Friedemann Zenke

This study develops a spiking retinal model trained to predict future visual stimuli, showing it captures retinal responses and processes across species.

This paper introduces a neural circuit theory for representation learning that focuses on predicting in representation space rather than input space, using a recurrent joint-embedding predictive architecture to learn object identities and dynamics from sensory inputs.

The paper systematically mutagenizes human bZIP domains and uses deep mutational scanning to quantify interactions, revealing complex genetic architecture and enabling synthetic bZIP design.

The paper proposes that deviations from excitation-inhibition balance encode local error signals for synaptic plasticity, enabling efficient learning in neural networks.

This paper uses evolutionary strategies to meta-learn plasticity rules in spiking networks, finding stable rules for synapse types but facing challenges with complex co-active rules.

The paper investigates the computational role of balanced excitatory-inhibitory neuronal assemblies in zebrafish olfactory memory networks, showing they prevent runaway activity and correlations.

The paper builds a spiking network model of zebrafish olfactory cortex to study how E/I assemblies affect memory representation, finding continuous representations that enable fast classification and continual learning.

The paper develops a spiking neural network to predict natural movies, revealing key differences between excitatory and inhibitory V1 neurons that aid in sensory prediction.

This paper introduces a plasticity rule that enables deep neural networks to learn invariant object representations through a combination of Hebbian and predictive learning, showing alignment with biological observations.

This paper introduces nonlinear transient amplification (NTA) as a mechanism for transient amplification in recurrent neural networks with short-term plasticity, showing it enables stimulus-selective transients without dynamical slowing or pathological activity.

This study shows that burst-dependent synaptic plasticity enables coordination of learning in hierarchical circuits by solving complex tasks through simulations and mathematical analysis.

This paper investigates how surrogate gradient learning performs in spiking neural networks, showing its robustness to different surrogate derivatives and the importance of regularization for effective learning.

This paper uses meta-learning to discover synaptic plasticity rules that can produce desired neural functions, re-discovering known rules and finding new ones for complex tasks.

The paper examines the interaction between Hebbian and homeostatic plasticity, finding that slow homeostatic processes are insufficient to stabilize Hebbian-driven instabilities and proposing the need for rapid compensatory mechanisms.