Last November, Google announced that it would provide its machine learning framework as an open architecture, hoping to leverage the collective power of various machine learning platforms to build cumulative experience and thereby drive the scale of AI system applications. Amazon then went a step further by announcing that it would open source its own machine learning software, DSSTNE, and publish the source code architecture on its GitHub forum.
Amazon announcedAmazon will make the source code for its DSSTNE machine learning software open source, and will also release the source code architecture through its GitHub forum. According to Amazon, DSSTNE performs better when there's less reference data for comparison, with overall processing speed approximately 2.1 times faster than Google's TensorFlow machine learning framework. However, TensorFlow still offers greater efficiency for analyzing and learning large amounts of data.
DSSTNE was originally implemented by Amazon as a product recommendation feature within its online retail platform. By comparing user browsing behavior on Amazon services, it predicts and assesses which products users might be interested in, allowing for targeted advertising. This type of machine learning uses reference data, specifically the content opened when a user first browses Amazon's online retail platform, and quickly searches for corresponding products. This allows users to easily find more similar products, thereby increasing their purchasing opportunities.
Compared to TensorFlow, which uses Google's own cloud servers to analyze and compare massive amounts of data, allowing the system to learn various "experiences", DSSTNE uses a native architecture to accelerate calculations on multiple GPUs. It can also configure different GPUs based on computing needs, using parallel computing mode to achieve faster processing speeds. Therefore, it can complete data analysis in a shorter time than many machine learning software.
