unearth.wiki

Reservoir Computing

/ˈrez.ə.vwɑː kəmˈpjuː.tɪŋ/ Dynamical Systems Theory
Definition A computational paradigm that utilizes a complex, non-linear dynamical system (the "reservoir") to map inputs into a high-dimensional space. Unlike traditional neural networks where all weights are trained, in Reservoir Computing, the internal dynamics of the reservoir are fixed, and only a simple readout layer is trained.

The Mycelial Reservoir

In the Myceloom framework, specific focus is given to physical reservoirs—specifically, biological ones. Mycelial networks naturally exhibit the properties required for a good reservoir: complex non-linear responses, fading memory (hysteresis), and high dimensionality.

Researchers can treat a living mycelial network as a "black box" computational resource. By stimulating it with input signals (chemical or electrical) and measuring its complex response, the biological network performs the difficult task of transforming data into separable patterns. A simple digital classifier then reads these patterns.

Training the Readout, Not the Brain

This approach bypasses the immense energy cost of "training" a large neural network (backpropagation). Instead of forcing the network to behave in a specific way, we simply observe its rich natural behavior and learn how to interpret it. It is akin to learning to read the ripples in a pond rather than trying to engineer every water molecule's position.