Perceptual Autopilot for Distance Education Classrooms

1
Perceptual AutopilotforDistance Education
Classrooms

• -Project Description-

2
Distance Education Classrooms

• Remotely located classrooms connected via a
  broadband network (VPnet)
• Lecture delivered to all classroom on real-time
  basis video image and sound
• One instructor at a location
• AGMODE AccessGrid Middleware Optimized for
  Distance Education manages transmission of video
  and sound

3
(No Transcript)
4
AGMODE System Architecture

• AV devices to handle video and sound
• AccessGrid System Middleware
• AccessGrid middleware for scientific
  collaboration, motivated by vision of ideal
  virtual meetings among collaborators
• VPnet Fiber-optic broadband network
  (400Mbit/second)
• Requirement easy to use, low-cost operations
  autopilot for AV devices

5
(No Transcript)
6
PADE
7
PADE Autopilot

• Multi agent, blackboard architecture
• Real-time signal processing video and audio
  input streams
• Tokenized output compliant to XML
• Soft Computing a consortium of uncertainty
  management technologies (probabilistic reasoning,
  dempster-shafer, fuzzy logic, and more)

8
PADE System Architecture

• Blackboard information concerning states of
  class activities and signals captured from AV
  devices (signal-content-based histograms)
• Sensing agents generate histograms
• State agents recognized states of activities
• Control agents generate outputs in XML notation
  to be passed onto a lower layer
• House Keeping Agent garbage collection

9
(No Transcript)
10
Illustration of Process

• Sensory Information (sensing agents)
• Visual histogram on x-axis y-axis
• Audio Signal spectrum
• Determine the state (state agents)
• Determine actions (control agents)
• Housekeeping eliminating obsolete information
  from the blackboard (housekeeping agents)

11
Challenges

• Real-time, Adaptive intelligent agents
• Signal and image processing based on histograms
• Identifying states based on linguistic
  expressions with uncertainties
• Generating configuration of controls based on
  current states with uncertainties
• Reasoning linguistic expressions using soft
  computing methods

12
Specification (to be developed)

• Objects to be recognized
• Instructor
• Students
• Things (not person)
• States and transitions among states
• In-session
• Ask-question/draw instructors attention
• In-lecture/presentation
• Sensory information
• Audio
• Visual
• Others (e.g., position)
• Devices (minimum requirements)
• PZ-CCD (vision of entire classroom, for focus)
• Microphones
• Position sensor (small device on the instructor)

13
Project Outcomes

• Background research
• Scope of project sub-problem
• Feasibility of fundamental approach
• A simple computational simulation

14
Requirements

• Agent Architecture
• Embedded Artificial Intelligence (FRIL)
• Uncertainty Management
• Support Logic Belief Networks, DS
• Fuzzy Logic
• Evidential Logic Neural Networks
• Real-time Response
• Low-cost Computation

15
Tools

• FRIL
• Target Language to embed FRIL
• C
• C
• JAVA
• Simulation Tools MATLAB, Octave, R
• Data Set UCI KDD/ML Repository

16
Development Environment

• PC UNIX is ideal
• LINUX
• Free BSD
• Windows
• OK, but I cannot offer much help.