Training CNN Models with Machine Intelligence NVIDIA DIGITS - - PowerPoint PPT Presentation

training cnn models with
SMART_READER_LITE
LIVE PREVIEW

Training CNN Models with Machine Intelligence NVIDIA DIGITS - - PowerPoint PPT Presentation

Training CNN Models with Machine Intelligence NVIDIA DIGITS Modern Infrastructure http://mi2.live What is MI2? MI2 Webinars focus on the convergence of machine intelligence and modern infrastructure . Every alternate week, I deliver


slide-1
SLIDE 1

Machine Intelligence Modern Infrastructure

http://mi2.live

Training CNN Models with NVIDIA DIGITS

slide-2
SLIDE 2

What is MI2?

MI2 Webinars focus on the convergence of machine intelligence and modern infrastructure. Every alternate week, I deliver informative and insightful sessions covering cutting-edge technologies. Each webinar is complemented by a tutorial, code snippets, and a video. MI2 strives to be an independent and neutral platform for exploring emerging technologies.

Register at http://mi2.live

slide-3
SLIDE 3

Objectives

  • What is a Convolutional Neural Network (CNN)?
  • Key Concepts
  • Overview of NVIDIA DIGITS
  • Demo
  • Summary
slide-4
SLIDE 4

Lifecycle of an ML model

Historical Data Training Dataset Testing Dataset Algorithm Evaluation Model

Production Data Prediction
slide-5
SLIDE 5

What is Deep Learning?

Input Value 1 Input Value 2 Input Value 3 Input Layer Hidden Layer Output Layer
slide-6
SLIDE 6

What is Deep Learning?

slide-7
SLIDE 7

The Neuron

slide-8
SLIDE 8

The Neuron

  • Neuron
  • Axon
  • Dendrite
slide-9
SLIDE 9

Neuron in an Artificial Neural Network

X1 X2 X3

Neuron Y

Output Value Input Value 1 Input Value 2 Input Value 3
slide-10
SLIDE 10

Neuron in an Artificial Neural Network

X1 X2 Xm

Neuron

Y1 Y2 Yp

Output Value 1 Output Value 2 Output Value p Input Value 1 Input Value 2 Input Value 3
slide-11
SLIDE 11

Artificial Neural Network

X1 X2 X3

Neuron Y

W1 W2 W3 Input Value 1 Input Value 2 Input Value 3 Output Value
slide-12
SLIDE 12

Deep Learning

slide-13
SLIDE 13

Lifecycle of an ML model

Historical Data Training Dataset Testing Dataset Algorithm Evaluation Model

Production Data Prediction
slide-14
SLIDE 14

Forward Propagation

slide-15
SLIDE 15

Back Propagation

slide-16
SLIDE 16

What are Convolutional Neural Networks?

Input Image Convolutional Neural Network Output Label (Image

Class)

slide-17
SLIDE 17

What are Convolutional Neural Networks?

Convolutional Neural Network Dog Convolutional Neural Network Cat

slide-18
SLIDE 18

How Computers See Images?

slide-19
SLIDE 19

The Big Picture of CNN

slide-20
SLIDE 20

Full Connection

slide-21
SLIDE 21

Overview of NVIDIA DIGITS

  • Deep Learning GPU Training System™ (DIGITS)
  • Wrapper for Caffe, Torch , and TensorFlow frameworks
  • Provides a wizard-style graphical web interface
  • Can rapidly train highly accurate neural networks
  • Highly customizable and configurables
slide-22
SLIDE 22

What are we Building?

  • A CNN model trained through NVIDIA DIGITS
  • Model trained with images of dogs and cats
  • Installing NVIDIA Docker runtime on a GPU-based VM
  • Running NVIDIA DIGITS as a Docker container
  • Downloading the trained model for inference
  • Using Caffe CPU Docker image for inference
slide-23
SLIDE 23

DEMO

https://github.com/janakiramm/digits-tutorial

Training a CNN Model with NVIDIA DIGITS

slide-24
SLIDE 24

MI2 Sponsors

slide-25
SLIDE 25

Next Webinar

Bu Buil ilding ing Conv nvolutio iona nal Neural Networks s wit ith NVID VIDIA IA DIGIT IGITS

NVIDIA DIGITS is a wizard-style framework to build convolutional neural networks. Attend this session to learn how to setup and configure NVIDIA DIGITS for building deep neural networks based on popular frameworks such as Caffe and TensorFlow. In this webinar, I will walk you through everything from data preparation to hyperparameter tuning to model serving.

Thursday, February 7th, 2019 9:00 AM PST / 10:30 PM IST

Register at http://mi2.live