Machine Learning to Recognize Egyptian Hieroglyphs - WebGL and Deep Belief Networks
What needs to be done?
A server-side, machine learning algorithm needs to be built that will continuously retrain and learn from a steadily increasing set of drawn hieroglyphs that are mapped to the correct matching Unicode glyph.
What are the design constraints?
--The system can suggest more than one matching glyph. We are looking for the capability to narrow a single drawn glyph down to a selection of potential matching glyphs. There are 7,000 Egyptian glyphs and anything that can make the set of possible glyphs smaller will improve the speed of finding the correct matching glyph.
--We are open to ideas for the design approach. Currently our bias is towards implementing a Deep Belief Network based training system. We are open hearing other approaches if you have evidence that they will be successful.
What are resources to better understand the project?
There are over 7,000 Egyptian hieroglyphs. We will have image files for all of them - and an increasing number of image files of people's tracings of them.
If you wish to see the set of hieroglyphs, you can view them on this page by clicking the icons: http://www.ancientwords.net/full.h
What skills are needed?
WebGL (a subset of OpenGL ES 2.0)
Machine Learning via Deep Belief Networks
What would we like to hear from you?
Help us understand the approach you suggest (if it is different than we have expressed here)
Direct us to any evidence that supports your recommendation (if different)
Provide a rough estimate in hours for the project.