Train Faster, Perform Better: Modular Adaptive Training in Over-Parameterized Models

TL;DR

This paper introduces a modular neural tangent kernel to analyze learning dynamics in over-parameterized models and proposes Modular Adaptive Training (MAT), a selective update strategy that reduces computation and improves performance.

Contribution

It develops the concept of modular neural tangent kernel (mNTK) and proposes MAT, a novel training method that selectively updates modules based on eigenvalues to enhance efficiency and accuracy.

Findings

MAT nearly halves training computational cost.

MAT outperforms baseline training methods.

Modules with higher mNTK eigenvalues learn features more effectively.

Abstract

Despite their prevalence in deep-learning communities, over-parameterized models convey high demands of computational costs for proper training. This work studies the fine-grained, modular-level learning dynamics of over-parameterized models to attain a more efficient and fruitful training strategy. Empirical evidence reveals that when scaling down into network modules, such as heads in self-attention models, we can observe varying learning patterns implicitly associated with each module's trainability. To describe such modular-level learning capabilities, we introduce a novel concept dubbed modular neural tangent kernel (mNTK), and we demonstrate that the quality of a module's learning is tightly associated with its mNTK's principal eigenvalue $\lambda_{\max}$. A large $\lambda_{\max}$ indicates that the module learns features with better convergence, while those miniature ones may impact generalization negatively. Inspired by the discovery, we propose a novel training strategy termed Modular Adaptive Training (MAT) to update those modules with their $\lambda_{\max}$ exceeding a dynamic threshold selectively, concentrating the model on learning common features and ignoring those inconsistent ones. Unlike most existing training schemes with a complete BP cycle across all network modules, MAT can significantly save computations by its partially-updating strategy and can further improve performance. Experiments show that MAT nearly halves the computational cost of model training and outperforms the accuracy of baselines.