Understanding how intelligence works and how it can be emulated by machines is an age old dream and arguably one of the biggest challenges in modern science. Learning, with its principles and computational implementations, is at the very core of this endeavor. Recently, for the first time, we have been able to develop artificial intelligence systems able to solve complex tasks considered out of reach for decades. Modern cameras recognize faces, and smart phones voice commands, cars can see and detect pedestrians and ATM machines automatically read checks. In most cases at the root of these success stories there are machine learning algorithms, that is, software that is trained rather than programmed to solve a task. Among the variety of approaches to modern computational learning, we focus on regularization techniques, that are key to high-dimensional learning. Regularization methods allow to treat in a unified way a huge class of diverse approaches, while providing tools to design new ones. Starting from classical notions of smoothness, shrinkage and margin, the course will cover state of the art techniques based on the concepts of geometry (aka manifold learning), sparsity and a variety of algorithms for supervised learning, feature selection, structured prediction, multitask learning and model selection. Practical applications for high dimensional problems, in particular in computational vision, will be discussed. The classes will focus on algorithmic and methodological aspects, while trying to give an idea of the underlying theoretical underpinnings. Practical laboratory sessions will give the opportunity to have hands on experience.
RegML is a 20 hours advanced machine learning course including theory classes and practical laboratory sessions. The course covers foundations as well as recent advances in Machine Learning with emphasis on high dimensional data and a core set techniques, namely regularization methods. In many respects the course is a compressed version of the 9.520 course at MIT.
The course started in 2008 has seen an increasing national and international attendance over the years, with a peak of over 90 participants in 2014.
RegML 2020 will take place at the Department of Informatics, Bioengineering, Robotics, and Systems Engineering (DIBRIS) of the University of Genova, Via Dodecaneso 35, 16146 Genova.
Morning classes will be held in room 506. Afternoon labs will take place in rooms SW1, SW2, 217 and 218. Directions will be provided at the DIBRIS entrance in Via Dodecaneso 35.
Here you can find a list of hotels near the department (~ 20' walk) or in the city centre (~20' by bus).
Here is a list of places where you can go for lunch.
For more info write to:
modiana [at] unige [dot] it
vigogna [at] dibris [dot] unige [dot] it
Sapienza University of Rome
The past decade in computer vision research has witnessed the re-emergence of "deep learning", and in particular convolutional neural network (CNN) techniques, allowing to learn powerful image feature representations from large collections of examples. CNNs achieve a breakthrough in performance in a wide range of applications such as image classification, segmentation, detection and annotation. Nevertheless, when attempting to apply the CNN paradigm to 3D shapes, point clouds and graphs (feature-based description, similarity, correspondence, retrieval, etc.) one has to face fundamental differences between images and geometric objects. Shape analysis, graph analysis and geometry processing pose new challenges that are non-existent in image analysis, and deep learning methods have only recently started penetrating into these communities. The purpose of this tutorial is to overview the foundations and the current state of the art on learning techniques for non-Euclidean data. Special focus will be put on deep learning techniques (CNN) applied to Euclidean and non-Euclidean manifolds for tasks of shape classification, retrieval and correspondence. The tutorial will present in a new light the problems of 3D computer vision and geometric data processing, emphasizing the analogies and differences with the classical 2D setting, and showing how to adapt popular learning schemes in order to deal with non-Euclidean structures. The tutorial will assume no particular background, beyond some basic working knowledge that is a common denominator for students and practitioners in machine learning and graphics.
Max Planck Institute for Intelligent Systems
|Mon 29th||9:30-11:00||506||Class 1||Introduction to Statistical Machine Learning||-|
|11:30-13:00||506||Class 2||Tikhonov Regularization and Kernels||-|
|14:00-16:00||SW1-SW2-217-218||Lab 1||Binary classification and model selection||-|
|Tue 30th||9:30-11:00||506||Class 3||Early Stopping and Spectral Regularization||-|
|11:30-13:00||506||Class 4||Regularization for Multi-task Learning|
|14:00-16:00||SW1-SW2-217-218||Lab 2||Spectral filters and multi-class classification||-|
|Wed 1st||9:30-11:00||506||Tutorial 1||Emanuele Rodolà - Geometric Deep Learning||-|
|11:30-13:00||506||Tutorial 2||Krikamol Muandet - Hilbert Space Representation of Probability Distributions||-|
|Thu 2nd||9:30-11:00||506||Class 5||Sparsity Based Regularization||-|
|11:30-13:00||506||Class 6||Structured Sparsity||-|
|14:00-16:00||SW1-SW2-217-218||Lab 3||Sparsity-based learning||-|
|Fri 3rd||9:30-11:00||506||Class 7||Data Representation: Dictionary Learning||-|
|11:30-13:00||506||Class 8||Data Representation: Deep Learning||-|