Sadamori Kojaku
@skojaku.bsky.social
89 followers 58 following 76 posts
Assist. Prof. in 🇺🇸 . Interested in Network Science and Computer Science. Love Sailing, Kendo, Tennis, and Cooking Chinese and Italian. My favorite book: 11ぴきのネコ.
Posts Media Videos Starter Packs
skojaku.bsky.social
Deep dive into AI podcast
Reposted by Sadamori Kojaku
cna25.bsky.social
Was your work accepted to #ComplexNetworks2025? It's time to register! Don't wait any longer and take advantage of our early bird discount.🐣

🚨This offer is only available until October 17th.🚨

#ComplexSystems #NetworkScience

Register here:
👉 www.billetweb.fr/complex-netw...
Reposted by Sadamori Kojaku
adrian-currie.bsky.social
If you’re interested in practicing the philosophy of science in practice why not come along to an online workshop “the philosophy of science in practice - in practice”? #philsci
skojaku.bsky.social
Looking forward to watching it! The limitation is known in graph ML 1, w/ some remedies (e.g., BUDDY). I wonder if the SOTA is still falling behind the traditional ones. (we found it was behind [2] but curious why)

1. proceedings.neurips.cc/paper/2018/h...

2. openreview.net/forum?id=K9z...
Link Prediction Based on Graph Neural Networks
proceedings.neurips.cc
Reposted by Sadamori Kojaku
aaronclauset.bsky.social
This was a fun talk to give. Our new results show that "Meta-learning optimizes predictions of missing links in real-world networks" (arxiv.org/abs/2508.09069). tl;dr: graph neural networks are a limited tool for predicting missing links in most networks
www.youtube.com/watch?v=dRYz...
GraphGeeks Talk: Meta-Learner That Finds the Best Link Prediction Model
YouTube video by GraphGeeks
www.youtube.com
Reposted by Sadamori Kojaku
netscience.bsky.social
Network Science, and three of its avengers, featured in the latest @veritasium.bsky.social video (one of the largest youtube channels for Science education)!! youtu.be/CYlon2tvywA?... @stevenstrogatz.com Duncan Watts @barabasi.bsky.social
Something Strange Happens When You Trace How Connected We Are
YouTube video by Veritasium
youtu.be
skojaku.bsky.social
One of the time-consuming tasks in paper writing is to curate bibtex for references. So I automated bibtex curation with an Alfred app that fetches entries from dx.doi.org using DOI or title. It saves time, reduces errors, and maintains consistent LaTeX keys. github.com/skojaku/tobi...
GitHub - skojaku/tobibtex
Contribute to skojaku/tobibtex development by creating an account on GitHub.
github.com
skojaku.bsky.social
My notes in obsidian are so disorganized (and yes this reflects what’s in my brain 😅). There are many pieces I thought important but misplaced/not synthesized. Wonder a better more systematic way to organize notes.
Reposted by Sadamori Kojaku
skojaku.bsky.social
I really like the idea that we learn something by contrasting expectations and observations. When the observations do not fit, we seek for alternative explanations, advancing science.
So I'm encouraged to see the apparently "failed" results in front of me now. Could be a seed of great discovery!
1 | Ellen Rothenberg: inhabiting the data
open.spotify.com
skojaku.bsky.social
Honored to be invited as a keynote speaker at Complenet'26! Looking forward to sharing insights on network science and representation learning!
complenet.bsky.social
📢 Keynotes unlocked: 3/12. Thrilled to have Sadamori Kojaku @skojaku.bsky.social as a keynote speaker in Complenet’26! Join us to learn more about #networks and #neuralembeddings!

🌐 Info & registration: complenet.weeblysite.com
🚨 Call for contributions open – submit by Nov 15, 2025!
Reposted by Sadamori Kojaku
lhd.bsky.social
Our Physical Review Letter looks at how to get power-law distributions of cascade size without tuning or self-organization to criticality by allowing cascades to improve in quality and jump over gaps or dead-ends

TL;DR complex-stories.uvm.edu/friends-funn...

Paper: journals.aps.org/prl/abstract...
complex-stories.uvm.edu
skojaku.bsky.social
Yes. The diff of idea is more like a mental model of the KL divergence at the idea level. In the language of KL divergence, I often don’t have accurate reference probability q (prior works) to gauge the difference (a part of novelty) from the focal probability (the subject of a paper).
skojaku.bsky.social
A phrase that occurred to me today, the "diff" of an idea. A diff of an idea is a statement about the difference from the prevailing ideas. I feel some non-negligible number of papers well cover many papers but miss the "diff", and this obscures the novelty.
Reposted by Sadamori Kojaku
skojaku.bsky.social
good point! Another LLM validates the questions on prompt injection and content match, but not the clarity and specificity. I'll add that check. Thanks for the feedback!

My initial test on a vague question finds that LLM wins, likely because many possible answers. github.com/skojaku/llm-...
github.com
skojaku.bsky.social
Creating questions at a right level of challenge helps me learn nuances and go beyond surface-level understanding. But why not let students to do that?

I created a pedagogical tool that lets students to create challenging questions that stump LLMs.

Repo 👉: github.com/skojaku/llm-...
GitHub - skojaku/llm-quiz
Contribute to skojaku/llm-quiz development by creating an account on GitHub.
github.com
Reposted by Sadamori Kojaku
cna25.bsky.social
Yian Yin develops novel computational tools to understand how individual, social, and environmental processes independently and jointly promote (or inhibit) scientific progress and innovation achievements, from artistic productions to public policy. #ComplexNetworks2025
Reposted by Sadamori Kojaku
cna25.bsky.social
Brennan Klein is interested in developing tools and theory for characterizing dynamics, structure, and scale in networks, but also in creating and analyzing large-scale datasets that reveal inequalities in the U.S., from epidemics to mass incarceration. #ComplexNetworks2025
skojaku.bsky.social
gpt-oss is very very good. Too good to believe that this is an open model only having 20B parameters.... Tested on math, coding, and natural conversations.
skojaku.bsky.social
Separate the effect of geography from mobility choices by comparing observed against random mobility choices to a location in a space irrespective of distance, revealing the effect of the shapes of landmass, location distributions, and geometry of space.
Decoupling geographical constraints from human mobility
Driven by access to large volumes of movement data, the study of human mobility has grown rapidly over the past decades. The field has shown that human mobility is scale-free, proposed models to gener...
arxiv.org