29/11/2018
A few ideas from the first Data Science Nancy Meetup
1) Radu Stoica, Professor at Lorraine University and Researcher with "Elie Cartan" Institute of Lorraine
• spatial data : data sets with two components, position and characteristics;
• application domains : image analysis, environmental sciences, astronomy, industry;
• question one data scientist is interested to answer: "What is the pattern hidden in the data ?"
• the key hypothesis is: the pattern is a complex entity made of simple objects that interact;
• the answer may be brought to answer it by stochastic modelling shall be used;
• is about using prior knowledge of the studied phenomenon in order to build the models;
Examples of applications:
• animal epidemiology: sub-clinical mastitis for dairy herds in France;
• image analysis: road and hydrographic networks;
• cosmology: spatial distribution of galaxies;
• industry : spatio-temporal distribution of leaks on a water distribution network;
Note: things become more complex when time variable occurs;
Scientific trends in Data Science:
• contribute to answer fundamental scientific questions;
• transfer the developed knowledge to students and to social actors;
• be aware of the general impact: social, environmental, ethical;
• propose real win-win partnerships;
Question: what is the pattern hidden in the data? Writing a computer program to describe the color distribution and the horse silhouette is not a trivial challenge.
Figure: Robert Delaunay: Paysage nocturne (le fiacre). Centre Pompidou Metz.
2) Marianne Clausel, Professor with Lorraine University and Researcher with Elie Cartan Institute of Lorraine
• Textual data are at the core of AI;
Analysing textual data might be a challenging given the fact that:
• the format is not predefined;
• there is a specific vocabulary;
• may evolve in time (for a blog);
• there might be potential interdependencies inside and outside the document itself;
The question arising is how to stock, to classify and to analyse them;
A few examples of questions:
• creation of an interactive map allowing to visualize patents?
• can one predict the evolution of financial markets via social networks? Using sentiment analysis;
• analysis of the correspondence between the employees;
• analyzing health tweets: tweets, human sensors;
• can we anticipate the spread of diseases?
Julien Trombini, co-founder and President of Two-i SAS, a video content analytics start-up, and researcher with Elie Cartan Institute of Lorraine.
Several ideas on Big Data:
4 Vs of Data: Volume (Scale of Data), Velocity (Analysis of Streaming Data), Variety and Veracity (Uncertainty of Data);
Machine Learning techniques:
• Supervised learning: classification and regression;
• Unsupervised learning: clustering, dimensionality reduction, and association rule learning.
