Sign In

Communications of the ACM

ACM TechNews

NVIDIA Lets You Peer Inside the Black Box of Its Self-Driving AI


The neural network architecture is designed to highlight areas that contribute most strongly to the behavior of the car’s deep neural network.

Nvidia's neural network software highlights (in green) the areas on which it focuses as it makes driving decisions.

Credit: Nvidia

NVIDIA researchers have developed a neural network for autonomous driving that emphasizes what areas it is focusing on in a video image in an attempt to instill more accountability into artificial intelligence.

During testing, the researchers found the network tends to focus on road edges, lane markings, and parked cars as human drivers would.

"What's revolutionary about this is that we never directly told the network to care about these things," says NVIDIA's Urs Muller.

He notes this is a good starting point for understanding the neural network's reasoning mechanisms. "I can't explain everything I need the car to do, but I can show it, and now it can show me what it learned," Muller says.

He thinks this type of strategy could become increasingly relevant as deep learning is applied to virtually any challenge that involves large quantities of data, such as those in medicine, finance, and military intelligence.

From Technology Review
View Full Article

 

Abstracts Copyright © 2017 Information Inc., Bethesda, Maryland, USA


 

No entries found