| dc.description.abstract | The brain transforms an exuberant number of incoming noisy signals at imperfect noisy receptors and converts them into meaningful percepts that allows us to seamlessly interact with a complex world full of sensory information. This process involves a number of intricate neural computations and thus understanding perception requires a deeper dive into the underlying neural computations. Specifically, it requires probing what features are extracted from incoming sensory signals, when the stimuli features are extracted and finally where in the brain different stimulus features are primarily located. To address these questions, I studied brain activity at different scales spanning from action potentials, local field potentials, and current source density within microcircuits in macaque primary visual cortex, to whole brain recordings of humans listening to visual and auditory objects, to direct comparisons between brain activity and computations found in artificial neural networks. To characterize stimulus specific information, I used decoding methods for neural activity, as well as representational similarity analysis to compare patterns contained within neural activity and layer activations of artificial neural networks. Within a localized microcircuit in V1, which can be grossly divided to granular (middle), supragranular (top), and infragranular (bottom) layers, I found that stimulus features embedded in spiking data were distributed with unique spatiotemporal profiles within the V1 microcircuit and importantly differed from the spatiotemporal profiles acquired from LFP and CSD signals, showing the unique stimulus feature information contained within each of these signals. | |
