[new preprint] On the Importance of Multistability for Horizon Generalization in Reinforcement Learning
A formal account of why agents fail to generalize to longer episodes, and why multistability is the property they are missing.
Julien Brandoit
🧠
[new preprint] Hardware-Software Co-Design of Scalable, Energy-Efficient Analog Recurrent Computations
The full version of our analog recurrent network work, including how the power cost actually scales with network size.
Learning to think (again)
A daily code challenge I solve without any LLM, to keep the habit of thinking about code.
[conference] Spotlight at ICML 2026
Our work on improving the performance and training stability of parallelizable RNNs for ultra-low power applications got accepted as a Spotlight at ICML 2026!
[conference] Accepted at ICNCE 2026
We made an analog RNN that does keyword spotting with less than a microwatt of power -- and it just got accepted at ICNCE 2026!
[new release] Conductance-based neuron dataset generator and pre-generated spike-train datasets
Open-source Python toolkit for simulating large populations of conductance-based neuron models, plus six ready-to-use spike-train datasets on Zenodo.
[conference] I will present at the Benelux Meeting 2026
We will present our works at the Benelux Meeting 2026 ! I'm part of two works.
[conference] Accepted at COSYNE 2026
Our work on population inference and neuromodulation has been accepted for a poster presentation at COSYNE 2026!
[conference] Presented at NeurIPS 2025 Data on the Brain & Mind Workshop
I presented our work on fast reconstruction of degenerate neuron populations at the NeurIPS 2025 workshop in San Diego, CA!
[new preprint] A fast and tiny current as common generator of slow regular pacemaking in brain and heart
New collaborative work identifying a conserved pacemaker mechanism across neuronal and cardiac systems.