Abstract: The efficiency of neural network computations is heavily influenced by numerical precision management. Traditional floating-point (FP) representations offer high precision but are ...