A single parameter w is trained on
J(w)=(3w−3)2
using RMSProp:
gt=dwdJw=wt−1,st=βst−1+(1−β)gt2,wt=wt−1−st+ϵηgt
The settings are
| quantity | value |
|---|
| starting weight w0 | 3 |
| starting accumulator s0 | 0 |
| decay β | 0.9 |
| learning rate η | 0.1 |
| ϵ | 10−8 (small enough to ignore in the arithmetic) |
Exactly three updates are performed. What is the value of w3?
Round your answer to 2 decimal places. Carry at least four decimals through the intermediate steps.