Advanced School and Workshop on Soft Computing and Complex Systems

Broad structure

The workshop will be organised around three main activities: lectures, given by well known international experts, teamwork by the attendees, to a solve particular problem proposed by the lecturers and, finally, short presentations by the students about their own work and interests.

Each Lecture will last for one and a half hour. There will be a small break between the lectures. Each team will have a lecturer assigned to it to provide some help if needed. The results of each group will be in the form of a written document, that will be the possible basis for a paper to be submitted to an international conference.

Lecture 1A: 

Neural Computation and Applications in Time Series and Signal Processing.

• Speaker: Georg Dorffner
• Synopsis: The main architectures for neural computation will be reviewed. The particular architectures for time series prediction and signal processing will be studied.

Lecture 2A: 

Analog Computation

• Speaker: Félix Costa
• Synopsis: There will be short introduction to the new promising area of analog computation. Some mathematical results, about the power of this approach, will be presented as well as the implications for the theory of computation.

Lecture 1B: 

Neuro-Fuzzy Modeling

• Speaker: R. Babuska
• Synopsis: Clustering in its several forms will be studied for structure learning from data in fuzzy systems. Parameter optimisation through neural networks, composing neuro-fuzzy systems will be reviewed and experimented.

Lecture 2B: 

Intelligent Control

• Speaker: R. Babuska.
• Synopsis: The main recently developed techniques for advanced Intelligent Control will be studied and experimented. On-line learning techniques, the problems of dimensionality, will be discussed.

Lecture 1C: 

Multi-criteria Genetic Optimisation

• Speaker: Carlos Fonseca
• Synopsis: Genetic optimisation will be reviewed in its general formulation. The particular case of multi-criteria will then be developed and discussed.

Lecture 2C: 

Speaker: Juergen Schimdhuber

Universal learning algorithms based on the theory of universal induction and Kolmogorov complexity, with applications:

• Optimal Ordered Problem Solver
• Theory of Universal Learning Machines. Recent Results of an Ongoing Research Project.
• The Speed Prior. A New Simplicity Measure for Near-Optimal Computable Predictions (based on the fastest way of describing objects, not the shortest).
• Generalized Algorithmic Information, Generalized Algorithmic Probability, and Super Omegas

 

Recurrent Neural Networks

RNNs are artificial neural networks with adaptive feedback connections. From training examples they can learn to map input sequences to output sequences. They can implement almost arbitrary sequential behavior. RNNs are biologically more plausible and computationally more powerful than other adaptive models such as Hidden Markov Models (no continuous internal states), feed-forward networks and Support Vector Machines (no internal states at all)