【Seminar】"How Circuits Learn: Projection Operators, Memory, and Computation in Analog Hardware" by Dr. Frank Barrows

Analog computing has several potential advantages beyond efficiency gains. The central opportunity is algorithmic: physical constraints (Kirchhoff's laws, conservation, dissipation, phase locking) let us embed a problem directly into a dynamical circuit so that computation is performed by continuous-time evolution toward fixed points, limit cycles, or steady states. 

In this talk I will outline a theory of physical computation in which hardware primitives are compared by the operators they realize: fixed-point maps, projectors, attractor basins, and compositional rules. I will ground this perspective in recent work on memristive and resistive networks, where circuit topology induces projection operators that determine reachable fixed points, capacity bounds, and the composability of trained submodules; exact physical learning rules for passive resistor networks; and coupled oscillator networks whose attractor landscapes support autonomous learning and generative modeling.