Tag: model compression

A new study systematically benchmarks five CNN compression techniques across GPUs, CPUs, and noisy data, finding that the best choice ...

A new theoretical framework trains neural networks natively on the Leech lattice, compressing weights to 0.76 bits per parameter while ...

Researchers in South Korea have developed a progressive distillation framework called ASBQ that trains one-bit binary neural networks to match ...

A hybrid compression framework combining knowledge distillation, quantization-aware training, and dynamic inference achieves 92.3 percent semantic accuracy at 1.5 watts ...

Researchers at the National University of Defense Technology have systematically surveyed memory-efficient training techniques for large Transformer models, showing how ...

Researchers compressed transformer-based tabular neural networks by up to 38-fold using knowledge distillation and deployed them on an NVIDIA Jetson ...

Researchers have released EmbedKD, an open-source toolkit that brings verified diagnostics, retrieval protocols and reproducible deployment checks to the compression ...