MADAS Seminar Series 2017/18 
Econet: a Metanet system for pattern recognition. A real Deep Learning
Paolo Massimo Buscema
Semeion Centro Ricerche su Reti Neurali Artificiali (Roma) and University of Colorado (Denver)

Wednesday 21 March @14:30

Topic 1: Image Pattern Recognition


On tasks such as the recognition of handwritten digits, traditional methods from machine learning and computer vision have always failed to beat human performance.

Inspired by the importance of diversity in biological system, we built an heterogeneous system that could achieve this goal. Our architecture could be summarized in two basic steps.

  1. We generate a diverse set of classification hypothesis using Convolutional Neural Networks (CNN) and recent innovative Neural Networks (SNN).
  2. Then, we forward all the Machine Learning already trained in the a parliament of classifiers where all the hypothesis, despite of their accuracy or methodology, are considered.  The judges of the parliament are a new family of Meta Classifiers, whose name is Met Net.

We have applied this new strategy of “real” Deep Learning on the very competitive MNIST handwriting benchmark and our method is seems to be promising. It  surprisingly shows that artificial diversity to be the key for success in decision making.


Topic 2 :Data Fusion and  Theory of Impossible Worlds


One of the most challenging targets in Machine Learning research is the data fusion between datasets with completely different attributes (variables) and observations (records).

In the classical theory of possible worlds and its updating, an individual can be transferred from a source world (dataset A) to a destination world (dataset B), if, and only if, at least one of its attributes is shared in both worlds.

Thus, according to the classical theory, it is impossible to connect  two datasets with void intersection among their attributes (properties)  and with void intersection among their records (individuals).

We are to going to present a way to make such impossible transfer actually possible.

We have named this approach: “Theory of Impossible Worlds” (TIW for short).



Paolo Massimo Buscema

Professor and Computer Scientist, expert in Neural Networks and Artificial Adaptive Systems.

Director of Semeion, Research Centre of Sciences of Communication (Rome – Italy), Institution recognized by the Italian Ministry of University and Scientific Research (MIUR). Full Professor Adjoint at the Department of Mathematics and Statistics Sciences, University of Colorado (Denver, USA). Founder e Member of the Center for Computational and Mathematical Biology (CCMB – University of Colorado).

Adviser of the Presidency of the Council of Ministers, the Italian Government. From 2003 to 2007: Adviser of New Scotland Yard, London, UK.

He has designed, constructed and developed new models and algorithms of Artificial Intelligence. Author of scientific publications on theoretical aspects of Natural Computation, with over 250 titles (scientific articles, essays, and books (24) on the same subject) and over 70 Software Systems used in many University and international Research Centres. Inventor of 28 international patents.

His current projects are: 1. EEG in autism diagnosis (I-FAST Algorithm). 2. Geographic Profiling (TWC Algorithm). 3. Theory of Impossible World (ANNs working with many data sets simultaneously, not linked each others). 4. Deep Learning and Real Deep Learning using an eco system of different ANNs cooperating into a set of Meta Nets. 5. Data Matrix Theory: a algebraic theory of non linear operators using adaptive algorithms (ANNs and Evolutionary Systems).