Thrilled to share our new paper in Science @science.org: “Scientific Production in the era of Large Language Models.” - LLMs boost productivity, especially for non-native speakers. - Writing complexity is now "decoupled" from scientific quality. - LLMs help discover more diverse literature.

A long-overdue interactive version finally done (Huge thanks to Google Gemini) Complex Systems Topic Network: Interactive Visualization bingdev.binghamton.edu/sayama/compl...

We are looking for a PhD student to work with us on network science methods for biomedicine. The student can be enrolled in any graduate program. #NetworkMedicine #ComplexSystems.

GPT Inference (greedy, top-k, beam search, temperature control).

word2vec: from the structural linguistic perspective. Written by me. Illustrated by Nano Banana Pro.

🧵1/ A bit late to the party, but I want to introduce a pet project of mine—and my first single-author paper! 🎉 It tackles something that has quietly frustrated me for years: how we measure the similarity of two clusterings/community structures.

Nano Banana Pro unbelievably lowers the entry barrier for people interested in creating short Manga. This blog post describes the recipe (in Japanese). I get this Manga version of my lecture note on the position encoding in transformers, **in the first shot** 😮. note.com/konho/n/na8d...

海南鶏飯を一日かけて作ったのだけど、子供は食べないばかりかぐちゃぐちゃにされて心が折れた。１０分でできるパスタ出してあげるよ。次から🫠。

I'm looking for a PhD student to join my lab at UAlbany for Fall 2026! We are data scientists studying the creation, flow, and impact of information. How do scientists collaborate? How does AI shape information access? What drives local innovation? Want to answer these questions? details below!

🚨 New paper on ArXiv: “Uncertainty quantification and posterior sampling for network reconstruction” TL;DR; We present an efficient method to sample the entire ensemble of possible network reconstructions that are compatible with an indirect observation, e.g. a dynamics. Short thread: 1/N

TLDR: communities are undetectable even if they are because of the variabilities of network topology. We look further into edge weights, showing that a higher variability in edge weights makes communities harder to be detected. Our results inform when to use edge weights and better edge reweighing.

Is #innovation born or made? Why do some places keep innovating while others fall behind? Is innovation the product of cities—or of deeper laws? Our new paper in npj Complexity explores these questions through the lens of a century of US patent data.🧵👇 nature.com/articles/s44...

When are communities detectable in weighted networks? You can find the answer in our latest preprint. Great collaboration with @filipisilva.bsky.social @skojaku.bsky.social arxiv.org/abs/2511.00214

Wanted to create an interactive *public* visualization for networks/embeddings? I created a minimum GitHub Page template for HeliosWeb visualization. @filipisilva.bsky.social 👉 Demo: shorturl.at/m9CnR 👉 Repo: shorturl.at/HJ2gn 👉 HeliosWeb: shorturl.at/H9bJU #dataviz #datascience #python

A semi-regular reminder: My colleagues and I are running an interdisciplinary online seminar series called Complex Systems Research Exchange (CREx🦖). sites.google.com/view/cxrex

Snakemake is essential for reproducible research—it organizes code, and acts as my second brain about code and data relations. A downside is its maintenance cost. It can be easily convoluted as a project progresses. This tool might resolve that. arxiv.org/html/2505.02...

I've been playing with Ollama Cloud that has a free cloud Qwen 3 Vision Language model, which is surprisingly good. I put together a small app that automatically renames screenshot files based on their content, which has been super helpful for organizing them. github.com/skojaku/scre...

I'd love to see someone try to estimate just how much time and money has gone into research that is either fully undermined by reliance on LLMs or fully pointless --- because obvious if you start from an understanding of what LLMs actually are. www.pnas.org/doi/10.1073/...

We will have two CoCo seminars on October 29th and 30th! Dakota Murray (U. Albany) and Anastasia Bizyaeva (Cornell) will come to Binghamton/CoCo :) Note the irregular schedule.

That's a great project to learn how LLMs are trained. $100 is not cheap but the cost to train LLMs comes down to the realm where I can try (just for fun). github.com/karpathy/nan...

2025 isn't over yet but it's unlikely that Twitter will see a resurgence of research discussions.

Deep dive into AI podcast

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...

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

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...

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

Come to SUNY and be part of the growing Northeast US complexity community! Professor and Chair, Department of AI and Society, University at Buffalo www.ubjobs.buffalo.edu/postings/58816

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...

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.

New paper by A. Varga + coauthors @skojaku.bsky.social & @filipisilva.bsky.social introduces a family of paper & author similarity measures based on the idea that papers are more similar if they're more likely to be retrieved during lit search via backward & forward citations doi.org/10.1162/qss....

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!

Honored to be invited as a keynote speaker at Complenet'26! Looking forward to sharing insights on network science and representation learning!

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...

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.

Q: How are resources consumed in transportation networks, and how does this shape the overall functioning of the system? We introduce the minimum-cost percolation framework and apply it to the U.S. air transportation system using publicly available data. 🔗 www.nature.com/articles/s41...

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-...

🚨🚨 Correction posted 🚨🚨 "The simpliciality of higher-order networks" in collaboration with @jgyou.bsky.social and @eikmeier.bsky.social is now posted! This was an intimate look into the strengths and weaknesses of the correction process. A short 🧵 epjdatascience.springeropen.com/articles/10....

You can now donate to support OSoMe!! osome.iu.edu/about/donate

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

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

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.

Transformers are graph neural networks applied to fully-connected token graphs. Interesting perspective though it is natural as the graph attention networks adopt transformer architecture.

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.

"our results indicate that academia pays a price by focusing attention and resources on superstars" Nice to see this supported by data analysis, now. H/T: @cxdig.bsky.social www.nature.com/articles/s41...

🚀Ready to shape the future of #NetworkScience? Submit to #ComplexNetworks2025! 📅Dec 9–11, Binghamton, NY 🇺🇸 📄 Full Papers & Extended Abstracts ⏳ Deadline: Sept 2, 2025 👉 complexnetworks.org/submission/ #ComplexSystems #CallForPapers #ComplexNetworks

Enjoyed to attend a talk at #ICML2025 on loss landscape during training for Gaussian mixture data using random matrix theory. Very interesting in that almost the same framework is used to study very different subject, the detectability limit of communities for degree hetero networks.

🚨🚨 New preprint just dropped! In collaboration with @colltoaction.bsky.social, Cliff Joslyn, @fralotito.bsky.social, Audun Myers, Joshua Pickard, Brenda Praggastis, and Przemysław Szufel, we define a new data sharing standard for higher-order networks. arxiv.org/abs/2507.11520

We are excited to share that this paper has been published in Scientific Data. The whole dataset is now available to download. Please see the paper for details! Link: doi.org/10.1038/s415...

UB's new Department of AI and Society is hiring faculty across ranks (Assistant, Associate, Full Professor). We’re looking for transdisciplinary scholars interested in building AI by society, for society. Start dates begin Fall 2025. More info: www.ubjobs.buffalo.edu/postings/57734

Decoupling geographical constraints from human mobility arxiv.org/abs/2405.08746