"Transforming visual information between areas V1 and MT" - Dr Nic Price, Monash University

An OIST neuroscience online seminars (ONOS) talk.

Abstract

The visual system is a complex, hierarchically-organised information processing network. Counter-intuitively, successive areas contain less information about a scene, but neural activity is structured to better represent specific information. For example, neurons in primate area MT convey little colour information, but motion direction can be linearly decoded from their activity. The ongoing activity of individual neurons is highly variable, meaning reliable computation depends on collaborative processing across neural populations. However, it remains unclear how visual information is reliably represented across neurons within an area, and how these representations are transformed between areas to extract specific stimulus properties. To address this, we record visually-evoked activity simultaneously from dozens of neurons in V1 and MT in marmosets.

We use decoding techniques to predict stimulus orientation or direction from activity across a neural population. This has allowed us to show that neural representations are affected by stimulus history: recent motion biases predictions in a manner consistent with perceptual illusions; and luminance and contrast changes affect orientation coding, again in a manner consistent with human sensitivity.

We study inter-area communication by comparing the timing of action potentials in V1 with local field potentials (LFP, a population measure of local dendritic activity) in MT. We have shown that action potentials preferentially occur at specific phases of the LFP, and that motion information is best communicated from V1 to MT at specific phases.

Collectively, this suggests that hierarchical information processing depends on action potentials in privileged subsets of neurons occurring in privileged time windows.

Zoom details

Zoom Meeting link:
https://oist.zoom.us/j/95700149323?pwd=VWQxY0FGQmxVQmtUTVZ2SXRnMGx6dz09

Meeting ID: 957 0014 9323
Passcode: 855303