Seeing the forest through the trees: Structure representation under noisy processing

Thesis

For humans to optimally handle the constant stream of input they receive from their surroundings, they need to be able to separate signal from noise. The understanding that information is structured can help humans to organize information better and improve the detection of signal in a noisy input. This thesis deals with two main themes: the neural representation of structure and how structure representation can adapt in the light of noisy processing. First, I investigate how the human brain represents the structure of magnitude and how through structural alignment it can learn efficiently about new stimuli that have the same underlying structure. Next I present evidence from a new task that magnitude is represented neurally along a single axis but separated in parallel lines based on the current context. In the following chapter I investigate through model simulations how normalization can confer robustness under accumulation noise and test these predictions in humans. Finally, I present neural evidence for a computational model of robust processing when two pieces of information have to be assessed simultaneously. Taken together I hope to demonstrate that neural signals carry a low-dimensional representation of structure and that the brain can adaptively change those representations to improve performance.

Authors: Luyckx, F

• DOI: 10.5287/ora-no6zk79pg
• Type of award: DPhil
• Level of award: Doctoral
• Awarding institution: University of Oxford
• Language: English
• Keywords: number representation; magnitude representation; selective integration; normalization; Decision-making; noisy processing; representational similarity; computational modelling; EEG
• Subjects: Psychology, Experimental; Cognitive neuroscience