Computational and Theoretical Neural Information Processing Laboratory
Unbelievable that it is out! Bayesian efficient coding, with Jonathan Pillow, is now published in PLOS Computational Biology. Jonathan and I started it when I was a postdoc in his lab at UT Austin; the oldest figure in our repository is from April 2012. Since then, the paper has gone through six versions on bioRxiv and 516 commits.
Abel’s paper on universal approximation for dynamical systems has been accepted at NeurIPS 2026, and we will present the poster in Sydney on December 9! He proved that Neural ODEs can approximate multistable dynamics, the basis of decision-making, rhythm generation, and analog memory, over the infinite time horizon.
Welcome Germán Abrevaya, who joined CATNIP Lab in July as a postdoctoral researcher at the Champalimaud Centre for the Unknown. He fills the postdoc position we advertised earlier this year.
Our Perspective on the cybernetic future of large-scale model-centric integrative neuroscience is now on arXiv. Predictive accuracy is not understanding: a Transformer-style foundation model that captures neural time series at scale does not, by itself, explain how behavior arises from the closed-loop coupling between brain, body, and environment. We argue for redirecting the model class itself toward integrative neurocybernetic models – closed-loop, embodied, dynamically structured, and organized around a controller pursuing latent objectives, with internal dynamics that correspond to biologically identifiable components.
COSYNE 2026 is in Lisbon this year (March 12–15 main meeting, March 16–17 workshops in Cascais). #cosyne2026
Update (July 2026): this position has been filled. Germán Abrevaya joined the lab as a postdoctoral researcher, and we are no longer accepting applications for this call. The original announcement is kept below for the record.
We have a new exciting line of work showing that the typical guarantees for universal approximation for neural networks do not hold for recurrent systems in general. Abel’s new work shows that a subclass of dynamical systems can have guaranteed approximation by Neural ODEs with connection to temporal generalization.
After the successful hosting of the new 3-day symposium in October 2025, the original website hosted at https://symposium.fchampalimaud.science/ has retired as it was part of the continuing cycle of Champalimaud Research Symposium series.
Information in spiking neurons on latent variables
Perfect mathematical objects and their shadows in the brain
Closing remarks for the NSF CAREER award IIS-1845836
The Neural Dynamics Lab, also known as the CATNIP Lab, headed by Memming Park at the Champalimaud Centre for the Unknown, Lisbon, Portugal, is looking for talented next generation scientists and engineers to advance neurotechnology and neural data analysis tools to better understand neural dynamics. Our broad goal is to obtain an effective systems-level description of relevant neural dynamics in the context of cognitive functions and dysfunctions, such as working memory, decision-making, motor control, and disorders of consciousness.
This year’s COSYNE had great content in all areas of computational neuroscience. In this post we give highlights of some of the posters, talks, and workshop sessions that we found to be interesting or relevant.
After a cold trip, we finally made it to Breckenridge. I felt lucky to get on a shuttle before the storm that blocked the way. The resort was really a nice and warm place to heal my headache.
I only mentor high school students through the Simons Summer Research Program (SSRP). Please refrain from contacting me directly before apply to SSRP, if you are interested in working in the lab.
Workshop Speakers:
Workshop Speakers:
Workshop Speakers:
Workshop Speakers:
Neurons are the fundamental unit of computation of the brain, however, they do not work alone when we perceive a tiger and decide to run away. The fundamental question in systems neuroscience is to understand how neurons interact with each other to generate large-scale dynamics that implements cognitive behavior and support dynamical neurological disease such as Parkinson’s disease at the same time.
CATNIP Lab is looking for awesome people: postdocs, graduate & undergraduate students interested in computational neuroscience and machine learning.