Unusual semantic applications of CUDA Options

[DigitalDirect]

We are interested in exploring the use of CUDA technology, in the DirectCompute environment ultimately, for two kinds of specialized processing, both essentially in the area of extracting meaning from ordinary narrative text (e.g., Wikipedia, NY Times movie reviews) and photographs/movies.

1. Port typical pattern-recognition-based semantic text extraction technology (www.semanticv.com) to the GPU
2. Extract meaning from pictures, including facial recognition (identification), scene categorization, etc.

Please reply directly to Nicholas.Bedworth@digitaldirect.com

We are in GMT -10.

[Prautes]

We are interested in exploring the use of CUDA technology, in the DirectCompute environment ultimately, for two kinds of specialized processing, both essentially in the area of extracting meaning from ordinary narrative text (e.g., Wikipedia, NY Times movie reviews) and photographs/movies.

1. Port typical pattern-recognition-based semantic text extraction technology (www.semanticv.com) to the GPU
2. Extract meaning from pictures, including facial recognition (identification), scene categorization, etc.

Please reply directly to Nicholas.Bedworth@digitaldirect.com

We are in GMT -10.

Very interesting, Machine vision and natural language understanding.
It's been a long time since I've heard of anyone exploring such ideas.
Probably since the cold war era.

I remember a Small Business Innovative Research briefing given many years ago at the Roma Air Development Center, RADC.
The system being developed was a combined AI and machine vision system. The business plan
was to develop the capability to turn Niagara Mohawk's (a large electric utility) massive paper data base
into something useful. The work included machine reading of engineering diagrams and making joint inferences
from data extracted from both text and graphical sources. I was struck by the estimates of the value
that such a capability would be to Niagara Mohawk.

There was a RFP from an intelligence branch of the Air Force that wanted to be able to semi-automatically
interpret/distill massive amounts of classified and non-classified mixed text, graphics and images. This of course would also involve
a combination of natural language inference, machine "vision", and contextual inference to resolve (or quantify) ambiguities.

With the end of the cold war and the resulting AI winter most of this sort of research came to an end. I always figured that
the futures traders and the like had continued similar developments but kept quiet about them.

Of course the raw processing power we have now is much greater and you would thing that the potential
for discovery would be immense. On the other hand no one works with
hardware and software like connection machines front-ended by a Symbolics anymore.

Galadriel:
"The world is changed. I feel it in the water. I feel it in the earth. I smell it in the air.
Much that once was is lost, for none now live who remember it."

I have also been giving some thought to the use of GPU processors for parallel inference.
(Their use in the image processing part of machine vision is fairly obvious.) It would be an interesting capability.
Once can imagine combining advanced scientific models and heuristic search techniques to search for
new materials, etc. But, there seems to be sort of an systemic mental block regarding such things at the current time.

[mtk]

Probably since the cold war era.

Cold war era??.... mmm...
Extremely famous consumer applications
Google, Google Translator, Powerset? DARPA Grand Challenge?
OpenCV? Midomi? Shazam ?

I think the field has been very active since cold war...