Advanced Micro Devices had 49 patents in artificial intelligence during Q2 2024. AMD has developed methods and apparatus for creating less computationally intensive nodes in neural networks without the need for retraining, reducing computation while maintaining accuracy. They have also designed a programmable integrated circuit with a dynamic function exchange module for efficient data transformation. Additionally, they have introduced a distributed cache network for machine learning, optimizing data caching based on access frequency. Furthermore, AMD has implemented a method for selecting bit width during training of neural networks and developed a design tool for pruning weights in neural networks based on configurable data types and circuit structures. GlobalData’s report on Advanced Micro Devices gives a 360-degree view of the company including its patenting strategy. Buy the report here.

Advanced Micro Devices had no grants in artificial intelligence as a theme in Q2 2024.

Recent Patents

Application: Optimizing low precision and sparsity inference without retraining (Patent ID: US20240211762A1)

The patent filed by Advanced Micro Devices Inc. describes an apparatus and method for efficiently creating less computationally intensive nodes for a neural network. The system includes a processor and memory storing input data values to process during inference of a trained neural network. The processor determines which node values to represent in a floating-point format with less precision during inference, using selection criteria to reduce computation while maintaining accuracy above a threshold. The updates to representations occur during inference without retraining, reducing the number of layers, nodes within a layer, and weight values per node to inspect.

The apparatus and method involve selecting node values of a trained neural network in a floating-point format based on selection criteria, generating output values during inference using a different floating-point format, and replacing the original format based on target metrics. The second format has less precision, and subsets of layers and available formats are chosen based on the type of neural network. The target metrics include accuracy and data storage comparisons between output values generated using different formats. The computing system includes a processor and memory to store weight values, with the processor selecting, generating, and replacing node values in different formats based on criteria and metrics to optimize computational efficiency while maintaining accuracy.

To know more about GlobalData’s detailed insights on Advanced Micro Devices, 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.