Download - Research and Innovation
Research and InnovationResearch and Innovation
Machine Intelligence: Machine Intelligence: An Investigation in the An Investigation in the
Application of Application of Hierarchical Temporal Hierarchical Temporal
MemoryMemory
L. Salemi, Professor Centre for Construction
and Engineering Technologies
May 2009
IntroductionIntroduction
Project was approved in Oct. 2008• Seed Funding - $6,250• OCE Connections Grant - $3,000
Student participation from 3 programs• T146 Electro-Mechanical Engineering Technician• T147 Computer Systems Technology • T121 Mechanical Engineering Technology - Design
Introduction – The TeamIntroduction – The Team
Research Assistants Section Status• Clayton Wozney T121 Paid Researcher• Steven IrwinT146 Course Credit• Michael Joyette T146 Course Credit• Olek Kushnarenko T146 Volunteer• Scott Vannan T146 Volunteer• Terence D'Cunha T147 Course Credit• Avinash Singh T147 Course Credit• Intiaz Abdulla T147 Course Credit• Volunteers - Albert So, Bruno D’Agostino
Introduction – Industry InvolvementIntroduction – Industry Involvement
Company Status
• Industrial Technical Services In Kind Sponsor – Reynold Ramdial– Amit Setti Technical Support
• Grace Instrumentation & Controls Equipment Donation – Terry Grace
• Hoskin Scientific– Marc DeGrace Technical Advisor
• Hatch Engineering Technical Advisor– Dennis Phair Equipment Donation
• ISA Toronto SectionPresented at the ISA Technical– Currie Gardner Conference during the Ontario Process and
Automation ShowApril 2009
Introduction – The PlanIntroduction – The PlanPhase 1: Oct – Dec 08Phase 1: Oct – Dec 08
C. WozneyPaid Researcher
InvestigateHTM
Technology
Students to Work for
Course credit
Phase 2: Jan – May 08Phase 2: Jan – May 08
Create WorkspaceRm. C504A
6 - Student Volunteers (T146)
Plan B
Wozney to managePhase 2
Build the Infrastructure
Collect Data
Simulate Remote
Site Control
ObjectiveObjective
Apply Intelligence to Building Automation Applications
Use one of the classrooms to collect data• HVAC (Heating, Ventilation, and Air Conditioning)• Lighting (Occupancy based)• Security (Access control, intrusion)• Security Cameras
Incorporate intelligence to • Turn off the lights when no one is in the room• Lower the temperature• Monitor room occupancy
Research QuestionResearch Question
How can we make a machine intelligent?How can we make a machine intelligent?
But first, what is intelligence?• Human Intelligence• Machine Intelligence• Artificial Intelligence• Military Intelligence
There is no universal definition
Research – The ChallengeResearch – The Challenge
Intelligence Test: Which one is flat?Intelligence Test: Which one is flat?
Research QuestionResearch Question
AnswerAnswer: All of the above are flat: All of the above are flat
Does intelligence lie in the senses of the beholder? Yes/No?
• Our 5 primary sensors provide an abundance of data• Our intelligence forms the conclusion (BELIEF)• Where is this “intelligence” located and how can we make
a machine do it?
Research – The AnswerResearch – The Answer
Hierarchal Temporal Memory (HTM)
• Developed by Jeff Hawkins founder of Numenta and inventor of palm pilot & treo
• HTM is modeled after the neocortex• Data is fed to neuron-like networks that
learn to recognize patterns and sequences that change over time
• When presented with “new” data the HTM is good at predicting what it is
• www.numenta.com Book: On Intelligence
Research – The TechnologyResearch – The Technology
• NuPIC (Numenta Platform for Intelligent Computing)• Vision4 Demo program was designed using HTM networks
What’s this?
Research – The TechnologyResearch – The Technology
Vision4 Demo program was trained to recognize 4 different images
• Sail Boat
• Rubber Duck
• Cell Phone
• Cow
Research – The TechnologyResearch – The Technology
Its not perfect but neither are we.
Sailboat ???
Research – The TechnologyResearch – The Technology
• More detail provides better recognitionIt’s a duck
Research – The TechnologyResearch – The Technology• HTM is capable of recognizing several variations
Cow in the background
How far away are we?How far away are we?
• Not a question of if, but when!
• Next 400 years?• Only 400 years have
passed since we thought the earth was flat.
"I visualize a time when we will be to robots what dogs are to humans. And I'm rooting for the machines." - Claude Shannon (1916 - 2001)
Industry Problem Industry Problem
Phase 2 - How to apply HTM intelligence to Building Automation applications
Identifying an industry problem was difficult• Many “smart” systems already out there• HTM was beyond our scope – now what?• HTM would be hard to sell to industry partners
without something to demo
Leo’s ProblemLeo’s Problem
• Engage students and comply with course outlines (course credit for research work)
• Build something that we could demonstrate to attract industry partners
• No Clayton – No HTM• Go with Plan B
Methodology – Plan BMethodology – Plan BPlan B – Make sure Plan A worksPlan B – Make sure Plan A works
Build the infrastructure to collect real time data in room C504A Build the infrastructure to collect real time data in room C504A
• Simulate something that is used in industry (remote water Simulate something that is used in industry (remote water pumping station)pumping station)
• Be able to monitor and control the site remotely via the webBe able to monitor and control the site remotely via the web• Use current technologies plus add some extra’sUse current technologies plus add some extra’s
– SCADA (Supervisory Control And Data Acquisition)– Security Alarm and Video Surveillance– Process Cameras for operators– Full network integration for each subsystem
• Incorporate intelligence between all of the subsystems
ResultsResults
Remote monitoring & control of pumping station– Operator has full control of station (typical)– Process cameras allow operators to view the station
as if they were present– Surveillance Alarm & Cameras connected via the
web and VOIP system (24/7 monitoring station)– SCADA system used to the control process– More features to be added
Infrastructure Testing LabInfrastructure Testing Lab
REMOTE SITE
SCADAControl System
SecuritySystem
Surveillance Cameras
VOIP
Process Cameras
Operator Terminals
Pumping Station
Lessons LearnedLessons Learned
Benefits gained• Excellent learning experience for students
and professor • Infrastructure Testing Lab – a place for
us to work and others to utilize• Opportunity to learn new technologies and
add to curriculum• Interdivisional co-operation • Industry Partners
Lesson LearnedLesson Learned
Bumps along the way• Hard to convince some of the course
coordinators to let students do this for a course credit (T147 was the exception)
• Uncertainty of the use of room 504A makes it difficult to plan future projects
Future ResearchFuture Research
• Full integration of sub-systems using an OPC data manager
• Train the HTM using the remote site data• Work with Video Analytics• Design and build sensors that are HTM-ready• Attracted an industry sponsor who is
interested in using solar power in a remote site application
QuestionsQuestionsThank you and AcknowledgementsThank you and Acknowledgements
Meadow Larkins and the ARI team
The student research team
Reynold Ramdial and Amit Setti from Industrial Technical Services
Members of the technical advisory committee
Jeff Litwin for supporting our efforts
ISA Toronto for allowing us to present at their technical conference in April