Stochastic conjugate gradient algorithm with an inertial extrapolation step for nonconvex optimization in machine learning

Key Points

• The study reveals improvements in convergence rates of stochastic gradient methods for nonconvex optimization.
• Key metrics show a notable reduction in computation time by up to 30%.
• The approach employs an inertial extrapolation step to enhance traditional stochastic gradient algorithms.
• These findings highlight the effectiveness of combining inertial methods with machine learning applications.