Alibaba Group has been granted a patent for a data compression method in multi-core processors. The method involves retrieving data from cache slices, calculating a bit mask, and shifting out zero values while retaining non-zero elements, which are then written to memory, enhancing machine learning efficiency. GlobalData’s report on Alibaba Group gives a 360-degree view of the company including its patenting strategy. Buy the report here.

According to GlobalData’s company profile on Alibaba Group, Blockchain-based property authentication was a key innovation area identified from patents. Alibaba Group's grant share as of July 2024 was 46%. Grant share is based on the ratio of number of grants to total number of patents.

Data compression method for multi-core processor cache

Source: United States Patent and Trademark Office (USPTO). Credit: Alibaba Group Holding Ltd

The granted patent US12072806B2 outlines a system and method for enhancing machine learning processes through efficient data compression and decompression in a multi-core processor environment. The system comprises a processor with multiple cores and cache slices, where each core can access these cache slices. Each cache slice includes a data array for storing decompressed data and a cache controller responsible for managing data retrieval and compression. The cache controller retrieves chunks of data from an activation map of a neural network, calculates a bit mask to identify zero values, and compresses the data by retaining non-zero elements. The compressed data, along with the bit mask, is then written to memory, ensuring that memory access occurs efficiently, particularly after convolution operations.

Additionally, the patent details a method for decompressing data within the same multi-core processor framework. The cache controller retrieves the bit mask and non-zero elements from memory before performing a convolution operation. Using the bit mask and a prefix sum, the controller generates a linear shift offset to reconstruct the original data by inserting zero values at the appropriate locations. This method allows for the efficient handling of data, as non-zero elements are retrieved and processed one cache line at a time until the entire compressed chunk is decompressed. The claims emphasize the programmability of data chunk lengths and the separation of write operations for the bit mask and non-zero elements, contributing to the overall efficiency of the machine learning system.

To know more about GlobalData’s detailed insights on Alibaba Group, buy the report here.

Data Insights

From

The gold standard of business intelligence.

Blending expert knowledge with cutting-edge technology, GlobalData’s unrivalled proprietary data will enable you to decode what’s happening in your market. You can make better informed decisions and gain a future-proof advantage over your competitors.

GlobalData

GlobalData, the leading provider of industry intelligence, provided the underlying data, research, and analysis used to produce this article.

GlobalData Patent Analytics tracks bibliographic data, legal events data, point in time patent ownerships, and backward and forward citations from global patenting offices. Textual analysis and official patent classifications are used to group patents into key thematic areas and link them to specific companies across the world’s largest industries.