shakedown.social is one of the many independent Mastodon servers you can use to participate in the fediverse.
A community for live music fans with roots in the jam scene. Shakedown Social is run by a team of volunteers (led by @clifff and @sethadam1) and funded by donations.

Administered by:

Server stats:

242
active users

#dynamicalsystems

0 posts0 participants0 posts today

Can time series (TS) #FoundationModels (FM) like Chronos zero-shot generalize to unseen #DynamicalSystems (DS)? #AI

No, they cannot!

But *DynaMix* can, the first TS/DS foundation model based on principles of DS reconstruction, capturing the long-term evolution of out-of-domain DS: arxiv.org/pdf/2505.13192v1

Unlike TS foundation models, DynaMix exhibits #ZeroShotLearning of long-term stats of unseen DS, incl. attractor geometry & power spectrum, w/o *any* re-training, just from a context signal.
It does so with only 0.1% of the parameters of Chronos & 10x faster inference times than the closest competitor.

It often even outperforms TS FMs on forecasting diverse empirical time series, like weather, traffic, or medical data, typically used to train TS FMs.
This is surprising, cos DynaMix’ training corpus consists *solely* of simulated limit cycles & chaotic systems, no empirical data at all!

And no, it’s neither based on Transformers nor Mamba – it’s a new type of mixture-of-experts architecture based on the recently introduced AL-RNN (proceedings.neurips.cc/paper_f), specifically trained for DS reconstruction.

Remarkably, DynaMix not only generalizes zero-shot to novel DS, but it can even generalize to new initial conditions and regions of state space not covered by the in-context information.

We dive a bit into the reasons why current time series FMs not trained for DS reconstruction fail, and conclude that a DS perspective on time series forecasting & models may help to advance the #TimeSeriesAnalysis field.

What is the connection between fractal geometry and systems at a critical point undergoing phase transition? This is one of the more useful ideas that has emerged from the study of dynamical systems, but often it's buried too deep into the study of modeling for most people to encounter it-- then it gets explained badly in pop-science books.

At last here is a video that will set you right:

youtube.com/watch?v=vwLb3XlPCB

Just moved to neuromatch.social, so here it goes (again), #introduction :

Hi everyone, I'm a last year undergrad in #Neuroscience & #ComputerScience at McGill. I'm doing #ComputationalNeuroscience research in the Baillet Lab at The Neuro (MNI), focusing on whole-brain dynamical models of coupled neural masses calibrated to #MEG #Neuroimaging data (more details @ neurolife77.github.io/ if anyone is curious).

I am also the VP of the #MachineLearning committee at PharmaHacks, a hackathon that blends #Biology & #DataScience with a focus on #Pharma.
@neuroscience #neurodon

------------------------ Bonus ------------------------

Since I have the space to put it in the same post now, thanks to the freedom in post length from this new server, here's a bonus:

I regularly share links to preprints that catch my attention and tag them with: #arxivfeed

I started doing this because I thought that the arxiv bots on mastodon were not super efficient, but after doing it for about a month I'd say it's also a good way to keep some form of history of my nightly exploration of the literature in my fields of interest. I usually share stuff about #ComputationalNeuroscience, #Neuroimaging, #DynamicalSystems, #MachineLeaning, #ArtificialIntelligence, etc.

Disclaimer: I usually only read the abstract or skim through them at the time of posting.
Disclaimer 2: I am definitely not consistent.

neurolife77.github.ioDominic Boutet - WebsiteDominic Boutet personal website. I am a Neuroscience and Computer Science student at McGill University...

Hello everybody, here is my #introduction.

I am Jitse (he/him), originally from the Netherlands but working in the School of Mathematics at the University of #Leeds, UK for 10+ years. Research interests: #NumericalAnalysis and #DynamicalSystems. Lately working on Fourier extension, geometric numerical integration and applications in particle methods in plasma physics and compartments models in chemistry. I contribute to #Spyder, an open=source IDE for Python geared towards scientists.