On May 16th at 1:00 PM in room 135
“Artificial Intelligence and Machine Learning: Linearly Evident Information and Hidden Insights in Data”
The information extractable from a dataset can either be evident and thus identifiable with simple linear multivariate algorithms, or it can be so implicit in the dataset and seemingly “weak” that it remains invisible to classical analysis tools. In this latter case, adaptive algorithms capable of computing highly non-linear functions need to be used. However, often deep ANNs (Artificial Neural Networks) already present in the literature are not sufficient for this purpose. Instead, new adaptive algorithms and new ANNs, not based on the classical learning technique known as “gradient descent”, have proven effective. Three cases of this kind will be shown:
a. Hidden Relationships: The issue of an investigation into a criminal gang that was about to be closed without having arrested the real leaders, but only the “small fish”;
b. Anomaly Detection: The problem of “fraudulent” and/or “inappropriate” refunds for healthcare expenses within a supplementary healthcare fund;
c. Implicit Marketing: The analysis of a small tourist dataset showing the generic characteristics of 80 travelers visiting 32 cities worldwide.
