Google llc (20240289285). EXPLOITING INPUT DATA SPARSITY IN NEURAL NETWORK COMPUTE UNITS simplified abstract
Contents
EXPLOITING INPUT DATA SPARSITY IN NEURAL NETWORK COMPUTE UNITS
Organization Name
Inventor(s)
Dong Hyuk Woo of San Jose CA (US)
Ravi Narayanaswami of San Jose CA (US)
EXPLOITING INPUT DATA SPARSITY IN NEURAL NETWORK COMPUTE UNITS - A simplified explanation of the abstract
This abstract first appeared for US patent application 20240289285 titled 'EXPLOITING INPUT DATA SPARSITY IN NEURAL NETWORK COMPUTE UNITS
The abstract of this patent application describes a computer-implemented method that involves receiving input activations, determining if they are zero or non-zero values, storing them in memory, and providing them to a computational array.
- Input activations are received by a computing device.
- The controller determines if each input activation is zero or non-zero.
- Non-zero input activations are stored in a memory bank.
- An index is generated with memory address locations of non-zero input activations.
- The controller provides the input activations to a computational array from the memory bank.
Potential Applications: - This technology can be used in machine learning algorithms for processing input data efficiently. - It can be applied in neural networks for optimizing memory usage and computational performance.
Problems Solved: - Efficient handling of input activations with zero and non-zero values. - Streamlining data storage and retrieval processes in computational arrays.
Benefits: - Improved memory utilization in computing devices. - Enhanced performance of computational arrays in processing input data.
Commercial Applications: Title: "Efficient Input Activation Handling Technology for Computational Arrays" This technology can be utilized in AI systems, data processing applications, and hardware acceleration solutions for various industries such as healthcare, finance, and autonomous vehicles.
Questions about Efficient Input Activation Handling Technology: 1. How does this technology improve the efficiency of computational arrays in processing input data? 2. What are the potential implications of using this technology in machine learning algorithms for real-time applications?
Frequently Updated Research: Researchers are continuously exploring ways to enhance the performance of computational arrays by optimizing memory utilization and data processing efficiency. Stay updated on the latest advancements in this field to leverage the benefits of this technology effectively.
Original Abstract Submitted
a computer-implemented method includes receiving, by a computing device, input activations and determining, by a controller of the computing device, whether each of the input activations has either a zero value or a non-zero value. the method further includes storing, in a memory bank of the computing device, at least one of the input activations. storing the at least one input activation includes generating an index comprising one or more memory address locations that have input activation values that are non-zero values. the method still further includes providing, by the controller and from the memory bank, at least one input activation onto a data bus that is accessible by one or more units of a computational array. the activations are provided, at least in part, from a memory address location associated with the index.
