US Patent Application 18355243. Systems and Methods for Machine-Learned Models Having Convolution and Attention simplified abstract

Systems and Methods for Machine-Learned Models Having Convolution and Attention

Organization Name

Google LLC

Inventor(s)

Zihang Dai of Cupertino CA (US)

Mingxing Tan of Newark CA (US)

Quoc V. Le of Sunnyvale CA (US)

Hanxiao Liu of Santa Clara CA (US)

Systems and Methods for Machine-Learned Models Having Convolution and Attention - A simplified explanation of the abstract

This abstract first appeared for US patent application 18355243 titled 'Systems and Methods for Machine-Learned Models Having Convolution and Attention

Simplified Explanation

The patent application describes a method for computer vision that reduces computational cost and improves accuracy.

• The method involves using a machine-learned convolutional attention network to process input data.
• The convolutional attention network consists of multiple stages and includes at least one attention block.
• The attention block uses a relative attention mechanism, which combines a static convolution kernel with an adaptive attention matrix.
• This approach improves the generalization, capacity, and efficiency of the convolutional attention network compared to existing models.

Original Abstract Submitted

A computer-implemented method for performing computer vision with reduced computational cost and improved accuracy can include obtaining, by a computing system including one or more computing devices, input data comprising an input tensor having one or more dimensions, providing, by the computing system, the input data to a machine-learned convolutional attention network, the machine-learned convolutional attention network including two or more network stages, and, in response to providing the input data to the machine-learned convolutional attention network, receiving, by the computing system, a machine-learning prediction from the machine-learned convolutional attention network. The convolutional attention network can include at least one attention block, wherein the attention block includes a relative attention mechanism, the relative attention mechanism including the sum of a static convolution kernel with an adaptive attention matrix. This provides for improved generalization, capacity, and efficiency of the convolutional attention network relative to some existing models.