ERC Foundations of Cognition (FOUNDCOG)
FOUNDCOG NewsSee a summary of some of our latest results on the news item Making Machines that Learn Like Humans
FOUNDCOG OverviewHow does the human mind develop? The FOUNDCOG project aims to address this question by using neuroimaging of infants, in their critical first year after birth, to capture the development of the brain and its functions. We will interpret the results by comparing them to computational models of learning from artificial intelligence. In addition to characterising the development of the human mind, we aim to understand how it is disrupted by brain injury and to produce new diagnostic tools for neonatologists.
The First Year of Life Human babies undergo a period of huge developmental growth over their first year of life. When compared to other mammals, the time it takes for humans to learn to walk, talk and communicate is extremely lengthy. Yet, we emerge with highly developed brains, and the capacity for hugely complex cognitive functions. What changes are taking place over this crucial first year? Why must we remain dependent on our caregivers for so long? How can we unlock some of the secrets of our wonderful infants' minds? We aim to answer some of these questions by using MRI of infants from 3 to 12 months old, examining the functions of their visual system and how this brain region develops as they learn to navigate the world. Deep Learning as a Model of Development Deep learning, the technology underlying rapid advances in modern artificial intelligence, continues to prevail as an invaluable model of human vision as it functions in the brain. We believe that this utility may extend to modelling infant learning. By working with state-of-the-art unsupervised deep neural networks, our team is testing possible mechanisms of learning to categorise visual classes, and we are thinking about how we might test these theories in our neuroimaging studies. In turn, we expect to find a reciprocal benefit for the machine learning community; the possibility of implementing more human-like learning mechanisms into these artificial neural networks certainly holds exciting potential. See our current vacancies. |
FOUNDCOG is funded by an Advanced Grant from the European Research Council from 2018-2023.
|