Anda di halaman 1dari 32

A) DLI TEACHING KITS

Deep Learning Institute


NVIDIA Corporation
GEP Program Update, Singapore.
18 July 2017
1
DEEP LEARNING INSTITUTE
Overview

DLI Mission: Helping the world to solve the most


challenging problems using AI and deep learning.
We help developers, data scientists and engineers
to get started in training, optimizing, and
deploying neural networks to solve real-world
problems in diverse disciplines such as self-
driving cars, healthcare, consumer services and
robotics.
For Everyone For Developers, Data Scientists, Researchers

INTRO MATERIALS ON-SITE WORKSHOPS SELF-PACED LABS

CASE STUDIES PARTNER COURSES TECHNICAL BLOGS


DLI Teaching Kit
Available free to educators!
developer.nvidia.com/teaching-kits

4
DLI
DLI TEACHING
TEACHING KIT
KIT
Available free nowGoals
Module for educators!
‒ Based on Yann LeCun’s Deep Learning course at
Teach
NYU academic theory and application of DL harnessing the PyTorch
and Torch frameworks
‒ Comprehensive teaching materials
‒ Lecture slides
Technical
‒ Hands-onsubjects:
labs/solutions
‒ Quiz/exam problem sets/solutions

‒ GPU resources Intro to ML/DL Unsupervised Learning


Applied Image Classification Generative Adversarial Networks
‒ Free NVIDIA DLI Qwiklab tokens for students
Applied Object Detection Optimization Techniques Early adopters include:
‒ Up to $125 of Amazon Web Services
Convolutional NNs (AWS) Recurrent NNs
GPU instances for students
Applied per term
Image Segmentation 15K students
Natural to be Processing
Language
trained in 2017
Energy-based Learning And more!
https://developer.nvidia.com/teaching-kits

5
6
CLOSING THE EXPERIENCE GAP
Supporting expertise

Instructor workshops, conferences, sponsorships and exhibits


Enablement web pages
Getting started guides/videos
Email updates
Feedback and enhancement requests
B) ACCELERATED COMPUTING (TEACHING KIT)
Breaking the Barriers to GPU Education in Academia
Co-developed by UIUC and NVIDIA for
qualified educators

Comprehensive teaching materials


3rd Ed. PMPP E-book by Hwu/Kirk
Lecture slides and notes
Lecture videos
Hands-on labs/solutions
Larger coding projects/solutions
Quiz/exam questions/solution

GPU compute resources


NVIDIA online free Qwiklab credits
AWS credits

developer.nvidia.com/teaching-kits
7
CUDA Parallel Related
Programming Computation Case Studies Programming
Model Patterns Models

CUDA Memory Advanced MRI


Histogram MPI
Reconstruction

Data Electrostatic
Management CUDA Python
Stencil Potential
using Numba
Calculations

CUDA
Parallelism
Model Reduction Deep Learning OpenCL

Dynamic
Parallelism Scan OpenACC

CUDA
Libraries Sparse Matrix OpenGL

Unified
Merge Sort
Memory

Graph Search

8
C) ‘Jet’ – A Smart,
Autonomous, Low-cost
AI Robot Kit

developer.nvidia.com/jet-robot
9
AGENDA

Introduction to ‘Jet’ and Jet Toolkit


Robotics Teaching Kit with ‘Jet’ Overview
‘Jet’ Teaching Kit Contents

10
INTRODUCTION TO ‘JET’

Smart, autonomous, low-cost AI robot based on


NVIDIA Jetson

Compute-intense computer vision, AI, and self-


driving capabilities

Utilizes Robot Operating System (ROS), scales from


K-12 projects to real industrial applications

Available globally for sale as full kit via partner


Servocity

Software includes ‘Jet’ Toolkit and Robotics


Teaching Kit with ‘Jet’ for educators

11
JETSON TX1

12
JETSON TK1

Tegra K1 Development Kit

Runs Ubuntu Linux

CUDA

OpenCV optimized for Tegra

13
JET OVERVIEW
Chassis

Jet chassis is constructed using Actobotics parts


These are machined metal parts with various aluminum channel, hubs, wheels, and
brackets.

14
JET OVERVIEW
Chassis

15
JET OVERVIEW
Electronics

Jet electronics consists of:


NVIDIA Jetson TX1/TK1

Arduino Mega

H-bridge and motors

3 sonar sensors

GY-521 accelerometer/gyroscope

Single Logitech Webcam

3S (11.1V) 5000mAh LiPo battery

16
JET OVERVIEW
Jet Toolkit Software

Build Instructions

Example Code and Educational Materials

- Introduction to Robotics, 'Jet', and Robot Operating System (ROS)


- Sensors and Actuators
- Obstacle detection and avoidance using sonar
- Image processing and edge detection using blurs, Canny Edge Detection, and
Hough Lines Transform
- GPU-accelerated HAAR Cascade Classifiers for facial detection

developer.nvidia.com/jet-robot
17
JET OVERVIEW
Architecture Design

Encoder Jetson TX1/TK1 Camera


readings USB

Left USB
Motor
H-Bridge Arduino
Shield Mega
Right
Motor I2C

Sonar Sonar Sonar Accel/Gyro (GY-521)


Module Module Module

18
Inserting video: Insert/Video/Video from File.
Insert video by browsing your directory and selecting OK.

File type that works best in PowerPoint is: .wmv

VIDEO FILE
19
Inserting video: Insert/Video/Video from File.
Insert video by browsing your directory and selecting OK.

File type that works best in PowerPoint is: .wmv

VIDEO FILE
20
PURCHASE ‘JET’ VIA SERVOCITY
Full Bundle Now Shipping Globally

TX1 ‘Jet’ Robot Kit


USD$756.79 Academic Price
USD$945.99 Regular Price

TK1 ‘Jet’ Robot Kit


USD$620.79 Academic Price
USD$775.99 Regular Price

https://www.servocity.com/tx1-jet-robot-kit
https://www.servocity.com/tk1-jet-robot-kit

21
Robotics Teaching
Kit with ‘Jet’
Available to Instructors Now!
developer.nvidia.com/teaching-kits
22
ROBOTICS TEACHING KIT WITH ‘JET’
Available Now Free for Educators!

‒ Co-developed with CalPoly University

‒ Comprehensive teaching materials


‒ Lecture slides
‒ Hands-on labs/solutions
‒ Quiz/exam problem sets/solutions
‒ Open-ended coding projects
‒ Free Qwiklabs and AWS access

‒ Robot hardware kit via Servocity includes:


‒ NVIDIA Jetson TX or TK1
‒ All mechanical, chassis, electronics
‒ https://www.servocity.com/tx1-jet-robot-kit
‒ https://www.servocity.com/tk1-jet-robot-kit

developer.nvidia.com/teaching-kits
23
ROBOTICS TEACHING KIT WITH ‘JET’
Module Goals

Teach multidisciplinary, GPU-


accelerated, autonomous Robotics
Technical subjects
Robot Operating System
Sensors
Computer Vision
Machine Learning
Dead Reckoning
Path Planning
Localization
Control
Obstacle Avoidance

24
TEACHING KIT MODULES
Robotics with ‘Jet’

• Course Introduction and Overview


Module 1 • Introduction to Robotics
• Introduction to Jetson TK1/TX1
Course Introduction • ‘Jet’ Overview
• Introduction to ROS

• Sonar
Module 2 • Accelerometer, Gyroscope
Sensors and Actuators • Camera
• Motors, Encoders

• ​Introduction to Computer Vision


• Image Representation
Module 3
• Edge Detectors
Computer Vision
• Hough Transform
• Image Filtering and Moments

• ​Machine Learning with Neural Networks


Module 4 • ​Neural Networks Models
Machine Learning • cuDNN
• Training and Usage using Caffe

25
TEACHING KIT MODULES
Robotics with ‘Jet’

Module 5 • Dead Reckoning


• Odometry Model for Differential Drive
Dead Reckoning • Sensor Fusion and Kalman Filters

Module 6 • Path Planning


Path Planning • A* Path Planning

Module 7 • Control Systems


Control • PID Control

• Robot Localization
Module 8
• Monte Carlo Localization
Robot Localization
• Particle Filters

Module 9 • Mapping
Mapping • SLAM
• ​Motivation
Module 10 • Harvester
Final Projects • ​Robot Capture the Flag Game
• Color Follower

26
TEACHING KIT CONTENTS
Lecture Slides

29 total slide decks from 10


modules .pptx format

27
TEACHING KIT CONTENTS
Hands-on labs/solutions

13 total labs from 10 modules

1-2 week assignments

Includes description, objectives, prerequisites


and open-ended questions

Includes Pseudo-code and solution code


templates

Latest source code and instructions always on


BitBucket

.docx and .pdf formats

28
TEACHING KIT CONTENTS
Questions/answers sets

Multiple choice, including rationale for


answers
Students should be able to answer from
lecture content
9 total questions/answers sets from 10
modules
.docx and .pdf formats

29
TEACHING KIT CONTENTS
Larger coding projects/solutions

3-4 week, open-ended, multidisciplinary,


final semester projects
Not tied to specific modules
3 total projects specifications
.docx and .pdf formats
Sample solutions in source code coming
in a future release!

30
OTHER RESOURCES
Deep Learning Institute (DLI) Online Labs

Live, hands-on, self-paced learning


environment to in practical, applied deep
learning and GPU accelerated computing

Labs includes interactive instructions, coding


and Q/A

Hosted in the cloud

Students only needs a web-browser and


internet access

Labs are timed

Free tokens with Teaching Kit

31
Q&A
developer.nvidia.com/jet-robot

32

Anda mungkin juga menyukai