←
Francesco Martinuzzi173 posts
Francesco Martinuzzi avatar

Francesco Martinuzzi

@MartinuzziFra

Postdoctoral researcher | Chaos and ML | @mpi_pks

◷ Joined February 2015▧ 12 Photos and videos
Francesco Martinuzzi avatar
Francesco Martinuzzi@MartinuzziFra·9 Sep 2025

What is the minimally effective way to construct a reservoir computer that can learn chaotic dynamics? In my new paper, I demonstrate that deterministically constructed reservoirs with minimal topologies (MESNs) consistently outperform conventional random-initialized ESNs. 1/2

Francesco Martinuzzi avatar
Francesco Martinuzzi@MartinuzziFra·Jun 24

Links: Paper: arxiv.org/abs/2510.21252 torchrecurrent: github.com/MartinuzziFrances… RecurrentLayers.jl: github.com/MartinuzziFrances… LuxRecurrentLayers.jl: github.com/MartinuzziFrances…

Francesco Martinuzzi avatar
Francesco Martinuzzi@MartinuzziFra·Jun 24

These libraries aim to make that workflow easier by providing unified implementations of many recurrent cells and higher-level layers across Python and Julia, with a consistent interface and room for customization.

Francesco Martinuzzi avatar
Francesco Martinuzzi@MartinuzziFra·Jun 24

There are many recurrent cells in the literature beyond the standard RNN / GRU / LSTM family, but trying them in practice often means digging through papers, finding old reference implementations, adapting code across frameworks, and hoping the details match.

Francesco Martinuzzi avatar
Francesco Martinuzzi@MartinuzziFra·Jun 24

Great news for recurrent neural networks (RNNs) enthusiasts! Over the past few months, I’ve been working on a small ecosystem of open-source libraries for RNNs: • torchrecurrent • RecurrentLayers.jl • LuxRecurrentLayers.jl

Francesco Martinuzzi avatar
Francesco Martinuzzi@MartinuzziFra·9 Sep 2025

Why it matters: - More reliable than random-initialized models - Lower tuning overhead, less trial and error - Stronger stability across runs 📄 Read more: pubs.aip.org/aip/cha/article… 2/2

Dr. Chris Rackauckas avatar
⟳ Francesco Martinuzzi retweeted
Dr. Chris Rackauckas@ChrisRackauckas·26 May 2025

New blog post: How chaotic is chaos? How some AI for Science / SciML papers are overstating accuracy claims. stochasticlifestyle.com/how-… #julialang #sciml #chaos #ergodic #ai4science

ScaDS.AI Dresden/Leipzig avatar
⟳ Francesco Martinuzzi retweeted
ScaDS.AI Dresden/Leipzig@Sca_DS·11 Apr 2025

⭕️ This week @Sca_DS. Our scientists (and their green colleagues🍃) are moving up. In recent years, @Sca_DS has grown rapidly. With the additional space on the 5th floor, our center in Leipzig has now more than doubled in size. This week the move was mostly completed. #RoomToGrow

Post attachment
Claudio Gallicchio avatar
⟳ Francesco Martinuzzi retweeted
Claudio Gallicchio@claudiogallicc1·13 Jan 2025

🚀 Call for Papers 🚀 Special Session: Reservoir Computing in the Deep Learning Era: Theory, Models, Applications, and Hardware 📍 IJCNN 2025 | 30 June - 5 July 2025 | Rome, Italy 🔗 lnkd.in/da9pMn-S 📅 Submission Deadline: January 30, 2025 (extended)

Post attachment
Ruben Ohana avatar
⟳ Francesco Martinuzzi retweeted
Ruben Ohana@oharub·2 Dec 2024

Generating cat videos is nice, but what if you could tackle real scientific problems with the same methods? 🧪🌌 Introducing The Well: 16 datasets (15TB) for Machine Learning, from astrophysics to fluid dynamics and biology. 🐙: github.com/PolymathicAI/the_… 📜: openreview.net/pdf?id=00Sx57…

ScaDS.AI Dresden/Leipzig avatar
⟳ Francesco Martinuzzi retweeted
ScaDS.AI Dresden/Leipzig@Sca_DS·4 Nov 2024

🌳On Wednesday, the 3rd annual Topic Area Meeting of the Earth and Environmental Sciences @ScaDS_AI took place. With great success! 🌍Throughout the day, more than 30 participants discussed their research across environmental sensor types, scales and ML architectures. #AppliedAI

Post attachmentPost attachmentPost attachment