Erik Andreas Martens

This is joint work with Rok Cestnik — many thanks for a great collaboration!

The method performs well for attractor reconstruction, bifurcation diagram estimation, and prediction with partial or noisy data. Interestingly, adding small noise during training improves long-term stability — promising for surrogate models and digital twins.

We introduce a scalable next-generation reservoir computing (NGRC) framework based on a pseudorandom nonlinear projection of time-delay embedded inputs. This decouples feature-space dimensionality from observation size, giving much more flexibility.

🚀 New Paper Published!
Proud to share the latest work with: “Next-generation reservoir computing for dynamical inference” published in Chaos!
📄 doi.org/10.1063/5.03...

#ReservoirComputing #DynamicalSystems #AIforScience

It's been a great pleasure to collaborate with @bickmath.bsky.social on Constrained network adaptation - published now in Chaos as a featured article!

Sounds like I missed out on a great session!

We should use it as a talk title. Or give a presentation at some meditation retreat.

Free yourself - use constraints and reduce dimensionality!

Multiple Timescale Dynamics of Network Adaptation with Constraints Adaptive network dynamical systems describe the co-evolution of dynamical quantities on the nodes as well as dynamics of the network connections themselves. For dense networks of many nodes, the resul...

Adaptive network dynamics are too high dimensional? Constrain! Or maybe adaptation may be naturally constrained to low-dimensional dynamics? Joint work with
@math-martens.bsky.social
now on the arXiv at arxiv.org/abs/2507.06359.