Projects

CONNECTS: Cognitive and Neural Computations of Semantics

In our everyday lives, we rely on existing relations among elements in our environment (i.e., semantic information) to interact efficiently with the world. This information can either be used to facilitate understanding by exploiting redundant (congruent) evidence or to signal out salient stimuli by highlighting unexpected (incongruent) elements. This duality raises fundamental questions about when and how our brains utilise stored semantic knowledge as its influence often differs by cognitive domain.

This seemingly paradoxical state represents a cognitive puzzle that questions whether the presence of (in)congruent contextual information in a given situation has a positive or negative impact on how we perceive, process, and remember information. CONNECTS seeks to solve this paradox with a multi-method approach, including behavioural, neural (fMRI and EEG), and computational data from artificial neural networks.

If you want to know more about the project, visit our site.

How do memories inform predictions?

One of the assumptions of predictive processing accounts is that high-order representations, and/or structures, send information downwards (i.e., feedback signals) to lower levels on the processing hierarchy. However, the nature of those high-order representations is still unknown.

In this project we point at memory traces as a candidate source of predictions. Using fMRI and functional retinotopy we aim at relating stored memory representations to feedback signals in the early visual cortex. Moreover, using high field fMRI we are tracing the feedback signals from episodic memories down to the layer level.

Predictive processing and the acquisition of information

Unexpected events are more informative than expected ones by mere definition: unexpected events carry information that the system could not anticipate and, therefore, are a perfect situation for our brains to update their knowledge. Conversely, fully expected events can be seamlessly integrated within our schemas, thus facilitating their processing and leading, in turn, to an easier access from memory.

In this project we aim at exploring this apparent contradiction by assessing the effects that different degrees of prediction fulfillment have on the acquisition of new information and the updating of previous knowledge.