Project Details
Projekt
Temporal dynamics of unsupervised contextual learning in cortico-collicular loops
Applicant
Livia de Hoz, Ph.D.
Subject Area
Cognitive, Systems and Behavioural Neurobiology
Term
since 2023
Project identifier
Deutsche Forschungsgemeinschaft (DFG) - Project number 520284049
Familiarity with the acoustic context (defined as sounds with stationary statistics in space) shapes our listening process. Statistical learning of contextual information is thus essential for adaptive sensing, yet we know little about the circuits involved in this implicit (unsupervised, non-reinforced) learning of the acoustic context. Our own work has shown changes in sound representation in the auditory midbrain (the inferior colliculus, IC) associated with contextual learning in an enriched environment. The IC is a processing hub in the auditory pathway, receiving inputs from other auditory structures, notably feedback from primary auditory cortex (A1) as well as from non-auditory structures such as motor and visual cortex or amygdala, placing it in optimal position to integrate contextual information. While there is ample evidence for a role of A1 and, specifically the cortico-collicular feedback projection, in reinforced associative learning, learning studies have typically used short stimuli more descriptive of foreground than contextual sounds. On the other hand, protocols designed to study aspects of contextual processing (streaming, scene analysis protocols) have focused on the ongoing neuronal adaptation that accompanies the build-up of an expectation in these regular sound streams, rather than on learning across repetitions. To better understand the circuits involved in contextual learning, we propose to combine the learning perspective and stream segregation protocols, with an aim to characterize sound representation in stations of the cortico-collicular loop-the IC and A1- (Objective 1), as well as the role of the cortico-collicular pathway (Objective 2) in contextual learning across different time windows. We will study activity in awake mice as contextual learning takes place, trial to trial and day to day. Our previous work found that changes in IC representation depended on the predictability of the contextual sound, a predictability that we believe was determined by the activation of the sound by the mouse movement. To assess the role of self-activation in predictability, we will incorporate comparisons between passive versus active (self-generated) exposure in the experimental design. In addition to measures of frequency representation we will use the time-course of adaptation as a measure of contextual learning over short- (minutes) and long-term (days) time windows using well-established streaming protocols. Finally, to better understand perceptual variables of contextual sounds, we will use behavioural assays to study the effect that contextual learning has on foreground detection (Objective 3). These experiments will contribute to our understanding of unsupervised learning in general and to the role of early auditory loops in particular.
DFG Programme
Priority Programmes
