Consejo Superior de Investigaciones Científicas · Universidad de Sevilla
sp    en

What has Deep Learning ever done for us? The Convis toolbox for modelling visual responses from retina to cortex built on PyTorch

Jacob Huth, Institute of Vision, Paris (France)

Deep convolutional neural nets are a hot topic in machine learning. Their success stems in part from the computational frameworks that were developed for their specific use case, such as Theano, Torch and Tensorflow.
I will give an introduction on how the main features of these frameworks, such as GPU accelerated operations and automated differentiation, can be used for non-deep-learning tasks, such as implementing LN-cascade vision models or fitting parameters in a model for Retinal Ganglion Cells.

JACOB HUTH studied Cognitive Science in Osnabrueck with a specialization in Neuroinformatics and received his Masters degree there in 2014. Since 2015 he has been working as a PhD student in Angelo Arleos lab at the Institute of Vision in Paris to investigate perceptual decline in old age and how it can be explained by neural mechanisms. He developed Convis, a vision model toolbox based on convolution and automated differentiation, written in Python.

Instituto de Microelectrónica de Sevilla IMSE-CNM
13 Marzo 2018 · Salón de Grados · 11:00h.