Unconventional computer
Recently, there has been a growing interest into the development of hybrid wetware silicon devices focused on its applications in non-linear media computing and more specifically in cultured neurons. The objective of this research is to exploit the partially understood operations of vitro –neural networks to carry out more advanced computations effectively as compared to traditional architecture in order to facilitate the understanding of how the nervous system works. The outcomes of this study have come with it the increase in machines that have improved capability of cell culturing, wetware silicon interfacing as well as neurobiology. Medically these systems have been used in cases such as prosthetics and study of degenerative diseases.
Furthermore, research into vitro networks has led to the potential discovery of generic neurons behaviors since they are created from dissociated cells hence the self-organization of characteristics identifiable. This has led to achievements such as simple computation-like behavior.
A case study presented by Shahaf and Maron has established behaviors in cultures networks obtained from pre-determined neuron (electrode) stimulus removal. It has been found out that the stimulus is behind the driving response of the network resulting in induced learning upon its removal. All these are established via prior simulation.
Critical view of the multi-electrode cathode array
Multi-electrode array (MEA) technology has now been made available by commercial companies. Multi-electrode cathode arrays have provided for long term recording of large neural network as compared to traditional single number of neuron recordings done by use of patch clamps. Recent studies have also established a fundamental property for such network. That is: Tetanic-constant volley-electoral stimulations that shown to induce excitatory, inhibitory and neural responses to simple electrical stimulations.
Recent presentation by Shahaf and Maron has showed how immediate stimulus was seen immediately after a few iterations were applied. They conclude their ability in establishing the possibility of obtaining similar behavior from the aforementioned re-aggregated neural network prior to a set of conditioned using their stimulation protocol.
It has also been found that some excitatory networks displayed exhaustion after periodic increase in spiking response. This may as well explain the consistency of the ability to obtain ‘learning’ curves in tandem to Shahaf and Marons work on the same.
Related to this, it could be well established that there have been a pattern recognition resulting upon the spatial separation of electrodes from the stimulation points to enable differential learning. This is one part of multi-electrode cathode array that depicts that majority of electrodes for either input happens to be maximally separated.
Furthermore examination on control experiments undertook by Shahaf and Maron established similar responses as earlier seen which is excitatory, inhibitory and neural in nature. It has been expected that maybe considering the fact that networks are randomly formed from computers which exhibit excitatory characteristics in nature.
Therefore all these studies and researched shows that neuronal networks grown from disassociated cells on multi-electrode cathode array have their behaviors changed upon electrical stimulations by dominantly exhibiting effects that support the underlying properties of present networks.
Universal neural computation
The realization of neural computation can be achieved through the logic AND, NOT and OR. For instance spiking response of the network have exhibited a form of logical NOT operations since stimulations have caused recent reductions in spiking behavior. This was established on the presumption of a stimulus as a logical ‘1’ input and two standard deviations from spontaneous behaviors as logical ‘0’ thereby a NOT effect was realized under such experiments.
It can now be said that this function, combined with the NOT gate concludes the possibility of a universal computation on conditions that the neuronal networks are appropriately interconnected and trained. This has suggested a great improvement in the realization of a universal computation.
Current establishments of Vitro neural network behaviors.
The recent findings on the exploitations of multi-electrode cathode arrays in network behavior has drawn immense interest in behaviors of networks as well as its computation properties. This has further showed how logic gates properties can be adopted in universal computations. This has even encouraged future works to explore development in utilization of other logic gates operations in vitro research.
The case for neural coding with precise spikes
Neural brain spike case study
It is believed that the human brain consists of vast numbers of neurons estimated to be in ranges of 10 to 100 billion. These neurons are distributed although the anatomical structures. In each of these structures different types of neurons exist each having different connectivity patterns, different responses and performs different tasks. The neurons communicate with each other through spikes or action potentials.
Despite the fact that the working of individual neurons can be reasonably established, detailed understanding of the information processing behind spiking neurons still remains a mystery. Considering the development of sophisticated methods advanced through neuroscience, it serves to support the present knowledge that neurons do not communicate by frequency and these findings remain quite controversial and has then become a topic of intense debate.
The continuous research on brain spikes brings out a clear argument on simple rate-coding for neurons that has to now demonstrate convincingly beneficial schemes in making networks more efficient.
Accuracy and reliability of real spikes
In regards to the Vistro experiments, several works have been carried out under several researches that have demonstrated the potential of neural computations.
The blow fly studies
One of the studies conducted by De Ruyter van Steveninck and Bialek was on time-response of the blow fly’s motion sensitive neuron H1 during its flight. They were able to establish the decoding method for reconstructing the known original environment as perceived by the blow fly’s input receptors. Furthermore they have managed to decode the signal of H1 and this has helped them to estimate the blow fly velocity. One key finding was the accuracy of the decoded signal was observed to increase as the spikes were observed with greater temporal precision.
Secondly the timing relationships between neural response and stimulus were obtained with millisecond precision. A conclusion of this results suggested that the neurons of the fly’s visual system employs a multi layered coding scheme where information conveying emitted by spikes enables the fly to encode stimulus by modifying their inter-spike interval.
Studies on invertebrates
In another research done on cats by Liu and company, they remarked that the degree of precision only becomes apparent when an adequate length of stimulus sequence was specified to determine the neural response. This emphasized that the relevance of cell response must be controlled in order to observe cells instinct response precision.
Another study on invertebrates as by Bierholm and company added to the present finding that reported a reliable precision of single spike in the order of 2-3ms, but only when the input was at a high frequency of (3-30Hz, 30-200pA).
The case of Spike-time coding in neural populations
A report postulated by deCharms and Merzenich on macaque’s primary auditory cortex showed that the populations of neurons can coordinate the relative timing of their action potentials resulting in spike occurring closer together in time during continuous stimuli. This presents a way by which neurons can signal stimuli even when their firing rated don not change.
Conclusion of neural information processing
Furthermore two important questions are now being raised: one being related to neurophysiology methods that suggests sophisticated methods are covering increasing sophisticated spiking patterns of neuron behaviors. This tends to put a burden of proof on convectional knowledge. A second question relates to the future of artificial neural networks in particular the finding of precise spikes that in principle just implies some parts of the brain is implemented with spiking neurons using reliable spiking timing.
Finally we can say that prior to these research findings, they suggest that despite the fact that artificial spiking can be implemented with far fewer biological spiking neurons, this can still be explicitly compensated in the same way through increased robustness or through large network effects.
All this presents a big challenge but will eventually go a long way in contribution to the cracking of the ‘neural code’.
References
Bohte, S. M. (2004). The evidence for neural information processing with precise spike-times: A survey. Natural Computing , 195–206.
Larry Bull, I. S. (2008). Towards Neuronal Computing: Simple Creation of Two Logic Functions in 3D Cell Cultures using Multi-Electrode Arrays. Int. Journ. of Unconventional Computing , 4, 143–154.
