Toshitake Asabuki

The paper proposes a mechanism where spike history gates plateau potentials in CA1 neurons, influencing when and where neural representations are rewritten based on prior activity.

This study shows that representational drift in neural networks arises from predictive excitatory-inhibitory plasticity, with changes in neuron preferences and stable ensemble coding.

This paper shows that feedback geometry in recurrent networks determines whether learned computations can be recombined through zero-shot parallel composition.

The paper shows that temporal predictability determines which evoked activity dimensions are included in spontaneous neural manifolds, affecting their dimensionality and geometry.

This paper shows that recurrent neural networks can learn new tasks without forgetting old ones by preserving task-specific latent dynamics through orthogonal task manifolds.

The paper compares synaptic kinetics across mammals and finds human neurons have faster excitatory transmission, leading to better pattern detection in noise during unsupervised learning.

This paper introduces a novel learning rule called Predictive Alignment for recurrent neural networks that suppresses chaos to generate patterned activities using biologically plausible plasticity.

This study tests if local neural circuits can generate predictive signals using a recurrent spiking network trained with local plasticity rules, replicating experimental prediction error patterns.

The paper shows that a recurrent spiking neuron network using predictive learning rules generates spontaneous activity resembling probabilistic sensory experiences, replicating behavioral biases in perceptual decision making.

This paper proposes biologically plausible synaptic plasticity rules to embed stochastic environmental dynamics into spontaneous neural activity in a spiking neural network model.

This study shows that intrinsic bursts, not isolated spikes, enhance learning of Levy flight trajectories in recurrent neural networks by synchronizing burst onsets across populations.

This paper introduces a recurrent gated neuronal network model with dendrites to solve context-dependent segmentation tasks using synthetic and calcium imaging data.

The paper introduces artificial dendritic neurons that use conditional entropy minimization to self-supervise temporal feature extraction from continuous information streams.

This paper introduces a class of unsupervised neural networks that identify repeated patterns in sequence data through mutual prediction between reservoir computing modules.