Quantization Deep Learning; Computational Infrastructure
noun
Definition: The process of mapping values, model parameters, activations, or signals from a high-precision numerical representation to a lower-precision discrete set, typically in order to reduce memory use, computation cost, and deployment complexity [Rifqie et al. 2024].
Example in context: “Techniques for compressing deep learning models, such as pruning, knowledge distillation, and quantization, can be used to provide a more efficient model implementation for deployment within practical hardware constraints.” [Nerella et al. 2024]
Synonyms: model quantization; low-precision quantization
Related terms: pruning; knowledge distillation; mixed precision; dequantization; model compression