Reduced precision

Use of lower-precision data types (e.g., 8-bit, 4-bit) instead of standard 32/64-bit floating-point to reduce computational/memory costs in machine learning systems.

References