18107210. METHOD AND APPARATUS FOR PROCESSING COMPUTATION OF ZERO VALUE IN PROCESSING OF LAYERS IN NEURAL NETWORK simplified abstract (Samsung Electronics Co., Ltd.)
METHOD AND APPARATUS FOR PROCESSING COMPUTATION OF ZERO VALUE IN PROCESSING OF LAYERS IN NEURAL NETWORK
Organization Name
Inventor(s)
Saptarsi Das of Bangalore (IN)
Sabitha Kusuma of Bangalore (IN)
Ankur Deshwal of Bangalore (IN)
Kiran Kolar Chandrasekharan of Bangalore (IN)
METHOD AND APPARATUS FOR PROCESSING COMPUTATION OF ZERO VALUE IN PROCESSING OF LAYERS IN NEURAL NETWORK - A simplified explanation of the abstract
This abstract first appeared for US patent application 18107210 titled 'METHOD AND APPARATUS FOR PROCESSING COMPUTATION OF ZERO VALUE IN PROCESSING OF LAYERS IN NEURAL NETWORK
Simplified Explanation
The patent application describes a method and apparatus for processing layers in a neural network. Here is a simplified explanation of the abstract:
- The method and apparatus fetch tiles of Input Feature Map (IFM) and kernel tensors.
- A convolutional operation is performed on the IFM and kernel tiles, taking advantage of the sparsity of both IFM and kernel.
- The result is a set of Output Feature Map (OFM) tiles that correspond to the IFM tiles.
Potential applications of this technology:
- Deep learning: The method can be used in deep learning models to improve the efficiency and speed of convolutional operations.
- Image recognition: The technology can be applied to image recognition tasks, such as object detection and classification.
- Natural language processing: The method can be used in neural networks for tasks like sentiment analysis and language translation.
Problems solved by this technology:
- Efficiency: By exploiting the sparsity of IFM and kernel, the method reduces the computational requirements of convolutional operations, making them more efficient.
- Speed: The method allows for faster processing of layers in a neural network, enabling real-time or near real-time applications.
- Resource utilization: By generating OFM tiles corresponding to IFM tiles, the method optimizes the use of computational resources.
Benefits of this technology:
- Improved performance: The method enhances the performance of neural networks by reducing computational complexity and improving efficiency.
- Faster processing: The technology enables faster processing of layers in a neural network, leading to quicker results.
- Resource optimization: By exploiting sparsity and generating corresponding OFM tiles, the method optimizes the use of computational resources, making neural networks more resource-efficient.
Original Abstract Submitted
A method and an apparatus for processing layers in a neural network fetch Input Feature Map (IFM) tiles of an IFM tensor and kernel tiles of a kernel tensor, perform a convolutional operation on the IFM tiles and the kernel tiles by exploiting IFM sparsity and kernel sparsity, and generate a plurality of OFM tiles corresponding to the IFM tiles.
