convolve project presentation

5
ALESSANDRO PAPPALARDO – [email protected] MADDALENA ANDREOLI – [email protected] STEFANO CAGNINELLI – [email protected] March 15, 2017 @ DEIB, Politecnico di Milano

Upload: politecnico-di-milano

Post on 08-Apr-2017

75 views

Category:

Engineering


0 download

TRANSCRIPT

Page 1: Convolve project presentation

ALESS ANDRO PAPPALARDO – ALESS ANDRO1.PAPPALARDO@MAIL .POL IM I . I TMADDALENA ANDREOL I – MADDALENA.ANDREOL I@MAIL .POL IM I . I TSTEFANO CAGNINELL I – STEFANO.CAGNINELL I@MAIL .POL IM I . I T

March 15, 2017 @ DEIB, Politecnico di Milano

Page 2: Convolve project presentation

2

CNN and real time Convolutional Neural Networks (CNNs) are powerful machine learning models for pattern recognition

Many practical scenarios for CNNs have hard real time requirements

Self-driving cars Visually impaired aids Autonomous drones

Page 3: Convolve project presentation

3

ConvolutionConvolution layers in CNNs use a kernel matrix in order to extract a feature from their input;

They require a huge amount of computations (particularly if the imput is big) and are one of the most time–consuming operations a CNN performs.

Page 4: Convolve project presentation

4

What Convolve aims to doOur goal is to implement a fast convolution algorithm on a FPGA fabric in order to speed up these layers and therefore the whole computation.

This will give CNNs even more competitive edge in the field of realtime image recognition.

Page 5: Convolve project presentation

5

ConvolveatNECSTfb.me/ConvolveatNECST

@ConvolveatNECSTtwitter.com/ConvolveatNECST

ConvolveatNECSTwww.slideshare.net/ConvolveAtNECST