Supervised Learning in a Multilayer, Nonlinear Chemical Neural Network

Design of Universal Chemical Relaxation Oscillator to Control Molecular Computation

“…We set up a universal chemical oscillator structure to solve the problem of how to stagger the previously disorderly reactions to make these reactions happen in the order we want. Much of the previous work mentioning chemical oscillators followed the logic of usability, how the oscillation is generated, how it is controlled by the model parameters and how the setting of initial values affects the oscillation properties are not involved in these work [8,1,10]. While theoretical analysis of the models and mechanisms that cause oscillations is improving day by day [28,24,23], it is feasible in theory to design transparent chemical oscillators according to actual needs.…”

“…Different from the harmonic oscillators used in previous work [8,10], we choose relaxation oscillation as underlying structure of our oscillator model for that mechanism of relaxation oscillation is clear and it is robust independent to initial points. While existence and property of harmonic oscillators are depend on the selection of initial point, which is not flexible in response to specific application requirements.…”

“…11 shows, if we just want the loop of system ′ 1 − ′ 2 to operate for times and output the termination value of ′ 1 , we just need to use truncated subtraction + 1 − 2 as termination component and add it into system ′ 1 − ′ 2 as Example 5.2 demonstrates. Similar cases include judging first and then performing the corresponding operation, such as module comparing output with threshold and module performing weight learning in supervised neural network based on chemical reactions [8], and case where the main computation is concentrated in the second half of the loop. For these situations, our strategy of loop termination as Example 5.2 is effective.…”

“…Our primary goal in designing chemical oscillator is to achieve efficient molecular computation. Recent attempts to build artificial neural networks in biochemical environments [2,3,4,5,8,13] have not only improved the computational power of molecular computers [30], but also helped advance the understanding of how living cells perform complex operations.…”

“…Oscillation phenomena are often encountered in chemical and biological systems such as Belousov-Zhabotinskii reaction [6] and circadian rhythm [7], while dealing with reaction orders by chemical oscillators is not a groundless rumour. Arredondo and Lakin [8] utilized a 20-dimensional oscillator extended from ant colony model [9] to order the parts of their chemical neural networks. Jiang et al introduced a different oscillator model with 12 species and 24 reactions [10], then chose two of these species to serve as clock signals.…”