An Improved Adaptive PID Optimizer with Enhanced Convergence and Stability for Deep Learning

TL;DR

This paper introduces IAdaPID-ADG, an enhanced adaptive optimizer for deep learning that improves convergence and stability by integrating non-increasing learning rates and gradient difference modulation, outperforming existing methods.

Contribution

The paper proposes a novel optimizer, IAdaPID-ADG, combining ideas from AMSGrad and DiffGrad to address convergence and stability issues in adaptive optimizers.

Findings

IAdaPID-ADG outperforms all competing optimizers on multiple datasets.

Ablation study confirms each component's contribution to performance.

The optimizer demonstrates significant improvements on benchmark and real-world datasets.

Abstract

Optimization is essential in deep learning. The foundational method upon which most optimizers are built is momentum-based stochastic gradient descent. However, it suffers from two key drawbacks. First, it has noisy and varying gradients, and second, it has an overshoot phenomenon. To address noisy gradients, Adam was proposed, which remains the most widely used adaptive optimizer. To address the overshoot phenomenon, a control-theory-based PID optimizer was proposed. To tackle both the limitations within a single framework, several variants of Adaptive PID (AdaPID) have recently been proposed. Although AdaPID performs well, it still inherits two critical drawbacks from Adam, namely convergence and stability issues. In this work, we address both these limitations. To fix the convergence issue, we uniquely integrate the idea of using a non-increasing effective learning rate into AdaPID (originally proposed in AMSGrad, an extension of Adam). To fix the stability issue, we innovatively integrate a gradient difference based modulation factor into AdaPID (originally proposed in DiffGrad, another extension of Adam). Combining both these ideas in AdaPID, results in our novel IAdaPID-ADG optimizer. We evaluate our proposed optimizer on multiple datasets, including benchmark datasets (MNIST and CIFAR10) and real-world datasets (IARC and AnnoCerv). The IAdaPID-ADG substantially outperforms all competing optimizers. Additionally, we perform an ablation study on the MNIST dataset to demonstrate the contribution of each added component.