Takuya Isomura

The paper shows that basal ganglia output in zebrafish destabilizes neural activity states to guide decision-making in a Go/No-go task.

This paper develops a neural network model that transfers active inference to new prism-shifted environments, showing faster adaptation and after-effects.

This paper shows synaptic pruning enables efficient Bayesian model selection by reducing neural network structures through connectivity parameters.

This paper uses active inference to model zebrafish neuronal dynamics during a go/no-go task, predicting individual learning trajectories and behavioural changes.

The paper validates the free-energy principle in vitro using rat cortical neurons, showing that free energy minimization predicts network self-organization and plasticity.

This paper shows canonical neural networks perform active inference by minimizing variational free energy through delayed plasticity and threshold adaptation.

This paper shows that biologically plausible cost functions in neural networks correspond to variational free energy under specific prior beliefs, enabling optimization of network parameters.

This paper demonstrates that a neural network using the error-gated Hebbian rule can learn to perform multi-context blind source separation by retaining context memories and extracting common features.

This study shows adult zebrafish use future state prediction errors to guide active avoidance behavior in a virtual reality setup.

The paper tests if in vitro neuronal responses align with Bayesian ideal observer models based on free energy minimization.

The paper proposes a model for social intelligence that uses multiple internal models to infer and learn from conspecific interactions, demonstrating success in a synthetic bird learning zebra finch songs.

The paper compares four recurrent neural network models of prefrontal cortex to identify common features in plastic changes during cognitive tasks, finding that plastic synapses are concentrated in small neuronal populations.