On Expressive Power of Quantized Neural Networks under Fixed-Point Arithmetic
This work addresses the theoretical limitations of quantized networks for efficient AI deployment, providing foundational insights but is incremental as it builds on existing expressive power studies.
The paper tackles the problem of understanding the expressive power of quantized neural networks under fixed-point arithmetic with round-off errors, showing that networks with popular activation functions like ReLU and Sigmoid can represent all fixed-point functions, and even those with binary weights achieve this capability.
Existing works on the expressive power of neural networks typically assume real parameters and exact operations. In this work, we study the expressive power of quantized networks under discrete fixed-point parameters and inexact fixed-point operations with round-off errors. We first provide a necessary condition and a sufficient condition on fixed-point arithmetic and activation functions for quantized networks to represent all fixed-point functions from fixed-point vectors to fixed-point numbers. Then, we show that various popular activation functions satisfy our sufficient condition, e.g., Sigmoid, ReLU, ELU, SoftPlus, SiLU, Mish, and GELU. In other words, networks using those activation functions are capable of representing all fixed-point functions. We further show that our necessary condition and sufficient condition coincide under a mild condition on activation functions: e.g., for an activation function $σ$, there exists a fixed-point number $x$ such that $σ(x)=0$. Namely, we find a necessary and sufficient condition for a large class of activation functions. We lastly show that even quantized networks using binary weights in $\{-1,1\}$ can also represent all fixed-point functions for practical activation functions.