A random network of Ag/Ag2S nanoparticles (NPs) was used as a physical system in reservoir computing (RC) because the network has nonlinear and memory characteristics. Atomic switching among the NPs caused nonlinear dynamical behavior of the random network. The Fourier transform of output signals indicated that the generated harmonics were far higher with a larger amplitude of the input sine wave because the atomic switching occurred only at high bias voltages. Higher accuracy was achieved in the waveform generation task with higher vias voltage. Based on the NARMA task results, we found that the capability of recalling historical data was effectively achieved for the input injection time within the time constant (0.2 s). These findings suggest that nonlinearity and memory function play a fundamental role in the designing and implementing in-materio RC devices.