Quantization and Compression

Low-Rank Matrix Factorization

Compression technique that decomposes a large weight matrix into two or more smaller matrices. It reduces the number of parameters and matrix multiplication operations, thus accelerating dense and convolutional layers.

← Indietro