Intel Labs has introduced a new approach to neural network-based object learning, enhancing the value of neuromorphic computing.

Intel Labs, in collaboration with the Italian Institute of Technology and the Technical University of Munich, announced on the 2nd that Intel has demonstrated a new object instance learning using a novel model on its neuromorphic research chip, Loihi.

This presentation targets future applications, including robots interacting with unconstrained environments in logistics, healthcare, and geriatric care. Furthermore, a novel interactive online object learning method using neuromorphic computing will enable additional object learning even after the robot is deployed.

Intel announced that its research results successfully implemented continuous interactive learning with speed and accuracy similar to or better than existing CPUs while using approximately 175 times less power than existing central processing units (CPUs).

To achieve this, the researchers limited the scope of learning to a single layer of plastic synapses, implementing a neural network architecture that describes different object views on Loihi by acquiring new neurons as needed. This neural network architecture was able to autonomously unfold the learning process while interacting with the user.

“When humans learn about a new object, they look at it closely, turn it around, and ask what it is. Then, they can instantly recognize that object in a variety of situations or environments,” said Yulia Sandamirskaya, director of robotics research at Intel’s Neuromorphic Computing Lab and lead author of the paper. “Our goal is to apply similar capabilities to future robots that operate in interactive environments, allowing them to adapt to unexpected situations and work more naturally with humans. Our achievements with Loihi further solidify the value of neuromorphic computing for the future of robotics.”