In this paper, we consider transferring the structure information from large networks to small ones for dense prediction tasks. Previous knowledge distillation strategies used for dense prediction tasks often directly borrow the distillation scheme for image classification and perform knowledge distillation for each pixel separately, leading to sub-optimal performance. Here we propose to distill structured knowledge from large networks to small networks, taking into account the fact that dense prediction is a structured prediction problem. Specifically, we study two structured distillation schemes: i) pair-wise distillation that distills the pairwise similarities by building a static graph; and ii) holistic distillation that uses adversarial training to distill holistic knowledge. The effectiveness of our knowledge distillation approaches is demonstrated by extensive experiments on three dense prediction tasks: semantic segmentation, depth estimation and object detection.

Rotary motion around a molecular axis has been controlled by simple electron transfer processes and by photoexcitation. The basis of the motion is intramolecular rotation of a carborane cage ligand (7,8-dicarbollide) around a nickel axle. The Ni(III) metallacarborane structure is a transoid sandwich with two pairs of carbon vertices reflected through a center of symmetry, but that of the Ni(IV) species is cisoid. The interconversion of the two provides the basis for controlled, rotational, oscillatory motion. The energies of the Ni(III) and Ni(IV) species are calculated as a function of the rotation angle.

Crown ethers have the remarkable property of recognizing and binding specific metal cations in complex mixtures. We propose to combine molecular recognition with molecular electric conductance. The question we address is: can the event of binding a cation be sensed by a change in conductance? Specifically, we study a short molecular wire (MW) containing a crown-6 molecule connected via sulfur atoms to two gold atomic wires acting as metallic leads. Upon binding a cation, the density of states of the system is only slightly affected. This reflects the fact that the cation binding is largely electrostatic in nature and is accompanied by little electronic reorganization. Yet, the cationic binding does significantly lower conductance. We also identify strong interference affecting the conductance. A striking feature is the insensitivity of conductance to the type of ligand with the exception of the proton.

No abstract