Reservoir computing: an interactive simulator
An online reservoir computing simulator — the scheme physical neural networks are built on: one delay line, one feedback loop, one nonlinear node. Dimensionality comes from time, as the line is split into virtual nodes. Only the linear readout is trained; the reservoir itself stays random and unchanged.
How to use the simulator
Start with the gain slider. Watch the regime label: fading, edge of chaos or chaos.
They show how far back the reservoir remembers the input. At the edge of chaos the bars reach furthest.
The dashed line marks your position on the memory curve. Bring it to the peak.
Turn the feedback gain and find the edge of chaos
Virtual node states
time left to rightEach row is a virtual node of the delay line, each column one loop cycle. Blue and red separate the sign of the state.
Memory by delay
How accurately the readout reconstructs the input fed 1…20 cycles ago.
Memory vs feedback gain
peak at —The curve is recomputed for every gain value. The dashed line is where you are.
Only the linear readout is trained — one solved system of equations, no backpropagation and no epochs. That is exactly why such schemes are built in hardware: the reservoir never needs changing.
This models a class of algorithms, not any particular device. Physical reservoirs are built in many media — from optical fibre to spin waves in yttrium iron garnet films.
