Evolutionary aspects of reservoir computing

Abstract

Reservoir Computing (RC) is a powerful computational paradigm that allows high versatility with cheap learning. While other artificial intelligence approaches need exhaustive resources to specify their inner workings, RC is based on a reservoir with highly non-linear dynamics that does not require a fine tuning of its parts. These dynamics project input signals into high-dimensional spaces, where training linear readouts to extract input features is vastly simplified. Thus, inexpensive learning provides very powerful tools for decision making, controlling dynamical systems, classification, etc. RC also facilitates solving multiple tasks in parallel, resulting in a high throughput. Existing literature focuses on applications in artificial intelligence and neuroscience. We review this literature from an evolutionary perspective. RC's versatility make it a great candidate to solve outstanding problems in biology, which raises relevant questions: Is RC as abundant in Nature as its advantages should imply? Has it evolved? Once evolved, can it be easily sustained? Under what circumstances? (In other words, is RC an evolutionarily stable computing paradigm?) To tackle these issues we introduce a conceptual morphospace that would map computational selective pressures that could select for or against RC and other computing paradigms. This guides a speculative discussion about the questions above and allows us to propose a solid research line that brings together computation and evolution with RC as a working bench.

Liquid brains, solid brains

“…Wood [41] suggests that understanding capabilities of different kinds of brains requires a more formal distinction between information processing and computation, terms that are often used interchangeably. Other powerful computational paradigms, such as Reservoir Computing might be more appropriate when dealing with biological information processing, as discussed in [42].…”

Locating the Inexhaustible: Material, Medium, and Ambient Information

“…The framework of reservoir computing suggests that the reservoir (i.e., a medium such as a liquid filter in liquid state machines) that exists independent of the task-specific readouts, if it has rich and diverse enough dynamics to differentiate the different sources of disturbances, could make available the opportunities for realtime, task-specific control of the medium-readout system. In reservoir computing, the only task of the reservoir is to have its dynamic state perturbed by some event (Seoane, 2019). In doing so, through its non-linear, convoluted dynamics, the reservoir is picking up the event and projecting it into the high-dimensional space that consists of various possible dynamic configurations of the reservoir, which could potentially render relevant features from the event more easily separable.…”

Fate of Duplicated Neural Structures

“…Evidence of computing principles compatible with reservoir-like behavior in CCs is scarce [ 56 , 168 , 169 , 170 , 171 ]. In contrast, there is abundant evidence of task-specificity.…”

“…Individual CCs of this cortex have grown and thickened through evolution to take care of the somatosensory processing of each whisker [ 167 , 172 ]. The purported versatility of CCs (including this capacity to commit to a task and change morphology over evolutionary time), make them an interesting candidate, as a computing unit, to test the ideas exposed in the Introduction [ 56 ]: What is the steady shape and dynamics of a CC under fixed computational and thermodynamic (energetic, input entropy, etc.) constraints?…”

“…For example, it could tell whether “brains” are likely to have a solid substrate (such as the cortex or laptops [ 39 ]) or a liquid substrate (such as ants, termites, or the immune system [ 39 , 48 , 49 ]). We could also derive phase diagrams saying when computing units behaving as reservoirs (as in Reservoir Computing [ 50 , 51 , 52 , 53 , 54 , 55 ]) are more efficient [ 56 ] or, in a different context, whether redundancy or regeneration is a favored strategy to overcome neural damage [ 37 ] ( Figure 1 d). Matter samples that undergo phase transitions often lose or gain some symmetry [ 57 , 58 ] (e.g., as a glass’s grid invariance fades upon melting).…”