Visual Perception to Higher Cognition

Exploring how simple sensory experiences turn into complex thinking is a fascinating topic for vision science. Researchers investigate how people make sense of object relationships, interpret and express emotions through facial expressions, navigate busy streets and sign reading, or appreciate the aesthetic value in art and nature.

At the heart of this exploration is the understanding that visual perception is not merely about seeing but involves a sophisticated network of cognitive processes. These processes interpret and give meaning to what we see, integrating visual information with prior knowledge, experiences, and expectations. This gives rise to various research questions. For instance, how do optical illusions mislead our perception and what does this reveal about our brain’s cognitive functions? How do attentional mechanisms contribute to our high-level cognitive processes, and how do these mechanisms differ from those employed in deep learning models?

Notably, the advent of vision-language models marks a significant stride in research on high-level visual cognition. For instance, the ability of large language models to tackle tasks like Raven’s Progressive Matrices — a benchmark for evaluating human abstract reasoning — underscores the potential of these technologies to achieve a level of visual intelligence akin to that of humans. Our research group is keen on further exploring various machine models to compare their capabilities with human cognitive processes, aiming to deepen our understanding of the neural mechanisms underpinning high-level cognitive functions.

Image from Huth et al., Nature, 2016