General Synaptics

Safety_Critical_Machine_Learning

One of the first (2002) Very-Deep Learning applications

MeteoNet™ was one of the first Very-Deep Learning applications on meteorology. MeteoNet™ is a software tool that processes thousands of temperature, humidity and pressure data on a national geographic scale (Italy) to make temperature and humidity forecasts for specific locations in the territory. MeteoNet™ was built with a Multilayer Perceptron with 10 hidden layers, 100 inputs and 100 outputs. MeteoNet™ has been developed as part of a research project of DIP.TE.RIS (DIPartimento del TErritorio e delle sue RISorse) department of University of Genova.

The training of the Multilayer Perceptron has been optimized with hybrid techniques based on EBP (Error Back Propagation), Genetic Algorithms and Simulated Annealing. This methodology allows the overcoming of local minima in gradient descent. Training the network took several months with the use of accelerators based on DSP (Digital Signal Processing) as at that time GPUs for machine learning did not exist.

Local temperature time series prediction

Meteonet™ has been realized for Dip.Te.Ris (Dipartimento del Territorio e delle sue Risorse) which is now called DISTAV (Dipartimento di Scienze della Terra, dell’Ambiente e della Vita)

Geographic locations temperature time series prediction

Four DSP acceleration boards have been used for the training process with thousands of meteorological data

WhoWeAre WhatWeDo Technology

Customers Partners News

General Synaptics

Aerospace_and_Defence_Machine_Learning_Company

VAT NUMBER:_IT02670700992

REA NUMBER: GE-503104

NATO_CAGE_CODE: AK845

Email:_luca.marchese@synaptics.org