Si Wu

The paper trains RNNs with short-term plasticity to perform decision-making tasks, showing that evidence accumulates during activity-silent periods through synaptic efficacy changes.

This paper proposes that theta sweeps in grid and place cells accelerate cognitive map learning by broadcasting sensory information to unvisited regions, and presents a computational model to test this hypothesis.

This paper introduces a unified Laplacian framework that links spatial representations to navigation policies, showing that spectral decompositions of the Laplacian can enable efficient goal-reaching in complex environments.

The paper introduces BrainScale, a system for scalable online learning in spiking neural networks, demonstrating its ability to train large-scale brain models with low memory and high efficiency.

The paper investigates how early visual experience reshapes microcircuit connectivity in the mouse visual cortex during the critical period, identifying specific synaptic changes in inhibitory neurons that underlie plasticity.

This study shows that place and grid cells complement each other through reciprocal connections to achieve robust and efficient spatial representation in the brain.

This paper proposes a neural circuit model for working memory sequence manipulation, demonstrating its ability to perform two-item swapping and generalize to complex permutations using algebraic principles.

BrainUnit integrates physical units into AI libraries to improve scientific computing accuracy and reliability without performance loss.

The paper presents a computational model explaining alternating theta sweeps in the hippocampal circuit using firing rate adaptation and continuous attractor dynamics, validated against experimental data.

This paper uses recurrent neural networks to model path integration and reveals a hierarchical processing pathway involving band and grid cells.

The paper identifies a unified central thalamus mechanism that explains diverse recovery outcomes in disorders of consciousness using electrophysiological data and machine learning.

The paper develops a computational model of the thalamic reticular nucleus (TRN) network with electrical synapses, showing that spindle oscillation arises from the critical state of the network.

This study identifies a neural mechanism involving firing rate adaptation that modulates hippocampal replay dynamics, showing correlations with theta sequence length and behavioral states.

This study shows that firing rate adaptation in a continuous attractor neural network explains theta phase precession and procession in hippocampal place cells.

The paper introduces BrainPy, a framework for brain dynamics modeling that supports simulation, training, and analysis across multiple scales with high performance.

The paper shows that recurrent neural circuits with Poisson spiking variability perform optimal Bayesian inference by sampling from posterior stimulus distributions.

This paper proposes a neural circuit model that generates temporally scalable sequences through Lie group principles, showing consistency with experimental data.

The paper models how non-uniform inhibition from PV+ interneurons shapes orientation selectivity in V1 during development, matching environmental features through circuit plasticity.

The paper shows that recurrent neural circuits can encode stimulus features through population codes, with variability aligned to stimulus directions and modulated by internal priors.

This paper develops a model of coupled neural circuits that perform distributed Bayesian inference in high-dimensional stimulus spaces using Langevin sampling.

The paper proposes a neural model for concurrent multisensory integration and segregation using congruent and opposite neurons to rapidly recognize stimuli based on sensory cues.