The U-M team devised an algorithm that provides a mathematical framework for how learning works in lattices called mechanical neural networks. This shows that materials can learn tasks and perform computations.
By using an algorithm based on backpropagation, the researchers demonstrated that physical materials could be trained to solve specific problems, like identifying species of iris plants.
Li noted that, while applications are still evolving, these mechanically learned materials could lead to smarter structures, such as airplane wings that adapt shape based on environmental factors.
The research not only opens avenues for mechanical systems but also might inform biologists about the functioning of biological neural networks by revealing parallel learning mechanisms.
#machine-learning #mechanical-neural-networks #backpropagation #physical-systems #computational-materials
Collection
[
|
...
]