DARPA Developing Video Parser

coondoggie writes with an article in Networkworld about a disconcerting DARPA project. From the article: "If a picture is worth a thousand words, the scientists at the Defense Advanced Research Projects Agency would like to make that about a billion with a new software intelligent program. DARPA this month said it will detail a new system it would like to see built known as the Visual Media Reasoning (PDF) program. The main idea is to develop an advanced software program that can 'turn 'dumb' unstructured, ad hoc photos and video into true visual intelligence.'"

Comments

Use biological *informed* systems.

[BWJones]

Why not design a connectomics informed system that mimics the neural retina and visual system? Something that takes the results of research like this and uses true biologically informed computing to do what neural systems are good at and silicon based systems are not so good at? After all, what they are looking at is a system that works like a retina works (more like a video camera and not a still camera), so why not go to the biology which is really good at comparing like streams of information and making like or not like decisions.

More traditional background on retinal design and research can be found here.

Re:Use biological *informed* systems.

[BWJones]

The eyes do not "see" in the sense of processing information. They turn light into nerve impulses. Ho-Hum. We've got that, in fact this isn't about that at all. They are dealing with already captured data anyway.

Actually they do process information. The neural retina is like a miniature parallel supercomputer at the back of your eye that does initial signal processing from the photoreceptors through the over 50 kinds (200 in other invertebrates) of neurons.

Nothing in biology can be applied to this problem directly, only perhaps simple ideas applied rigorously. Stop spouting your favorite rubbish.

No offense friend, but I can't figure out if this is a troll or that you are simply uninformed here. Biological neural systems are *very* good at discriminating differences in data streams. Nested neural systems then further refine those abstractions and you get more advanced logic. The problem in the past has been discerning what those connectivities are as most current models of neural connectivity grossly underpredict the biological reality of neural circuit complexity.