Testing Infants Online
Rhodri Cusack, Lorijn Zaadnoordijk
This talk will describe the potential of testing infants online, recording from the user's webcam, and summarize through the challenges and ways to address them.
Talk live broadcasting information
July 7th, 15:00 - 16:30 (BST) / 7:00am - 8:30am (PDT) -- Live Q&A
July 7th, 22:00 - 23:30 (BST) / 14:00am - 15:30am (PDT) -- Recording only
Deep neural networks as a model of learning during the helpless period of infancy
I will present evidence from neuroimaging that even in early infancy, many brain systems are surprisingly mature. I argue that human infants are not helpless because their brains are too immature to function, but rather because they require an extended period of experience-dependent learning before behaviour can develop. I present deep neural network models to begin to provide models of this critical hidden process.
Talk live broadcasting information
July 9th, 13:30 - 15:00 (BST) / 5:30am - 7:00am (PDT)
July 9th, 20:30 - 22:00 (BST) / 12:30 - 14:00 (PDT)
Q&A live broadcasting information
July 9th, 15:00 - 16:00 (BST) / 7:00am - 8:00am (PDT) (https://american.shortcm.li/vICIS-S29)
Conducting and improving internet-based preferential looking studies
Lorijn Zaadnoordijk, Rhodri Cusack
Looking behaviour has been a valuable measure in infant studies. Recently, there has been an increasing interest in conducting such studies online as opposed to in-lab. There are many benefits to acquiring infant data via online studies, but it has unique challenges that have to be addressed. Here we evaluate five challenges and present possible solutions.
July 6th , 12:00 - 13:30 (BST) / 11:00 - 12:30 (PDT)
Poster number: P1-G-126
Investigating the application of unsupervised deep neural networks as a model for infants’ visual brain development
Anna Truzzi, Rhodri Cusack
In neuroscience, deep neural networks are being used as models for the adults’ brain because when a network has been trained on image recognition tasks the way it represents visual stimuli becomes similar to how the brain visual areas represents them. However, it is still unclear which features are similarly represented and whether it is possible to use the learning process itself as a model for infants’ brain development. In this project we investigate which features make deep neural networks similar to the brain and whether and how those features changes before and after the networks has been trained.
July 6th, 12:00 - 13:30 (BST) / 11:00 - 12:30 (PDT)
Poster number: P1-B-26
Infant curiosity and effective learning: insights from computational modelling
Anna Kravchenko, Lorijn Zaadnoordijk and Rhodri Cusack
Previous studies of curiosity show that, at least sometimes, it is beneficial to learn about easier events before learning about complex events. However, it remains unclear whether this is always the case. We have designed and conducted two modelling experiments to explore the conditions under which learning can be facilitated by a simple-to-complex hierarchy.
July 8th, 12:00 - 13:30 (BST) / 11:00 - 12:30 (PDT)
Poster number: P3-F-94
Semantic relationships emerge from visual temporal co-occurrences: A statistical analysis of a learning mechanism in early infancy
Clíona O’Doherty, Anna Truzzi and Rhodri Cusack
Infants learn to recognise the objects they see around them through experience. Investigating a signal which may facilitate this process, we quantified the temporal co-occurrence patterns in a naturalistic movie dataset and the semantic quality of object clusters. It was found using a statistical analysis that temporal correlations of objects could be clustered into semantically meaningful categories. Learning across a time interval which was neither too short nor too long (1 min) allowed superordinate category clusters to be learned by a deep neural network learning model. Thus, the results presented here provide a possible learning signal which could be tested in future infant studies.
July 6th,12:00 - 13:30 (BST) / 11:00 - 12:30 (PDT)
Poster number: P1-G-120