18476033. PROCESSING DATA USING CONVOLUTION AS A TRANSFORMER OPERATION simplified abstract (QUALCOMM Incorporated)

PROCESSING DATA USING CONVOLUTION AS A TRANSFORMER OPERATION

Organization Name

QUALCOMM Incorporated

Inventor(s)

Dharma Raj Kc of Tucson AZ (US)

Venkata Ravi Kiran Dayana of San Diego CA (US)

Meng-Lin Wu of San Diego CA (US)

Venkateswara Rao Cherukuri of San Diego CA (US)

PROCESSING DATA USING CONVOLUTION AS A TRANSFORMER OPERATION - A simplified explanation of the abstract

This abstract first appeared for US patent application 18476033 titled 'PROCESSING DATA USING CONVOLUTION AS A TRANSFORMER OPERATION

Simplified Explanation

The abstract describes a method for processing image data using convolution as a transformer (CAT) operations. The method involves applying depth-wise separable convolutional filters and pointwise convolutional filters to generate output features.

• The method involves receiving a set of features associated with an image and applying convolutional filters to generate output features.
• The depth-wise separable convolutional filter is applied to the input features to capture spatial information.
• The pointwise convolutional filter is applied to the output of the depth-wise separable convolution to incorporate global information from both spatial and channel dimensions.
• The output features are modified to the three-dimensional shape and combined with the input features to generate an output set of features.

Potential Applications

This technology can be applied in various fields such as image recognition, object detection, and image classification.

Problems Solved

This technology helps in improving the efficiency and accuracy of image processing tasks by capturing spatial and global information effectively.

Benefits

The benefits of this technology include enhanced feature extraction, improved image analysis, and better performance in machine learning models.

Potential Commercial Applications

One potential commercial application of this technology could be in developing advanced image recognition systems for security surveillance.

Possible Prior Art

Prior art in this field may include traditional convolutional neural networks and other image processing techniques that focus on feature extraction and analysis.

Unanswered Questions

How does this technology compare to other convolutional methods in terms of computational efficiency?

This article does not provide a direct comparison with other convolutional methods in terms of computational efficiency. Further research or experimentation may be needed to evaluate the performance of CAT operations in comparison to traditional convolutional techniques.

What impact does the three-dimensional shape modification have on the final output features?

The article does not delve into the specific impact of modifying the output features to the three-dimensional shape. Understanding the significance of this modification could provide insights into the effectiveness of the CAT operations in image processing tasks.

Original Abstract Submitted

Systems and techniques are described herein for processing data (e.g., image data) using convolution as a transformer (CAT) operations. The method includes receiving, at a convolution engine of a machine learning system, a first set of features, the first set of features being associated with an image and having a three-dimensional shape, applying, via the convolution engine, a depth-wise separable convolutional filter to the first set of features to generate a first output, applying, via the convolution engine, a pointwise convolutional filter to the first output to generate a second output based on global information from a spatial dimension and a channel dimension associated with the image, modifying the second output to the three-dimensional shape to generate a second set of features and combining the first set of features and the second set of features to generate an output set of features.