HiMax: Characterization of the CogniMem Device
EE x96 Preliminary Design Review
Advisor: Tep Dobry
Sub Advisor: Neil Scott
Members: Raymundo Flores EE 296Darnel Balais EE 496
Presentation Overview:
Team and Member Introduction Brief Project Overview and Background Approach Project Block Diagram Controlled Environment Diagram CogniMem Device Diagram Problems End of Semester Project Goals Project Timeline References
Team and Member Introduction:
HiMax Group: Communication and Information Sciences,
UH Manoa– Develop and research applications using the Cognimem Chip.
Department of Electrical Engineering - UH Manoa– Develop and research baseline characteristic of the CogniMem
Neural Processor.– Create hardware and develop software
for baseline characterization.
Darnel Balais: Software ProgrammerRaymundo Flores: Hardware
Project Overview:
This technology is fairly new, so we propose to develop:
1. Procedures to accurately train the Neural Processor,2. Baselines for "high-level confidence" for pattern
recognition for a given physical environment.3. (For follow-on X96 Project), an application that uses
the CogniMem Neural processor, with a "built-in" camera, as an autonomous pointer that identifies any painting / object in a museum setting. The camera, then can be used to communicate to an ultra portable PC via broadband that gathers information about the identified painting.
Project Background:
CogniMem 1K Specs- Patented parallel
architecture- 1024 Parallel neurons- Unlimited neural network
expansion- Trained by example- Low power consumption
Image Recognition Board
www.recognetics.com
Approach:
Fabricate a mini-museum setting that includes a picture inside a frame that enables us to test the CogniMem processor/camera pattern recognition (imaging) capabilities.
Develop a program that trains the CogniMem camera to identify and view images from any angle, orientation and distance.
Project Block Diagram:
Controlled Environment
CogniMem Device
Controlled EnvironmentDiagram:
Computer Generated
Human Generated
CogniMem Device Diagram:
Problems:
Hardware and software has only been recently developed and product knowledge is still limited at this time.
– Help can easily be attained from Research Assistants within the HiMax Group.
– Weekly meetings
Hardware is very expensive and reacquiring a new chip and camera may take a long time.
– Take special care of equipment.– Backup Device
End of Semester Project Goals:
Our projected goals are: To develop "high-confidence level"
characterization of the CogniMem processor with respect to distance, orientation, and image complexity.
Have a working device for a museum application.
Project Timeline
Presentation Summary:
Key Technology: CogniMem Neural Processor (1024 parallel neurons)
Method of Modeling: Learn and build knowledge by example vectors (i.e. camera,
microphone). No cumbersome programming required. Challenges: Fairly new technology Minimal data/information available Lay "ground" work for future applications
Main System Diagrams: Controlled Environment CogniMem Device
References:
1. http://wiki.roadnarrows.com - Distributer2. http://www.general-vision.com - Researchers3. http://www.recognetics.com - Developers
?Any Questions?