Taro Toyoizumi

This paper proposes a theoretical model of schema learning through low-dimensional neural representations formed during sleep-driven dimensionality reduction.

This paper proposes a neural model linking NMDA hypofunction and dopaminergic hyperfunction to schizophrenia spectrum disorders through prediction error pathways, explaining clinical diversity as a bias-volatility spectrum.

The paper shows that CA3 network activity is distributed across mossy cell ensembles with pseudo-orthogonal representations, enabling efficient information compression in the hippocampal MC layer.

This paper develops a biologically plausible nonlinear dimensionality reduction algorithm inspired by Drosophila olfactory circuits and demonstrates its effectiveness on datasets like entangled rings and MNIST.

The paper shows chaotic neural dynamics enable probabilistic computations by generating samples of variables and trajectories through synaptic learning.

The paper models hippocampal CA3 as a Hopfield-like network that stores and retrieves correlated and decorrelated encodings, showing how inhibitory tone controls transitions between concept and example representations, and extends this to multitask learning in neural networks.

The paper reviews computational models showing that intrinsic synaptic dynamics can support stable network performance and may enhance information processing.

The paper develops a framework to predict the effects of neural perturbations using functional causal flow inferred from resting state activity in monkeys.

This paper introduces a Bayesian model of sense of agency that explains intentional binding and shows that causation can be inferred even from unintended actions when sensory cues are reliable.

The paper shows that intrinsic spine dynamics stabilize cell assemblies and affect learning performance, particularly in autism spectrum disorder models.

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.

The paper identifies two alternating neural populations in synchronized states using ICA, one associated with slow cortical activity and the other with theta oscillations from subcortical regions.