Conjugate-gradient-based Adam for stochastic optimization and its application to deep learning

TL;DR

This paper introduces a conjugate-gradient-based Adam algorithm that combines Adam with nonlinear conjugate gradient methods, demonstrating faster training of deep neural networks in fewer epochs through convergence analysis and experiments.

Contribution

It presents a novel optimization algorithm blending Adam with conjugate gradient methods, with proven convergence and improved training efficiency.

Findings

Fewer epochs needed for training deep neural networks.

Effective convergence properties demonstrated.

Outperforms existing adaptive stochastic optimizers.

Abstract

This paper proposes a conjugate-gradient-based Adam algorithm blending Adam with nonlinear conjugate gradient methods and shows its convergence analysis. Numerical experiments on text classification and image classification show that the proposed algorithm can train deep neural network models in fewer epochs than the existing adaptive stochastic optimization algorithms can.