Optimization algorithms that approximate Newton's method using gradient information instead of full second derivatives.