Scientific seminar at Como Campus
Introduction to The AI and CV revolution
Darian Frajberg
DEIB PhD student - Politecnico di Milano
Geometric Deep Learning: going beyond Euclidean data
Federico Monti
PhD student - Università della Svizzera Italiana, ICS (Institute of Computational Science)
Politecnico di Milano - Polo di Como (Valleggio building - VS8B Room)
October 30th, 2017
10.30 am - 12.00 pm
Contact:
Carlo Bernaschina
Research Line:
Data, web, and society
Darian Frajberg
DEIB PhD student - Politecnico di Milano
Geometric Deep Learning: going beyond Euclidean data
Federico Monti
PhD student - Università della Svizzera Italiana, ICS (Institute of Computational Science)
Politecnico di Milano - Polo di Como (Valleggio building - VS8B Room)
October 30th, 2017
10.30 am - 12.00 pm
Contact:
Carlo Bernaschina
Research Line:
Data, web, and society
Sommario
On October 30th, 2017 Darian Frajberg and Federico Monti will held a seminar in Como Campus (Valleggio building, VS8B Room), with two different talks:
10.30 am
Introduction to The AI and CV revolution
With the Multimedia revolution more and more data has become available and easily accessible. Extracting useful information from this raw data is a complex task.
Machine Learning techniques have become more and more common in this environment.
We will introduce technologies and approaches which are at the core of this revolution, with a particular focus on Computer Vision and Deep Neural Networks.
The presentation of "Introduction to The AI and CV revolution" is available at https://www.slideshare.net/darian_f
11.00 am
Geometric Deep Learning: going beyond Euclidean data
Deep Learning approaches are rapidly becoming ubiquitous in our life nowadays. From the cameras/microphones we all have in our smartphones, to the latest models of cars we use for commuting everyday, everything today is introducing Deep Learning for solving the most complicated and challenging tasks. Convolutional Neural Networks probably represent the most successful example of these techniques at present.
While CNNs have been extensively applied on data which has some intrinsic Euclidean structure (e.g. images, videos or audio signals), in the last years there have been several attempts to extend these particular techniques to data defined on non-Euclidean domains (i.e. graphs and manifolds). This trend falls under the name of Geometric Deep Learning and will be the main topic of this talk. We will revise the basics of signal processing and graph theory which are used for defining convolutions over graphs and manifolds. We will present some of the latest state-of-art Graph Convolutional Neural Networks which have been introduced in the literature. We will illustrate typical tasks that can be solved on non-Euclidean structured data and finally, we will present possible and interesting applications which currently represent the main objective of this new and emerging field.
The presentation of "Geometric Deep Learning" is available at http://geometricdeeplearning.com/slides/polimi2017.pdf
10.30 am
Introduction to The AI and CV revolution
With the Multimedia revolution more and more data has become available and easily accessible. Extracting useful information from this raw data is a complex task.
Machine Learning techniques have become more and more common in this environment.
We will introduce technologies and approaches which are at the core of this revolution, with a particular focus on Computer Vision and Deep Neural Networks.
The presentation of "Introduction to The AI and CV revolution" is available at https://www.slideshare.net/darian_f
11.00 am
Geometric Deep Learning: going beyond Euclidean data
Deep Learning approaches are rapidly becoming ubiquitous in our life nowadays. From the cameras/microphones we all have in our smartphones, to the latest models of cars we use for commuting everyday, everything today is introducing Deep Learning for solving the most complicated and challenging tasks. Convolutional Neural Networks probably represent the most successful example of these techniques at present.
While CNNs have been extensively applied on data which has some intrinsic Euclidean structure (e.g. images, videos or audio signals), in the last years there have been several attempts to extend these particular techniques to data defined on non-Euclidean domains (i.e. graphs and manifolds). This trend falls under the name of Geometric Deep Learning and will be the main topic of this talk. We will revise the basics of signal processing and graph theory which are used for defining convolutions over graphs and manifolds. We will present some of the latest state-of-art Graph Convolutional Neural Networks which have been introduced in the literature. We will illustrate typical tasks that can be solved on non-Euclidean structured data and finally, we will present possible and interesting applications which currently represent the main objective of this new and emerging field.
The presentation of "Geometric Deep Learning" is available at http://geometricdeeplearning.com/slides/polimi2017.pdf