How transferable are features in deep neural networks?

Event Date: 
13 February, 2017 -
16:00 to 17:00
Venue: 
L2
Event: 
Speaker: 
Akanksha Paul
Abstract: 
Deep neural networks trained on images learn general features in the initial layers, which could be used for transfer. However, features learnt eventually transit from general to specific features of a particular task. So, elaborating on the general to specific transition of features in transfer learning. Further, evaluating the performance of transfer with dissimilar data sets.