17965141. METHOD AND APPARATUS FOR COMPRESSING WEIGHTS OF NEURAL NETWORK simplified abstract (Samsung Electronics Co., Ltd.)
METHOD AND APPARATUS FOR COMPRESSING WEIGHTS OF NEURAL NETWORK
Organization Name
Inventor(s)
METHOD AND APPARATUS FOR COMPRESSING WEIGHTS OF NEURAL NETWORK - A simplified explanation of the abstract
This abstract first appeared for US patent application 17965141 titled 'METHOD AND APPARATUS FOR COMPRESSING WEIGHTS OF NEURAL NETWORK
Simplified Explanation
The abstract describes a method for compressing the weights of a neural network. The method involves compressing a weight set of the neural network and then determining modified weight sets by changing at least one of the weights. Compression efficiency values are calculated for these modified weight sets based on the results of compressing them. A target weight is determined among the weights that satisfies a compression efficiency condition. Finally, the weights are compressed by replacing the determined target weight.
- Method for compressing weights of a neural network
- Compresses a weight set of the neural network
- Determines modified weight sets by changing at least one weight
- Calculates compression efficiency values for the modified weight sets
- Determines a target weight satisfying a compression efficiency condition
- Compresses the weights by replacing the target weight
Potential Applications
- Efficient storage and transmission of neural network weights
- Faster inference and training of neural networks
- Improved performance of resource-constrained devices using neural networks
Problems Solved
- Reduces the size of neural network weights for efficient storage and transmission
- Addresses the challenge of compressing weights without significant loss of accuracy
- Enables deployment of neural networks on resource-constrained devices
Benefits
- Reduced storage requirements for neural network weights
- Faster transmission of neural network weights over networks
- Improved performance and efficiency of neural networks on devices with limited resources
Original Abstract Submitted
A method of compressing weights of a neural network includes compressing a weight set including the weights of a the neural network, determining modified weight sets by changing at least one of the weights, calculating compression efficiency values for the determined modified weight sets based on a result of compressing the weight set and results of compressing the determined modified weight sets, determining a target weight of the weights satisfying a compression efficiency condition among the weights based on the calculated compression efficiency values, and determining a final compression result by compressing the weights based on a result of replacing the determined target weight.
