fixed-point arithmetic

model compression (3302) neural network training (309) forward gradient (2) on-device training (7) memory footprint (21) hardware accelerator (11) integer-only inference (5) dynamic range (8) stochastic rounding (4) quantized training (13)