Home » Node » 18751

AI: Advances in Imagination

Christian Napoli
Data dell'evento: 
Martedì, 24 September, 2019 - 09:15
Aula A6

Modelling human-based phenomena is not an easy task. Moreover, implementing such a model by means of a traditional algorithmic approach - based on ab initio assumptions - gets nearly impossible.
On the other hand, leaving the idea of any prodromal hypothesis, it is possible to let models emerge by themselves. That can be done by applying specific mathematical methods and several advanced tools such as feature representation transforms and neural networks with a little pragmatic perspective on chaotic phenomena. Sometimes results gets as surprising as unexpected, but, eventually, quite interesting. The interpretation of apparently chaotic phenomena relies on our ability to extrapolate symmetries and mathematical relations, therefore it depends on how we represent the information that, while apparently hidden in the initial form, can be also described in a different manner. When we change frame - or basis, in a strictly mathematical fashion - we can decompose a signal in new sets of coefficients that could dramatically change our perspective. Signal decomposition techniques such as functional analysis, multiresolution transforms and polynomial decompositions, permits us to pack the information contained on a signal into a few significant numerical  coefficients. By doing so we could relate few coefficients to specific subsets of features emerging on the observed phenomena although initially unfathomable on the registered signal. This kind of informational assets is generally ideal for automatic classification and model extrapolation by means of machine learning techniques and neural networks.
Such models can be applied to many fields of research to understand and exploit non periodical phenomena as well as for classification purposes in many different fields. Such approaches constitutes optimal tools in order to understand, model and predict the behaviour of human groups as well as to better understand and predict important aspects of human-computer interaction.

In this talk the following cues will be mentioned:

        - Predictive and forecasting neural models
        - Dynamical time-evolving AI-based models
        - Continuous machine learning approaches
        - AI-oriented feature enhancement and suppression
        - Non-trivial human-based classification tasks
        - Context and environment driven neuro-classifiers
        - Behavioural models identification and predictive interaction
        - A new biometrics perspective and other things to come....

© Università degli Studi di Roma "La Sapienza" - Piazzale Aldo Moro 5, 00185 Roma