Unified Neural Scaling Laws
Provides a more accurate tool for predicting performance of large-scale neural networks across diverse domains, aiding resource allocation and architecture design.
The authors propose a unified functional form (UNSL) that accurately models and extrapolates the scaling behavior of deep neural networks as multiple dimensions (parameters, data, compute, etc.) vary simultaneously across vision, language, math, and RL tasks, outperforming existing scaling laws in extrapolation accuracy.
We present a functional form (that we refer to as a Unified Neural Scaling Law (UNSL)) that accurately models and extrapolates the scaling behaviors of deep neural networks as multiple dimensions all vary simultaneously (i.e. how the evaluation metric of interest varies as one simultaneously varies the number of model parameters, training dataset size, number of training steps, number of inference steps, amount of compute, and various hyperparameters) for various architectures and for each of various tasks within a varied set of upstream and downstream tasks. This set includes large-scale vision, language, math, and reinforcement learning. When compared to other functional forms for neural scaling, this functional form yields extrapolations of scaling behavior that are considerably more accurate on this set.