Vanilla Gradient Descent
Vanilla gradient descent, a fundamental optimization algorithm, is undergoing renewed scrutiny in various machine learning contexts, focusing on its application to complex models and its inherent limitations. Current research explores its efficacy in training non-linear models like decision trees and neural networks, investigating modifications like adaptive step sizes and sharpness-aware minimization to improve performance and generalization. These efforts aim to enhance the efficiency and robustness of vanilla gradient descent, potentially leading to more efficient training procedures and improved model accuracy across diverse applications.
Papers
October 20, 2024
August 17, 2024
July 5, 2024
June 7, 2024
April 2, 2024
February 9, 2024
September 30, 2023
June 28, 2023
February 28, 2023
January 6, 2023
November 4, 2022
October 23, 2022
September 12, 2022
June 25, 2022
June 9, 2022
February 3, 2022
February 2, 2022