Nostalgic Adam: Weighting more of the past gradients when designing the adaptive learning rate

Abstract

First-order optimization algorithms have been proven prominent in deep learning. In particular, algorithms such as RMSProp and Adam are extremely popular. However, recent works have pointed out the lack of ``long-term memory" in Adam-like algorithms, which could hamper their performance and lead to divergence. In our study, we observe that there are benefits of weighting more of the past gradients when designing the adaptive learning rate. We therefore propose an algorithm called the Nostalgic Adam (NosAdam) with theoretically guaranteed convergence at the best known convergence rate. NosAdam can be regarded as a fix to the non-convergence issue of Adam in alternative to the recent work of [Reddi et al., 2018]. Our preliminary numerical experiments show that NosAdam is a promising alternative algorithm to Adam. The proofs, code and other supplementary materials can be found in an anonymously shared link.

arXiv comment17 pages and 7 figures

References in corpus (9)

• Adam: A Method for Stochastic Optimization
• ADADELTA: An Adaptive Learning Rate Method
• On the Convergence of Adam and Beyond
• Visualizing the Loss Landscape of Neural Nets
• The Marginal Value of Adaptive Gradient Methods in Machine Learning
• Improving Generalization Performance by Switching from Adam to SGD
• Closing the Generalization Gap of Adaptive Gradient Methods in Training Deep Neural Networks
• Dissecting Adam: The Sign, Magnitude and Variance of Stochastic Gradients
• AdaShift: Decorrelation and Convergence of Adaptive Learning Rate Methods

Cited by in corpus (17)

• A Comparison of Optimization Algorithms for Deep Learning
• AdaBelief Optimizer: Adapting Stepsizes by the Belief in Observed Gradients
• Review: Deep Learning in Electron Microscopy
• On the Convergence of A Class of Adam-Type Algorithms for Non-Convex Optimization
• Adam revisited: a weighted past gradients perspective
• AdaNN: Adaptive Neural Network-based Equalizer via Online Semi-supervised Learning
• Descending through a Crowded Valley - Benchmarking Deep Learning Optimizers
• An Adaptive and Momental Bound Method for Stochastic Learning
• A Sufficient Condition for Convergences of Adam and RMSProp
• AdaX: Adaptive Gradient Descent with Exponential Long Term Memory
• Towards Practical Adam: Non-Convexity, Convergence Theory, and Mini-Batch Acceleration
• A new regret analysis for Adam-type algorithms
• Variance Reduction on General Adaptive Stochastic Mirror Descent
• Gravity Optimizer: a Kinematic Approach on Optimization in Deep Learning
• Momentum Centering and Asynchronous Update for Adaptive Gradient Methods
• AdaL: Adaptive Gradient Transformation Contributes to Convergences and Generalizations
• Adaptive Gradient Method with Resilience and Momentum