Northwestern Computational Research Day keynote april 19, 2016 (1)
-
Upload
iqbal-arshad -
Category
Technology
-
view
667 -
download
0
Transcript of Northwestern Computational Research Day keynote april 19, 2016 (1)
Deep Neural NetworksDistributed IntelligenceUbiquitous Connectivity
Infinite Cloud
Architecting Artificial Intelligence Experiences
Iqbal ArshadSVP Engineering, Global Product DevelopmentMotorola Mobility
Key Computing Trends
Internet Mobile Ai: Starting Now
1994: Netscape Browser
1997: Google Search
2007: iPhone with Safari Browser
2008: Motorola Droid with Google Maps
+
+
Enabling New Experiences
Vision – A World of Holistic Ai Experiences
Network
Data + Machine Learning
Analytics + Apps
Smartphone
Smartwatch
Tablet
Smart IoT Devices
Experiences
Smart Environments
Knowledge Extraction
System Architecture
Prediction+
InferencePredictions,Alert, Advice
Feature ExtractionDescriptors, Text, Metadata
Sensor + Observational Data
DNN
Key Technical Enablers
Cloud Computing Mobile Computing Wireless Connectivity
GPU no longer just for graphics
Explosion in Cloud Capacity
Ubiquitous 4G LTE
Faster & FasterCPU & GPU Ahead of Traditional Needs
Sensor Laden Devices
2015 Lenovo Confidential. All rights reserved.
Distributed IntelligenceGeneral-Purpose GPUs (GPGPUs) are helping fulfill the law of accelerating returns
The path to growth no longer depends on packing more transistors onto a die,but packing more GPU cores into warehouses.
Key Technical Enablers
Open Source Ai tools exist ... but data is key
CNTK GPU Developer Kit
Getting the Data: Smart Phones + IoT in 2020
World Population: 7.7 billion
Smart Phones: 6 billion
IOT Devices: 20+ billion
Smartphones2 billion9% CAGR
Computers749 million 3% CAGR
Smart Watches91 million 80% CAGR
IoT Devices2 billion 21% CAGR
Market Disruption: World of DNN Products
Smart Car IoT DevicesSmart Phone
Market Disruption: World of DNN Products
Smart Car
Autonomous Cars: Disrupting Transportation
Cheaper Public Transport55% of taxi revenue is paid to drivers
Relax and let the car do the driving
Everybody Can Own a Car
Safer Driving94% of accidents involve human error
Better Commute ProductivityUS average daily commute is 90 min
Safer for Pedestrians150,000 pedestrian accidents per
year
More Efficient ParkingSelf parking cars need 15% less space
Connected Autonomous Cars
Central Control Unit (DNN Compute, Car Control)
Sensor System (GPS, Radar, Lidar)
OEM Cloud(training, external data management)
OnboardDNN
360 Degree Vision System
Crowd Sourced DNN(mapping, driver data)
Other Clouds Weather Police City
Smart Car: Autonomous Systems●Forward facing radar and
camera vision system
●Ultrasonic sensors for 360º object detection
●Onboard vision computing algorithms to measure distance, read signs and detect pedestrians
●DNN system trained by millions of miles of crowd sourced driving videos & data
●High precision maps gathered and enhanced by all drivers
●Designed for riding, not driving
●Rotating roof top camera with 64 laser beams to create 3D images of surrounding objects
●Smart GPS sensory system calibrated by sensor map data captured by cars on previous journeys
●Onboard computer vision algorithms use windshield camera to help car make safe and intelligent decisions on busy roads
●Ultrasonic sensors alert car to obstacles in rear and enable backup systems
Smart Car: Self Driving Car
Market Disruption: World of DNN Products
Smart Phone
Intelligent Smartphone Platform
● Contextual notifications delivered to your phone based on your location, calendar, search history, email and sensor information
● Personalized voice trigger technology that enables you to wake-up and control your phone even when it’s locked and turned off
● Smart gestures that enables you to control and interact with your phone using sensor input
● Smart sensors that track your biometric, physical activity and sleep
Beam FormedMulti Mic
Noise Reduction
Intelligent Smart Phones: Natural Language
DSPWake-UpTrigger
NeuralNet
GoogleNowApp
Noise Reduction
SpeechDetection
SemanticAnalysis
IntentClassifier
EvidenceRanker
GenerateResponse
NeuralNet
Market Disruption: World of DNN Products
IoT Devices
Smart IoT Devices: Autonomous Drones●Real-time object detection,
tracking and smart video-cast algorithms
●3D reconstruction of scenes to create a model of the environment
●Smart autopilot modes to enable safe navigation of specified flight paths
●Airport avoidance, and other cloud controlled safety measures to enable new use-cases
● Indoor and outdoor flight control
Market Disruption: Not Just Devices…
Building Ecosystems is the key
A New World: Transformed by Ai
Smart Transport Smart City Smart Enterprise
Smart Transportation ● Shared Transportation via enhanced car
services
● New Business Models for OEM & Service Provider
● Mobility Services in autonomous vehicles
● Dynamic Traffic Control Systems will monitor queue depths at intersections and toll gates adapting to demand to maximize flow
● Lower Cost per mile
Smart Cities●Utilities, police, fire and city will share
a common data set that powers a city wide DNN
●Crowdsourced mobile phone data recruited to report public safety / health information
●Drones gather thermal images to detect poorly insulated and drafty homes / offices
●Real-time gas & chemical detection by mobiles on the street, alerting authorities
●Smart autopilot modes enable safe navigation of specified flight paths in crowded airspace
Smart Enterprises●Nvidia, Google and IBM are building HPC
Hyperscale platforms to enable enterprises to leverage DNNs
● To remain competitive, every enterprise need to understand how they will use Ai to improve consumer experience.
●Companies will adjust pricing based on
supply chain and wide range of market analytics
●Clinicians will gain new insights through applications that scour vast amount of health data
Deep Neural NetworksDistributed IntelligenceUbiquitous Connectivity
Infinite Cloud
To survive, every company must understand how they will apply artificial intelligence technology to their business.
Soon our home, car, smart devices and city will work together as a single intelligent unit, sharing data and enabling new user experiences.
Product engineers of the future must consider device, cloud, data sources and the application of machine learning techniques to every product that they design.
Summary