Nakakuki Laboratory

Keywords: DNA computing, memory and learning, plasticity

​

[Purpose of research]

In this research theme, wet systems are answered by answering questions such as how to realize memory by chemical reaction, how to write and read memory, and what the learning rules are. We aim to develop a memory / learning molecular circuit. This is nothing but the aim of realizing artificial intelligence that operates on the principles of chemistry, so to speak, chemical AI. In other words, "What is a chemically implementable AI framework?" Is the core academic question of this research theme.

 

[Research content]

DNA computing is an academic field related to theory and technology for rationally designing and implementing artificial molecular reaction systems (hereinafter referred to as molecular circuits) that perform desired information processing. There are two main types of analogies in molecular circuits.

 

The first is a semiconductor integrated circuit. For example, a typical example is research in which logic functions such as AND, OR, and NOT functions in Boolean algebra are realized by DNA molecular circuits, and logic circuits are constructed by combining them.

​

The second analogy is the neural network of the brain. The brain is a collection of a huge number of neurons, and each neuron is a chemical reaction system in a small compartment. For example, if a space of 1 ㎤ is filled with 1 pL of neurons, a network system consisting of 1 billion neurons can be constructed.

 

In this research, based on the latter analogy, we will establish the theory and technology for realizing intelligent information processing such as the cranial nerve network with a molecular network.

 

Neurons in the cranial nerve network are, so to speak, thermoplastic learning molecular circuits with compartments. Here, plasticity is considered to be the property that the internal state of the system can be dynamically updated in response to an external stimulus. In recent years, the challenge of implementing the plasticity (as a requirement) that learning molecular circuits should have on DNA molecular circuits has begun worldwide. To realize circuit plasticization, a reset mechanism that transitions the terminal state after the reaction to the initial state again is required. On the other hand, the artificial molecular circuit is also a reaction mechanism based on the law of thermodynamics, and the reaction proceeds in the direction of minimizing the free energy, so once the reaction settles in the most stable state, in principle any further reaction will occur. It doesn't happen. At present, no method has been established for freely transitioning the initial and terminal states of such a molecular reaction system. Furthermore, in order to give the plastic molecular circuit the memory / learning ability, it is necessary to have a learning rule that stores the repeatedly applied input stimulus as an internal state and updates the initial state toward the acquisition of the function to be learned. It becomes. Currently, in the AI field, the error backpropagation method that can be implemented in von Neumann computers is used, but there are many unclear points about the learning principle of neurons in the cranial nerve network. On the other hand, what are the learning rules that can be implemented on the platform of chemical reaction system? The theme is almost unexplored research field.

 

For more specific research themes, please see the following sites.

​

Academic Transformation Area Research (A) Molecular Cybernetics / Planning Research (Development of Memory / Learning Molecular Circuits for Minimal Artificial Brain)