US Patent Application 18171433. METHOD OF TRAINING BINARIZED NEURAL NETWORK WITH PARAMETERIZED WEIGHT CLIPPING AND MEMORY DEVICE USING THE SAME simplified abstract
METHOD OF TRAINING BINARIZED NEURAL NETWORK WITH PARAMETERIZED WEIGHT CLIPPING AND MEMORY DEVICE USING THE SAME
Organization Name
Inventor(s)
METHOD OF TRAINING BINARIZED NEURAL NETWORK WITH PARAMETERIZED WEIGHT CLIPPING AND MEMORY DEVICE USING THE SAME - A simplified explanation of the abstract
- This abstract for appeared for US patent application number 18171433 Titled 'METHOD OF TRAINING BINARIZED NEURAL NETWORK WITH PARAMETERIZED WEIGHT CLIPPING AND MEMORY DEVICE USING THE SAME'
Simplified Explanation
The abstract describes a method for training a binarized neural network (BNN). In this method, a binarized weight set is created by applying a clipping function to a weight set. The BNN then generates output data by performing a forward computation on the binarized neural network using input data and the binarized weight set. A gradient of the weight set is calculated by performing a backward computation on the BNN using the loss calculated from the output data. The weight set is updated based on the gradient and the range of the clipping function is adjusted to train the BNN.
Original Abstract Submitted
In a method of training a binarized neural network (BNN), a binarized weight set is generated by applying a clipping function to a weight set. Output data is generated by sequentially performing a forward computation on the binarized neural network based on input data and the binarized weight set. A gradient of the weight set is generated by sequentially performing a backward computation on the binarized neural network based on loss calculated from the output data. The binarized neural network is trained by updating the weight set based on the gradient of the weight set and changing a range of the clipping function.
