Flexibility and optimization of neural codes in primate sensory cortex
by Tarana Nigam
Date of Examination:2024-06-19
Date of issue:2025-06-18
Advisor:Prof. Dr. Caspar Schwiedrzik
Referee:Prof. Dr. Caspar Schwiedrzik
Referee:Prof. Dr. Alexander Gail
Persistent Address:
http://dx.doi.org/10.53846/goediss-11343
Files in this item
Name:PhD Thesis_TaranaNigam.pdf
Size:5.02Mb
Format:PDF
This file will be freely accessible after 2026-06-18.
Abstract
English
Flexibility and efficiency are core characteristics of any intelligent system. We have a remarkable capacity to continually learn from the structure in the world, predict upcoming situations and flexibly respond in a context-dependent manner. These mental operations build upon the structured and rich representations of the external world which exist in the sensory cortex of the brain. But flexibility is attributed to higher association cortex – implying that stability and flexibility exists in different parts of the brain. In this thesis, I challenge this notion. I first ask how efficient and flexible neural codes for perception arise in the sensory cortex through learnt priors and predictions. To this end, I conduct functional magnetic resonance imaging in the macaque monkey face-processing system, a dedicated sensory hierarchy in the primate brain. I find that top-down predictions dynamically transform neural representations in sensory cortex enabling separable, robust and abstract neural codes for perception. This provides evidence that predictions by means of feedback pathways allow cortical processing hierarchies to achieve their computational goal in an efficient manner. If indeed the sensory cortex can flexibly and efficiently achieve its computational goals, how does it also achieve flexibility when faced with multiple task-demands? I investigate this using invasive electrophysiology in human epilepsy patients that flexibly perform multiple tasks and improve their task performance over repetitions. I test the idea whether the brain adopts the strategy of re-using multipurpose shared resources for multiple tasks that can enable rapid learning and generalization. Or instead utilizes a specialized neural strategy, where different tasks are optimized in different areas, providing lesser interference. I find that the sensory neural codes in temporal cortex exhibit flexible optimization for all the tasks – something that is traditionally ascribed only to the shared resources in the frontoparietal network. Further, I find that a network-level flexibility arises whereby the flexible optimization of sensory codes is accompanied by different parts of the frontoparietal network based on task demands - hence bringing about robustness through distributed computations. Therefore, I provide evidence that flexibility is a multiscale phenomenon that efficiently makes use of the different resources available in the brain, within sensory processing hierarchies in the temporal cortex and beyond. With this, I propose that flexibility of the sensory cortex is a special ‘structured’ kind of flexibility that efficiently makes use of the information from the external world and is complementary to the flexibility afforded by the association areas.
Keywords: plasticity; Non-human primate; Cognitive flexibility; visual perception; human intracranial electrophysiology; functional magnetic resonance imaging; statistical learning; predictive processing; Context-dependent flexibility; multitask learning; face processing; neural representation
