Learning Goal-Oriented Visual Dialog via Tempered Policy Gradient | IEEE Conference Publication | IEEE Xplore