Can Pattern Recognition and Machine Learning Be Added to Mobile Devices?

Can Pattern Recognition and Machine Learning Be Added to Mobile Devices?

General Vision makes and promotes a “neural network chip” for complex image recognition systems that are capable of object recognition, surface analysis, pattern matching, anomaly detection, target tracking and more.

A few years ago, the company made something of a splash in the artificial intelligence and machine learning industries for its role in a high-speed fish inspection system that was deployed on several vessels in the waters around Norway and Iceland.

This inspection system was used to monitor at-sea fish processing, including sorting fish by species, watching for damaged fish, or even monitoring a fish's positional orientation relative to automated cutting or filleting equipment.

Although I have not been able to find any objective evidence supporting the anecdote I am about to share, I was once told that this fish-monitoring system had even detected seasonal differences in Herring scales associated with relative water temperature.

I heard this story while attending a trade show in Japan. Apparently, the fish inspection system mentioned above had been in vessels for a few of months happily classifying something like 600 fish per minute. Then all of a sudden, the CogniSight system, as General Vision calls its solution, didn't know a herring from a ham.

Something had gone wrong. It seems that as water temperatures had dropped there had been a minute change in the fish scales — perhaps an increase oil content, the engineer who told me the story explained. The General Vision solution had been so discriminating that it detected something about fish that was previously unknown to biologist they get winter coats.

While I will again mention that I am not vouching for the accuracy of this account or any associated scientific claims, I was reminded of it this week when I found a colleague using Google Goggles to take a picture of an image on a computer screen. Google Goggles is an Android application that can do a search based on input from the mobile device's CMOS image sensor. The software can successfully identify many name-brand products, bar codes, or even famous landmarks or works of art. My colleague was trying to use it to identify a celebrity from a photograph.

Although Goggles was unsuccessful, it made me wonder if, in fact, advanced pattern recognition — like what General Vision's solution was doing — might have its place in mobile devices. Would a mobile phone that could identify your child from across a crowded playground, have a competitive advantage? What about a phone that could be taught to recognize your face for security purposes? How would you use this sort of feature in a mobile device?

Armando Roggio