Projects
Mass-spring reservoir computer
An input signal shakes a grid of masses on springs, and a linear readout trained live by ridge regression reads their displacements to reproduce a target function of that input. The damping, stiffness, and drive controls show how narrow the window is where a physical reservoir works at all.
Reservoir dog
An Echo State Network watches where you drag a cursor and learns to predict where you are going next, trained online with Recursive Least Squares rather than from an offline dataset. Once it has your pattern the follower stops trailing behind you and starts cutting corners ahead of you.
Echo state melody lab
An echo state network learns to play a melody back from memory, with the readout fitted by a single closed-form ridge solve instead of gradient descent. Free-run accuracy is reported next to teacher-forced accuracy, because the two diverge as soon as the output is fed back as the input.
