Communications of the ACM
These Neural Networks Know What They're Doing
Massachusetts Institute of Technology researchers have shown that a special class of deep learning neural networks is able to learn the true cause-and-effect structure of a navigation task during training.
Massachusetts Institute of Technology (MIT) researchers have demonstrated that a specific neural network can learn the cause-and-effect structure of a navigation task it is taught.
The researchers observed that a Neural Circuit Policy (NCP) system assembled by liquid neural network cells can autonomously control a self-driving vehicle using just 19 control neurons.
They determined that when an NCP is being trained to complete a task, the network learns to interact with the environment and factor in interventions, or to recognize if an intervention is altering its output, and then it can relate cause and effect together.
Tests put NCPs through various simulations in which autonomous drones performed navigation tasks.
MIT's Ramin Hasani said, "Once the system learns what it is actually supposed to do, it can perform well in novel scenarios and environmental conditions it has never experienced."
From MIT News
View Full Article
Abstracts Copyright © 2021 SmithBucklin, Washington, DC, USA
No entries found