Communications of the ACM
Brain Scan For Artificial Intelligence Shows How Software Thinks
Researchers at the Israel Institute of Technology have developed a way to take a snapshot of a neural network as it works through a problem.
Credit: Tom Zahavy, Nir Ben Zrihem, Shie Mannor
Neural networks are so complex it can be impossible to retrace the steps a deep-learning algorithm took to reach a given result. Israel Institute of Technology researchers have developed a new technique for taking snapshots of neural networks as they work through a problem.
The technique is like a functional magnetic resonance imaging (MRI) scan for computers, capturing an algorithm's activity as it analyzes a problem. The image enables the researchers to track different stages of the neural network's progress, including dead ends.
The researchers gave a neural network the task of playing three Atari 2600 video games: Breakout, Seaquest, and Pac-Man. They then collected 120,000 snapshots of the deep-learning algorithm as it played each of the games, and mapped the data using a method that enabled them to compare the same moment in repeated attempts at a game. The researchers say scans such as these could help others develop algorithms designed to solve real-world problems.
"If you’re deploying this technology in the real world, you want to understand how it works and where it might fail," says University of Wyoming professor Jeff Clune, who was not involved in the research. "If we can understand neural networks better, then we can understand their weaknesses and improve their strengths."
From New Scientist
View Full Article - May Require Paid Subscription
Abstracts Copyright © 2016 Information Inc., Bethesda, Maryland, USA
No entries found