US Patent Application 18249723. NEURAL ARCHITECTURE SEARCH SYSTEM AND SEARCH METHOD simplified abstract

NEURAL ARCHITECTURE SEARCH SYSTEM AND SEARCH METHOD

Organization Name

NIPPON TELEGRAPH AND TELEPHONE CORPORATION

Inventor(s)

Yuki Arikawa of Tokyo (JP)

Kenji Tanaka of Tokyo (JP)

Tsuyoshi Ito of Tokyo (JP)

Tsutomu Takeya of Tokyo (JP)

Takeshi Sakamoto of Tokyo (JP)

NEURAL ARCHITECTURE SEARCH SYSTEM AND SEARCH METHOD - A simplified explanation of the abstract

This abstract first appeared for US patent application 18249723 titled 'NEURAL ARCHITECTURE SEARCH SYSTEM AND SEARCH METHOD

Simplified Explanation

The patent application describes a neural architecture search system that improves the accuracy of neural networks while considering deployment constraints. Here are the key points:

• The system converts a constraint condition of a neural network implementation into a constraint condition for the network's architecture parameters.
• A learning engine unit trains the neural network and calculates its inference accuracy under a search condition.
• A model modification unit adjusts the architecture of the neural network based on the inference accuracy and the constraint condition.
• The goal is to find the best architecture that maximizes inference accuracy while meeting the given constraints.

Original Abstract Submitted

A neural architecture search system includes a deployment constraint management unit that converts a first constraint condition that defines a constraint of a system that implements a neural network into a second constraint condition that defines a constraint of a parameter that prescribes an architecture of the neural network, a learning engine unit that performs learning of the neural network under a search condition and calculates inference accuracy in a case where the learned neural network is used, and a model modification unit that causes the learning engine unit to perform the learning and the calculation of the inference accuracy while changing the architecture of the neural network on the basis of the inference accuracy and the second constraint condition so as to obtain the best inference accuracy.