RMSProp's Very First Step

RMSProp updates a parameter with

$s_t = \rho\, s_{t-1} + (1-\rho)\, g_t^2, \qquad \theta_{t+1} = \theta_t - \frac{\alpha}{\sqrt{s_t} + \epsilon}\, g_t$

Training has only just started, so the running average is still empty: $s_0 = 0$. Take $\rho = 0.9$, $\alpha = 0.01$, $\theta_0 = 2$, and $\epsilon = 10^{-8}$ (small enough to ignore in the arithmetic).

The first gradient is $g_1 = -20$.

What is $\theta_1$, to four decimal places?