« Learning Similarity with Siamese Neural Networks» by Hiba Al Qasir

Seminar by Hiba Al Qasir

Abstract

Siamese networks are a type of artificial neural network designed as two (or more) identical twin networks joined at their final layer by a distance layer. The distance layer is trained to predict whether the inputs are of the same category or not, by computing how similar or dissimilar the corresponding outputs of the twin networks are. These networks have found their ways in a broad set of problems such as facial identification, signature verification, etc. In our chairlift safety problem, we use Siamese networks to compare image features with features extracted from binary masks representing shape priors of the important objects in the scene.

This seminar will be held in english.