Kang Eun Jeon

We propose a one-shot training framework that reduces the training cost for multi-bit quantization networks using weight-space bias correction and bit-wise coreset sampling.