←
CoMuNe Lab883 posts
CoMuNe Lab avatar

CoMuNe Lab

@ComuneLab

Research Group for Multilayer Modeling and Analysis of Complex Systems: from cells to societies. Led by @manlius84 at @DFAunipd @unipadova

◷ Joined December 2017▧ 144 Photos and videos
CoMuNe Lab avatar
CoMuNe Lab@ComuneLab·14 Feb 2025

Kudos to the team, who worked tirelessly to build a robust study, as well as to the reviewers and the editor for the constructive peer review process. For more info about the density matrix literature, check the thread curated by @manlius84 end/

This tweet is unavailable

CoMuNe Lab avatar
CoMuNe Lab@ComuneLab·14 Feb 2025

Our paper adds to a long list of papers showing how StatPhys and ML can be used together to tackle very hard problems in #NetworkScience It's cool that behind such methods there is our density matrix framework: I can't foresee how many applications are still to be found 13/

CoMuNe Lab avatar
CoMuNe Lab@ComuneLab·14 Feb 2025

These features make the NFC a new technique to investigate how network function works across scales, providing a complementary multiscale perspective to structural techniques such as LRG. NFC is not a RG method: it doesn’t iteratively rescale or focus on fixed points. 12/

CoMuNe Lab avatar
CoMuNe Lab@ComuneLab·14 Feb 2025

By dropping fast modes above a given scale, LRG effectively renormalizes networks while capturing the backbone of slow information dynamics NFC prioritizes functional similarity over topological constraints: nodes w/ similar functional roles are grouped even if disconnected 11/

Post attachment
CoMuNe Lab avatar
CoMuNe Lab@ComuneLab·14 Feb 2025

Recently, a RG technique has been developed to exploit our density matrix framework: it is called the Laplacian RG. In LRG, nodes are grouped based on slow diffusion modes (low eigenvalues of the Laplacian), ensuring that information flow is preserved at a specific scale 10/

CoMuNe Lab avatar
CoMuNe Lab@ComuneLab·14 Feb 2025

Super-nodes are effective coarse-grained units that play similar roles in network function, not necessarily in direct connectivity, at variance w/ standard RG methods Our purpose is different and NFC provides the most reliable results when compared w/ structural methods 9/

CoMuNe Lab avatar
CoMuNe Lab@ComuneLab·14 Feb 2025

Super-nodes preserve the full information dynamics, retaining the diversity of diffusion pathways & signal speed across the network. NFC identifies redundant structures in information flow and merges nodes even if they are not directly connected but perform similar roles. 8/

Post attachment
CoMuNe Lab avatar
CoMuNe Lab@ComuneLab·14 Feb 2025

This is the largest difference with renormalization group techniques for networks, which instead are focused on structural scale invariance by means of specific ad-hoc rules. We sacrifice a bit interpretability to achieve an unexpectedly good results. 7/

Post attachmentPost attachment
CoMuNe Lab avatar
CoMuNe Lab@ComuneLab·14 Feb 2025

It's not possible to achieve this in analytical form, so we asked a machine to learn what should be done. Our Network Flow Compression (NFC) works like a charm and does its job by preserving information flow properties across all scales, rather than focusing on structure only. 6/

CoMuNe Lab avatar
CoMuNe Lab@ComuneLab·14 Feb 2025

Back to the original problem. By ensuring the partition function remains nearly unchanged across scales, we guarantee that key macroscopic flow properties (entropy/free energy) are preserved even after compression (coarse-graining). That's where StatPhys ends and ML starts 5/

Post attachment
CoMuNe Lab avatar
CoMuNe Lab@ComuneLab·14 Feb 2025

Providing a grounded definition for von Neumann entropy, we can use it in StatPhys and information-theoretic problems. Recently, we have used to define a variational approach for open systems, able to predict the emergence of sparsity in networks. 4/ nature.com/articles/s41567-0…

Post attachment
CoMuNe Lab avatar
CoMuNe Lab@ComuneLab·14 Feb 2025

The density matrix describes an ensemble of events unfolding on a quenched network, according to a given dynamics. For this reason it is suitable to describe complex information dynamics for multiscale analysis: journals.aps.org/pre/abstrac… journals.aps.org/pre/abstrac… 3/

CoMuNe Lab avatar
CoMuNe Lab@ComuneLab·14 Feb 2025

Problem: assume a process on a network with N nodes. How should we group nodes into n<N super-nodes while preserving the overall diversity of pathways & flow speed? NFC solves this by learning node groupings by means of our density matrix framework 2/ journals.aps.org/prx/abstrac…

CoMuNe Lab avatar
CoMuNe Lab@ComuneLab·14 Feb 2025

🚨 Excited for the latest paper from the lab, led by Zhang & Arsham Can we coarse-grain a #ComplexNetwork while preserving macroscopic network flows? Surprisingly, the answer is positive, and we have shown it by means of StatPhys + Machine Learning 🧵1/ nature.com/articles/s41467-0…