Welcom on Edge-ML Channel !

These educational videos present the Edge-ML solutions. They are for Data-Scientists and Data-Engineers, but also for Business Managers, Marketing and deciders. These videos are classified into 3 categories in order to address the different audiences in a better way.

Icone video mathMathematics and algorithmics

Icone video techniqueTechnique and implementation

Icone video usageApplications and use cases

Icone video math

Icone video usage

An introduction to Auto-ML Edge-ML (talk 1/4)

The classical approches of Auto-ML automatise the best practices of Data Scientists : they assess a great number of models and select the most precise.
Edge-ML offers a real break through thanks to MODL approch that allows to produce efficient models in a simple way with warranted robustness.

Icone video math

Supervised discretisation and grouping (talk 2/4)

This video presents the mathematical approches MODL used by Edge-ML in the phase of pre-treatment of data :

  • Supervised discretisation of numerical variables
  • Supervised grouping of the modality of the categorical variables
  • Robust filtering of the variables which are not-corelated to the target

Those stages come before the construction of the Set classifier which will be introduced in the next video.

Icone video math

Bayesian Ensemble classifier (talk 3/4)

This video presents the next step of automated pipe of Machine Learning implemented by Edge-ML. The aim is to build multi-varied classifier without parameters to optimise, starting from mono-varied pre-treatments (discretisation and grouping).

More precisely, this talk presents 4 ways of improvement of the Naive Bayes classifier, in order to make this classifier more accurate while keeping its robustness.

Icone video math

Icone video usage

Extraction of sequential rules (talk 4/4)

This video presents the MODL approch devised for the preparation of the sequential data (texts, web sessions, logs…). The aim is to identify the sub-sequences of the data set allowing to describe in a precise and robust way the distribution of the classes to be predicted (Auto Features Engineering).