What: Challenging the deep learning approach to word recognition with topography and predictive coding
Where: BCBL Auditorium and zoom room 2. (If you would like to attend to this meeting reserve at info@bcbl.eu)
Who: Thomas Hannagan, PhD. Cognitive Neuroimaging Unit, CEA, France.
When: Thrusday, May 26th at 2:30 PM.
I will present some recent deep learning models of word recognition in the cortex. Despite their success, these neural networks face significant challenges. Most importantly, unlike the cortex, typical deep learning models are not topographically organized, do not have feedback connections, and are not endowed with activation dynamics. I will present recent work studying how these features can be introduced into deep learning models.