A Neural Turing Machine Model with Novel Memory Module and Its Setting Method

Abstract

The invention discloses a neural Turing machine model with a novel memory module and a setting method thereof, belonging to the field of neural networks. The present invention adds a double-helix DNA structure in the memory module of the neural Turing machine model; at the same time, it improves the read-write head of the neural Turing machine model so that it can adapt to the storage data of the double-helix DNA structure in the memory module: neural network controller dynamics Read and write the time series data stored in the double-helix DNA structure, use the current offspring data of the double-helix DNA structure as the input of the next layer of neurons, and combine the current content read from the memory matrix to get the activation of the next layer of neurons value, and rewrite it into the double-helix DNA structure to replace the current child data, and the current child data before replacement is converted into the parent data of the current child data after replacement. At the same time, the invention also discloses its corresponding training setting method. The invention makes the update of the memory matrix of the memory module more detailed and improves the stability of the model.

Description

technical field [0001] The invention belongs to the field of neural networks, and in particular relates to a neural Turing machine model with a novel memory module. Background technique [0002] Neural Turing Machines (Neural Turing Machines, NTM) as a new machine learning model, its essence is based on the recurrent neural network to add an additional addressable external memory, so as to improve the performance of the neural network and solve the problems of the recurrent neural network. some flaws. Traditional machine learning algorithms or neural networks have some limitations in the face of the few sample data in behavior detection, the wide variety of sensor combinations, and the difficulty in associating high-level data with underlying logic. On the other hand, the development of artificial intelligence will face a huge bottleneck in the future. The following three points must be the first to accept the challenge: insufficient sample size, insufficient generalization...

Images