Statistical learning met a lightning theoretical and practical development during last years, since it allows to handle problems usual methods were powerless to solve.

This training is meant for ingeneers or researchers interested in applications in process modelling, shape recognizing (vision and speech), automatic classifying, signal processing, bioinformatics, non-linear, diagnostic, language treatment, etc..., and who think they have worked out all the traditional solutions in the field of modelling and data treatment.

At the end of the training, the student knows the bases of artificial learning, with a particular attention to two types of tachnics : neural networks and support vector machines (SVM). He also knows the main learning algorithms, the typical applications and performances, as well as the conceivable strategies to develop an industrial application.


GENERAL PRESENTATION OF ARTIFICIAL LEARNING

• What is artificial learning?
• Problems to solve to to realise an efficient application
• Introduction to neural networks and support vector machines; overview of their applications

NEURAL NETWORKS FOR STATIC MODELLING

• Neural networks and non-linear regression
• Learning of static neural networks
• Selection of variables : methodology and practical illustrations
• Performances appraisal of a model and selection of a model: crossed validation, leave-one-out, virtual leave-one-out, confidence intervals appraisal
• Selecting a model; examples of applications to industrial problems
• Semi-physical modelling: how to use available knowledge (physical, chemical, etc.) to structure a neural network; examples of industrial applications

SUPPORT VECTORS AND NEURAL NETWORKS MACHINES FOR CLASSIFYING

• Introduction to classifying : Bayes’ formula, bayesian classifiers
• The Perceptron algorithm and the linear support vector machines
• Non-linear support vector machines. The kernel trick.
• Appraisal of the probability for an object to be in a class with the help of neural networks
• Industrial examples

NEURAL NETWORKS FOR DYNAMIC MODELLING AND CONTROL

• Definitions and presentation of the appliance domains (filtering, control)
• Learning rules dynamic neurons neural networks
• Postulated models (determinists or probabilists) and associated predictive model
• Parameters appraisal for a neural predictor
• Non-adaptative and adaptative learning of loop neural networks
• Example of appraisal of parameters of a dynamic non-linear model

HOW TO START?

• Possible strategies for an enterprise that wish to apply artificial learning-based methods
• Case studies, with possibilities for the trainers to briefly expose their concerns

PRATICAL WORK

• · Practical work allows the trainers to run their support vector machine and the neural networks for modelling and classifying. Two software are used(on PC with Windows): a general software (Neural Networks Matlab’s tool box) and a software specialized in the development of industrial applications (Netral’s Neuro One).

SPEAKERS
Gérard DREYFUS, Dr of Sciences, Professor at the ESPCI
Pierre ROUSSEL-RAGOT, Ingeneer, Dr of Sciences, Lecturer at the ESPCI
Yacine OUSSAR, Ingeneer, Dr at the Pierre and Marie Curie University, Lecturer at the ESPCI
Jean-Luc PLOIX, President of Netral

Cost: 1 550 Euros VAT not included


Please contact us if you are interested in this training.

---

SECOND LEVEL
20th – 21st November 2008


APPLICATIONS AND REALISATIONS


This training is designed for ingeneers or researchers who have studied in the initiation course, or who already have equivalent thoric and practical knowledge. At the end of the training, the student is able, by using the case studies that have been presented, undertake a practical project by making good use of the technic(s) that best suits the problem.


NEURAL NETWORKS FOR AUTOMATIC RECOGNITION OF HANDWRITING

• The problem of shape recognizing the role of classification, neural classifiers.
• Hidden Markov Models (HMM) to model sequences; common facts about neural networks and Markov’s models ; learning algorithms
• Neural networks and HMM hybrids
• Case study and demonstrations: recognizing handwritten figures (postcodes), recognizing cursive writing (automatic reading check literal amounts)

DESIGN OF EXPERIMENTS

• Design of experiments: what for?
• Design of experiments for knowledge-based models and neural models

PROCESS COMMAND BY NEURAL NETWORKS; APPLICATIONS TO FOOD-PROCESSING

• Neural Technics to estimate indirect measurement
• Technics of neural command: open-loop command, inverse model command, command with intern model and prdictive command
• Applications to cooling process, drying process, micro-filtration process
• Predictive command of Champagne yeast production
• Prediction of the time the fermentation ends for beer and yoghurt production

TOPOLOGIC MAP AND APPLICATIONS

• Kohonen topologic maps
• Vector quantification
• Case study: applications of neural networks for teledetection

MATERIAL REALISATIONS

• Why electronic realisations?
• Didactical examples: nuclear physics, meteorology
• Material architecture types
• Available products on the market
• New approaches: FPGA, DSP

SPEAKERS
Szolt WIMMER, Ingeneer at Vision Objects
Jean-Luc PLOIX, President of Netral
Eric LATRILLE, Research Ingeneer at the INRA
Sylvie THIRIA, Professor Versailles Saint-Quentin en Yvelines University
Bruce DENBY, Professor at Pierre and Marie Curie University

Cost:1100 Euros VAT not included


Please contact us if you are interested in this training.

---

ORGANIZATION

Schedule: 9H30-17H00. Lunch in common

Address: E.S.P.C.I., 10 rue Vauquelin, 75005 PARIS

Cost and package:
1st level: 1 550 Euros VAT not included, 2nd level: 1 100 Euros VAT not included. Combined 1st and 2nd level: 2 200 Euros package VAT not included.
Charges include the lesson documents and lunch.

Discount: a 25% discount is granted to schools and colleges.

Next session: 23-27 March 2009, contact us.

---

GENERAL CONDITIONS

Continuous training: NETRAL is registered at the Direction Régionale du Travail, de l'Emploi et de la Formation Professionnelle d'Ile de France. Contact NETRAL to obtain the register number.

Payment conditions: Cash, by check or by bank transfer. Exchange costs and bankink costs are to be paid by the client.

Conditions required for cancellation: Lessons can be called off by the client in the following conditions:

• Less than 15 days before the beginning of the seminar: 50% of the payment will be required
• Less than 4 days before the beginning of the seminar: The whole payment will be required.

NETRAL reserves the right to cancel a lesson if it totals less than 8 students. Then, the customer will be allowed to choose another date that best suits him among those proposed in the calendar.

Contesting : In case of lawsuit or contesting, only the Nanterre Trade tribunal will be competent.

All the conditions above are considered as accepted by the client as soon as he has placed an order.

---

REGISTER

Please contact us if you want to register for this training or if you want to receive further information about it.

otherwise

Fill the enclosed form and send it to Netral.

© Netral - June 2008