Talos: A Weighted Speedup-Aware Device Placement of Deep Learning Models | IEEE Conference Publication | IEEE Xplore