Srdjan Ostojic

This paper develops a theory of multi-tasking in neural circuits, distinguishing between sequential and simultaneous task performance and introduces a model of task barcodes to parameterize task selectivity.

This study examines how primary auditory cortex encodes categorical responses during a perceptual task, finding that categorical representations emerge during sound presentation and are maintained during the delay period.

The paper compares low-dimensional dynamics in spiking neural networks with those in rate networks, showing that rate model predictions apply to spiking networks and enable nonlinear computations.

The paper develops a method to map local connectivity statistics onto global low-rank connectivity structures in excitatory-inhibitory networks, showing how local connectivity shapes global dynamics.

The paper shows that low-dimensional neural manifolds enable parametric control of flexible timing through tonic inputs in trained recurrent networks and confirms this in monkey behavior.

This paper investigates how sparsity affects low-rank recurrent neural networks, finding that low-dimensional dynamics persist even at high sparsity levels.

The paper investigates how population gating in the auditory cortex supports stimulus selection through interactions with the medial prefrontal cortex in a task-dependent manner.

This study shows that population structure and dynamical dimensionality complement each other in shaping neural computations, with non-random structures enabling flexible input-output mappings.

The paper shows that pre-trained RNNs in ecological tasks mimic rats' suboptimal decision-making in 2AFC tasks by gating across-trial evidence after errors.

The paper investigates OFF responses in the auditory cortex of awake mice using population analyses of two-photon calcium recordings and compares single-cell and network models to understand their underlying mechanisms.

The paper investigates how task-set learning relies on unsupervised temporal chunking of stimulus-response associations using behavioral and neural data.

The paper shows that plasticity of inhibitory synapses onto layer 5 pyramidal neurons modulates excitation-inhibition ratios, neuronal gain, and information transfer across cortical layers.

The paper develops methods to fit mechanistic spiking circuit models to single-trial spike train data, achieving accurate parameter estimation and validation using synthetic and experimental data.

The paper shows that primary auditory cortex encodes stimulus meaning more during task engagement than passive listening, with population-level changes in encoding.

The paper shows that knowledge acquisition and expression can be dissociated by using probe trials without reinforcement, revealing that performance variability stems from testing context rather than inherent ability.