Physical Learning in Nonreciprocal Active Spinner Materials

Learning is often viewed as an abstract computational process, yet it can also emerge from the physical dynamics of matter itself. Building on the Hopfield model, where static memory arises from symmetric interactions in equilibrium networks, we extend this concept to nonreciprocal active matter to encode dynamic memory. Simulations suggest that target oscillatory responses can be encoded in a network of spinning particles by tuning their spinning frequencies. We explore this in an experimental platform of self-spinning magnetic particles with tunable nonreciprocal interactions. Taken together, this approach will introduce a new class of adaptive active materials in which memory and learning emerge directly from physical interactions, realizing embodied computation in active materials.