Search
Sunday 26 January 2020
  • :
  • :

Why is DARPA Challenging Contractors to Research Neural Networks?

It has been said that a lot of technologies we take for granted have their roots in military research. For example, duct tape was originally a military product. The fact is that necessity is the mother of invention, and military necessity is responsible for one technological advancement after another. This should excite those in the tech industry hoping to push the boundaries of artificial intelligence.

According to a recent news report that emerged in mid-November 2019, the Defense Advanced Research Projects Agency (DARPA) is challenging contractors to put more research into neural networks. The agency is so high on neural networks that they are looking to fund the right research projects. DARPA wants to know if it is possible to run advanced neural network architectures on low-power systems.

You might be wondering why DARPA is so interested in the neural network concept. The answer should be easy enough to figure out once you understand what a neural network is and what it does.

Basics of Neural Networks

A neural network is just one component of the larger concept we call artificial intelligence. A neural network is a group of computer algorithms designed to loosely mimic human brain activity as they work together. How do they do that? By analyzing tons of data, recognizing patterns within that data, and comparing it to past data sets to reach some sort of conclusion – similar to how the brain makes decisions.

For example, a neural network could predict the likelihood of certain people developing a given medical condition. The network would be given past medical data as a starting point. It would then analyze data provided by the patients in question and look for patterns between that data and the historical data set. If the right patterns are recognized, the network would return results suggesting which of the patients are more likely to go on to develop the medical condition in question.

The one thing neural networks cannot do is think for themselves. Remember, these are just computer algorithms. They don’t know what information they are lacking. They don’t know how to go out and find information they do not yet have. They are incapable of any sort of thought – be it abstract or otherwise.

Smaller, Faster, and Better

Neural networks are already being utilized by the U.S. military to some degree. DARPA is interested in additional research because they aren’t happy with current capabilities. They want neural networks that are smaller, faster, and better. They want neural network technologies that can be made portable. They want networks that are more efficient, possess a higher level of logic, and can crunch greater volumes of data.

Rock West Solutions, a California company that works with commercial and military contractors on signal processing solutions, explains that the fundamental technologies necessary to give DARPA what it wants already exists. Now it is a matter of improving those technologies substantially.

Much of what will go into that effort involves signal processing. From Rock West’s point of view, all the data a neural network could process is not necessarily good data. One of the first requirements of improving a neural network is to apply advanced signal processing to weed out undesirable data so that it doesn’t get in the way.

At the end of the day, DARPA’s desire to further expand its use of neural networks is all about smarter self-defense. Rather than relying merely on raw military power to protect us from our enemies, DARPA wants to outsmart them as well. That would theoretically lead to better defense at a lower human cost.