Description
 

Artificial intelligence (AI) is undergoing an explosive phase — machines can now accomplish complex specific tasks at a level that exceeds human skills. At the basis of this performance is the ability to understand the sensory input from the external world and to associate it with effective strategies to achieve the desired goal.
 
This advanced school aims to combine different yet strongly coupled perspectives: first, theoretical approaches, which focus on principles, algorithms, and their applications to computer science; second, the relationship with experimental neuroscience, which has inspired the latest generation developments in AI and has, in turn, benefited from the ability of AI to investigate the computations underlying complex cognitive processes; third, applications such as robotics, gaming, etc.
 
The topics covered include:
 
•    deep learning and its relation to vision and language
•    reinforcement learning and decision making
•    sensorimotor learning
•    the ethics of artificial intelligence and its impact on society


Invited lecturers:
B. Biggio (Cagliari U.)
D. Braun (Ulm U.)
P. Dayan (Max Planck Institute for Biological Cybernetics)
J. Di Carlo (MIT)
B. Kappen (Radboud U., Nijmegen)
T. Lattimore (DeepMind, London)
M. Pelillo (Ca' Foscari U.)
J. Peters (Technische Universitaet Darmstadt)
L. Rosasco (U. Genova, MIT & IIT)
N. Tishby (The Hebrew University of Jerusalem)
R. Zecchina (U. Bocconi )


 
Financial support for students from India is available under the Pratiksha Trust Scholarships, ICTS Bangalore
Go to day
  • Monday, 12 November 2018
    • 08:00 - 18:50 Monday, 12 November
      • 08:00 Registration & Administrative Formalities 50'
      • 08:50 Welcome Address 10'
        Speaker: A. Celani (ICTP)
      • 09:00 Neural Reinforcement Learning 1: Prediction 1h45'
        Speaker: P. DAYAN (Max Planck Institute for Biological Cybernetics, Tuebingen)
      • 10:45 Coffee break 30'
      • 11:15 Tutorial - Introduction to the Visual System 1h45'
        Speaker: D. ZOCCOLAN (SISSA)
      • 13:00 Lunch break 1h30'
      • 14:30 Tutorial - Learning from Noisy Sequential Observations 1h45'
        Speaker: C. MATHYS (SISSA)
      • 16:15 Coffee break 30'
      • 16:45 Reverse Engineering Human Vision 1: Where are the Neural Computations? 1h45'
        Speaker: J. DICARLO (MIT)
        Material: Video
      • 18:30 Welcome reception 20'
  • Tuesday, 13 November 2018
    • 09:00 - 18:30 Tuesday, 13 November
      • 09:00 Neural Reinforcement Learning 2: Choice 1h45'
        Speaker: P. DAYAN (Max Planck Institute for Biological Cybernetics, Tuebingen)
      • 10:45 Coffee break 30'
      • 11:15 Reverse Engineering Human Vision 2: How Can We Model those Computations? 1h45'
        Speaker: J. DICARLO (MIT)
        Material: Video
      • 13:00 Lunch break 1h30'
      • 14:30 Perceptrons and Gradient Descent Learning Rules 1h45'
        Speaker: B. KAPPEN (Radboud U., Nijmegen)
        Material: Slides Suggested exercises
      • 16:15 Coffee break 30'
      • 16:45 Hands-on tutorial: An Introduction to Neural Networks 1h45'
        Speaker: A. PEZZOTTA (ICTP)
        Material: Material
  • Wednesday, 14 November 2018
    • 09:00 - 18:30 Wednesday, 14 November
      • 09:00 Neural Reinforcement Learning 3: The Self and the Other 1h45'
        Speaker: P. DAYAN (Max Planck Institute for Biological Cybernetics, Tuebingen)
      • 10:45 Coffee break 30'
      • 11:15 Reverse Engineering Human Vision 3: What Is Still Missing? 1h45'
        Speaker: J. DICARLO (MIT)
        Material: Video
      • 13:00 Lunch break 1h30'
      • 14:30 Control Theory and Path Integral Methods - 1 1h45'
        Speaker: B. KAPPEN (Radboud U., Nijmegen)
        Material: Slides Suggested exercises Video
      • 16:15 Coffee break 30'
      • 16:45 Information Theory of Deep Learning - Rethinking Statistical Learning theory for Large-scale Learning 1h45'
        Speaker: N. TISHBY (The Hebrew University of Jerusalem)
        Material: Video
  • Thursday, 15 November 2018
    • 09:00 - 18:30 Thursday, 15 November
      • 09:00 Information Theory of Deep Learning - The Special Role of Stochastic Gradient Descent and the Information Bottleneck Limit 1h45'
        Speaker: N. TISHBY (The Hebrew University of Jerusalem)
        Material: Video
      • 10:45 Coffee break 25'
      • 11:10 Group Photo 5' ( ICTP Cafeteria - terrace )
      • 11:15 Control Theory and Path Integral Methods - 2 1h45'
        Speaker: B. KAPPEN (Radboud U., Nijmegen)
        Material: Slides Video
      • 13:00 Lunch break 1h30'
      • 14:30 Adversarial Machine Learning (Part I) 1h45'
        Speaker: B. BIGGIO (Cagliari U.)
        Material: Slides Video
      • 16:15 Coffee break 30'
      • 16:45 Hands-on tutorial: Overfitting and Regularization 1h45'
        Speaker: A. ANSUINI (SISSA)
        Material: Material
  • Friday, 16 November 2018
    • 09:00 - 18:50 Friday, 16 November
      • 09:00 Information Theory of Deep Learning - What Do the Layers of Deep Neural Networks Represent? 1h45'
        Speaker: N. TISHBY (The Hebrew University of Jerusalem)
        Material: Video
      • 10:45 Coffee break 30'
      • 11:15 Adversarial Machine Learning (Part II) 1h45'
        Speaker: B. BIGGIO (Cagliari U.)
        Material: Slides Video
      • 13:00 Lunch break 1h30'
      • 14:30 Hands-on tutorial: Convolutional Neural Networks: Theory 1h45'
        Speaker: G. MATTEUCCI (SISSA)
        Material: Slides
      • 16:15 Coffee break 30'
      • 16:45 Hands-on tutorial: Convolutional Neural Networks: Practicals 1h45'
        Speaker: E. ANNAVINI (SISSA)
        Material: Material
      • 18:30 Get-together 20'
  • Monday, 19 November 2018
    • 09:00 - 19:30 Monday, 19 November
      • 09:00 Regularization in Learning Theory 1h45'
        Speaker: L. ROSASCO (Genoa U. and MIT)
        Material: Slides Video
      • 10:45 Coffee break 30'
      • 13:00 Lunch break 1h30'
      • 14:30 Robot Learning - 1 1h45'
        Speaker: J. PETERS (Technische U. Darmstadt)
        Material: Video
      • 16:15 Coffee break 30'
      • 16:45 Hands-on tutorial: Reinforcement Learning: theory 1h45' ( Adriatico Guest House - Informatics Laboratory )
        Speaker: D. HOFMANN (Emory U.)
        Material: Material
      • 18:30 Reception 1h0'
  • Tuesday, 20 November 2018
    • 09:00 - 20:15 Tuesday, 20 November
      • 09:00 Implicit Regularization 1h45'
        Speaker: L. ROSASCO (Genoa U. and MIT)
        Material: Slides Video
      • 10:45 Coffee break 30'
      • 11:15 Robot Learning - 2 1h45'
        Speaker: J. PETERS (Technische U. Darmstadt)
        Material: Video
      • 13:00 Lunch break 1h30'
      • 14:30 Typical Minima in Single-Layer and Multilayer Networks 1h45'
        Speaker: R. ZECCHINA (U. Bocconi, Milan)
        Material: Video
      • 16:15 Coffee break 30'
      • 16:45 The Central Role of Wide Flat Minima: Rarity, Geometrical Features and Accessibility 1h45'
        Speaker: R. ZECCHINA (U. Bocconi, Milano)
        Material: Video
  • Wednesday, 21 November 2018
    • 09:00 - 18:30 Wednesday, 21 November
      • 09:00 Regularization with (Random) Projections 1h45'
        Speaker: L. ROSASCO (Genoa U. and MIT)
        Material: Slides Video
      • 10:45 Coffee break 30'
      • 11:15 Robot Learning - 3 1h45'
        Speaker: J. PETERS (Technische U. Darmstadt)
        Material: Video
      • 13:00 Lunch break 1h30'
      • 14:30 Loss Functions for Deep Learning 1h45'
        Speaker: R. ZECCHINA (U. Bocconi, Milano)
        Material: Video
      • 16:15 Coffee break 30'
      • 16:45 Bandit Algorithms - 1 1h45'
        Speaker: T. LATTIMORE (DeepMind, London)
        Material: Video
  • Thursday, 22 November 2018
    • 09:00 - 18:30 Thursday, 22 November
      • 09:00 Computational Models of Sensorimotor Learning and Decision-making - 1 1h45'
        Speaker: D. BRAUN (U. of Ulm)
        Material: Video
      • 10:45 Coffee break 30'
      • 11:15 Computational Models of Sensorimotor Learning and Decision-making - 2 1h45'
        Speaker: D. BRAUN (U. of Ulm)
        Material: Video
      • 13:00 Lunch break 1h30'
      • 14:30 Bandit Algorithms - 2 1h45'
        Speaker: T. LATTIMORE (DeepMind, London)
        Material: Video
      • 16:15 Coffee break 30'
      • 16:45 Hands-on tutorial: Reinforcement Learning: Practicals 1h45'
        Speaker: M. ADORISIO (ICTP)
        Material: Material
  • Friday, 23 November 2018
    • 09:00 - 18:50 Friday, 23 November
      • 09:00 Computational Models of Sensorimotor Learning and Decision-making - 3 1h45'
        Speaker: D. BRAUN (U. of Ulm)
        Material: Video
      • 10:45 Coffee break 30'
      • 11:15 Bandit Algorithms - 3 1h45'
        Speaker: T. LATTIMORE (DeepMind, London)
        Material: Video
      • 13:00 Lunch break 1h30'
      • 14:30 Through the Philosopher’s Glass 1h45' ( Adriatico Guest House - Kastler Lecture Hall Area (Lower Level 1) )
        Speaker: M. PELILLO (U. Ca' Foscari, Venice)
        Material: Slides
      • 16:15 Coffee break 30'
      • 16:45 Hands-on tutorial: Adversarial attacks and representations decoding 1h45'
        Material: Material
      • 18:30 Reception 20'