Neuronal cultures and reservoir computing
Many real world information processing task are temporal. Living organisms receive input signals overtime and must take decision according to the processing of pieces of information arriving in series. This is also true for many technological applications such as prediction of weather, economy and finance, prediction of robot movements along its displacement, recognition and processing of vision and sound as they arrive.
However most popular machine learning algorithms have been developed to process non-temporal data. This is the case of formal neuronal network which implement layers of feed-forward structures and are at the basis of deep learning algorithms so successful now-days.
Recurrent Neural Networks introduce « recurrent » connections between the neurons consisting in the possibility of feedback loops, an output reintering the system as input. The recurrent connections cause the system to have potentially very complex dynamics and it has been theoretically shown that such recurrent neuronal network are very powerful tools for solving complex temporal machine learning tasks. However, the complexity introduced by the loops make these system difficult to control and tune.
Reservoir Computing is a variant of recurrent neuronal networks providing an intriguing computational framework for computation suted for temporal processing yet amenable to tunable control.
According to the Wikipedia entry on Reservoir computing :
« Typically an input signal is fed into a fixed (random) dynamical system called a reservoir and the dynamics of the reservoir map the input to a higher dimension. Then a simple readout mechanism is trained to read the state of the reservoir and map it to the desired output. The main benefit is that training is performed only at the readout stage and the reservoir is fixed. Liquid-state machines[2] and echo state networks[3] are two major types of reservoir computing »
The reservoir appears alike a homogeneous neuronal culture where neurons randomly connect.
So the question I have is : can we do reservoir computing using neuronal cultures ?