An Improved Adam Optimization Algorithm Combining Adaptive Coefficients and Composite Gradients Based on Randomized Block Coordinate Descent

Table 1

Description of parameters of Adam algorithm and its improvement.

Parameters	Description
	Learning rate
	Exponential decay rate of the first-order and second-order moment estimation, respectively
T, t	The maximum iterations and the current t time step, respectively
	Product of exponential decay rate of the first and second-order moment estimation at t time step, respectively, and
	The first-order moment vector at t time step
	The second-order moment vector at t time step
	Current gradient at t time step
	Adaptive coefficient
	Prediction gradient
	Random diagonal matrix at t time step
	The ith diagonal element of with independent identical Bernoulli distribution
	The parameter that needs to be optimized
	The sequence of the smooth convex loss function
	Global optimal position