Happy 3rd birthday to the Index of Complex Networks! 🥳
ICON was launched @netsci2016 and now lists 655 data sets and 5319 networks, spanning all domains of science and all types of networks #netsci2019icon.colorado.edu/#!/ 1/3
ICON is an open index 😎: Anyone can suggest a publicly available network data set to be included. Suggestions are lightly curated to ensure index consistency. Suggest one here: icon.colorado.edu/#!/suggestions (Check out the Quick Start guide for writing good suggestions) 2/3
ICON has also spun off several large corpora of real-world networks 🤩, which you can reuse for your own studies. So far: the ScaleFreeTest corpus (nearly 1000 networks) and the CommunityFitNet corpus (nearly 600 networks): icon.colorado.edu/#!/about 3/3
• • •
Missing some Tweet in this thread? You can try to
force a refresh
Most networks are incomplete (for reasons). Link prediction algorithms can help fill in the gaps or say which network model fits better. Many link prediction methods exist! But most claim to work well... Is one method the best? How close to optimal predictors are they? 🤔 2/8
We studied a whopping 203 link predictors (in 3 families: topological, model-based, and embedding) on 550 networks from 6 domains (social, biological, info., etc). Across domains/networks, no predictor is best, or worst! Not even close. Like a No Free Lunch Theorem result 3/8
Happy 4th birthday to the Index of Complex Networks! 🥳 ICON was launched @netsci2016 and now lists 689 data sets and 5403 networks, spanning all domains of science and all types of networks, in a fully searchable index. icon.colorado.edu/#!/ 1/5
ICON was created rebelliously ✊, to (1) help researchers use a wider variety of empirical networks, (2) quantify the structural breadth of real-world networks, and (3) move us beyond toy models (and orthodoxy), toward a better understanding of network structure and evolution 2/5
ICON has produced several large corpora containing hundreds of real-world networks 🤩. These can be used to test a new algorithm or to investigate structural patterns, using data that approximates the empirical distribution of complex networks: icon.colorado.edu/#!/about 3/5
Life continues: excited to share a new preprint 🎉 "Denoising large-scale biological data using network filters", with Andrew Kavran, which describes how to leverage networks to filter out independent noise in systems biology data sets [1/4] biorxiv.org/content/10.110…
Core idea: in sys bio data, networks tell us which values should (anti-)correlate, and pooling neighbor values can filter out independent noise. This is like image analysis, where networks are the system's "geometry," and community detection is analogous to image segmenting [2/4]
On synthetic data, where we can systematically vary the (anti-)correlation of neighboring values, network filters can dramatically reduce noise, especially when the noise and filter are well aligned, either en toto or locally in a "patchwork" (via community detection). [3/4]
Good models can help us estimate which and how much different factors may influence the covid-19 pandemic's future. This modeling preprint by @StephenKissler@ctedijanto E Goldstein @yhgrad@mlipsitch is 💯. Some key takeaways for me: [1/4]
1. Data on the two common human betacoronaviruses show evidence of seasonal forcing and some cross-immunity. 2. Given their molecular similarity (they're all betacoronaviruses), we might expect the same for SARS-CoV-2; some statistical evidence from China seems to agree [2/4]
3. This could produce seasonal waves of SARS-CoV-2 transmission (with a round 2 in late 2020; animation by @StephenKissler). 4. But how big the waves, when they occur, and how often, all depend on SARS-CoV-2 immunity strength and length, and on HKU1 and OC43 cross-immunity. [3/4]
Lewis Fry Richardson (1881-1953) is one of my all-time favorite scholars, and arguably one of the fore-fathers of complex systems. I'm happy to have contributed a chapter revisiting his seminal work on war sizes and timing to a new edited volume about him blogs.prio.org/2019/12/lewis-…
Here are some awesome things Richardson did: 1. He discovered fractals, through his work on the Coastline Paradox, which paved the way for (and was cited by) Benoit Mandelbrot’s celebrated work on fractal geometry: wikiwand.com/en/Coastline_p…
2a. Richardson was a pioneer in meteorology, and in the 1920s developed the first numerical model for weather forecasting (which wasn't practical because digital computers didn't exist yet): en.wikipedia.org/wiki/Lewis_Fry…