Associative memory in mechanical systems

Associative memory is the ability of recurrent neural networks to retrieve one of several stored configurations (`memories') using only partial or corrupted information about the desired memory. We find the conditions under which associative memory is seen in a general family of mechanical networks that include spring networks and folding sheets (origami). We show that the capacity for associative memory requires strong mechanical non-linearities and grows with the range of interactions and the dimensionality of the system.
We classify the different kinds of failure modes when memory capacity is exceeded. In addition to laying out design principles for multi-functional mechanical metamaterials, our work identifies limits on the ability of mechanical systems to learn and adapt to external signals.