Contributed Talk Sessions | Poster Sessions | All Posters | Search Papers

Poster Session B: Wednesday, August 13, 1:00 – 4:00 pm, de Brug & E‑Hall

Explaining neural mechanisms of age-related dedifferentation in the ventral stream through deep neural networks

Akilles Rechardt¹, Gabriele Bellucci, Robert M. Mok²; ¹Royal Holloway University of London, ²Royal Holloway and Bedford New College

Presenter: Akilles Rechardt

Healthy older adults show less distinct visual category representations compared to young adults in later parts of the ventral visual stream (VVS), a phenomenon known as dedifferentiation. However, the neural mechanisms causing this are unclear. We used a deep convolutional neural network to model the VVS and applied noise and synaptic damage to different layers of the model while reading out category distinctiveness from a late, category-selective layer that models inferior temporal cortex (IT). We expected greater damage to IT to cause stronger dedifferentiation. As predicted, greater damage led to greater dedifferentiation. However, damage to earlier layers of the model (e.g., V1) caused greater dedifferentiation in IT compared to damaging later layers. This suggests that age-related dedifferentiation in IT could result from damage to upstream areas of the network. Our findings also match structural brain imaging work indicating early to late VVS white matter tract integrity is related to the distinctness of category representations. In sum, our modelling approach for the first time provides a mechanistic explanation for age-related dedifferentiation.

Topic Area: Object Recognition & Visual Attention

Extended Abstract: Full Text PDF