For this problem use α=0.05, β1=β2=0.9 (here β2 is set to 0.9 rather than its usual 0.999 purely to keep the arithmetic light), and ϵ=10−8, small enough to ignore.
Start from θ0=4 with m0=s0=0. The parameter is bouncing across a narrow valley, so the first two gradients are equal in size and opposite in sign: