation in AI & Open Source: Free Remote Tickets for Students at PyCon DE & PyData
Join PyCon DE & PyData 2025 in Darmstadt (Frankfurt), April 23-25! Germany’s largest Python and Data Science conference with talks, workshops, and community events like DjangoGirls and sprints. Be…
2025.pycon.de
April 5, 2025 at 7:57 AM
👨🏫 Can you help spread the word? Our community supports future #AI professionals like engineers and data analysts. With backing from PySV and Pioneers Hub, we provide students free virtual passes to Germany's top AI conference. Learn more at 2025.pycon.de/blog/free-re....
April 3, 2025 at 3:20 PM
Reposted
April 2, 2025 at 3:40 PM
Reposted
Stanford scholars introduced an open-source AI agent that learns how to navigate websites by mimicking childhood learning – an approach that could lead to more efficient, transparent, and privacy-conscious AI: hai.stanford.edu/news/an-open...
@chrmanning.bsky.social @shikharmurty.bsky.social
@chrmanning.bsky.social @shikharmurty.bsky.social
An Open-Source AI Agent for Doing Tasks on the Web | Stanford HAI
NNetNav learns how to navigate websites by mimicking childhood learning through exploration.
hai.stanford.edu
March 28, 2025 at 7:00 PM
Stanford scholars introduced an open-source AI agent that learns how to navigate websites by mimicking childhood learning – an approach that could lead to more efficient, transparent, and privacy-conscious AI: hai.stanford.edu/news/an-open...
@chrmanning.bsky.social @shikharmurty.bsky.social
@chrmanning.bsky.social @shikharmurty.bsky.social
Reposted
#IchBinHanna & der 🐘 senden 1 Botschaft in die Koalitionsverhandlungen!
#WissZeitVG
@drkeichhorn.bsky.social @kubon.bsky.social
#WissZeitVG
@drkeichhorn.bsky.social @kubon.bsky.social
March 28, 2025 at 11:23 AM
#IchBinHanna & der 🐘 senden 1 Botschaft in die Koalitionsverhandlungen!
#WissZeitVG
@drkeichhorn.bsky.social @kubon.bsky.social
#WissZeitVG
@drkeichhorn.bsky.social @kubon.bsky.social
Interesting paper on Human Cognition in the Age of #AI or how not to brainrot arxiv.org/html/2502.12...
Protecting Human Cognition in the Age of AI
arxiv.org
March 24, 2025 at 9:11 PM
Interesting paper on Human Cognition in the Age of #AI or how not to brainrot arxiv.org/html/2502.12...
Reposted
Die amerikanische Forschung ist auch für Deutschland wichtig. Das ist ja kein Wettbewerb, den wir gewinnen wollen. Wenn es anderen schlechter geht, wirkt sich das auch auf uns und unseren gesellschaftlichen Fortschritt aus.
March 24, 2025 at 1:10 PM
Die amerikanische Forschung ist auch für Deutschland wichtig. Das ist ja kein Wettbewerb, den wir gewinnen wollen. Wenn es anderen schlechter geht, wirkt sich das auch auf uns und unseren gesellschaftlichen Fortschritt aus.